Search

At present Deep Neural Networks (DNN) have a dominant role in the AI-driven Autonomous driving approaches. This paper focuses on the potential safety risks of deploying DNN classifiers in Advanced Driver Assistance System (ADAS) systems. In our experience, many theoretically sound AI-driven solutions tested and deployed in ADAS have shown serious safety flaws in practice. A brief review of practice and theory of automotive safety standards and related body of knowledge is presented. It is followed by a comparative analysis between DNN classifiers and safety standards developed in the automotive industry. The output of the study provides advice and recommendations for filling the current gaps within the complex and interrelated factors pertaining to the safety of Autonomous Road Vehicles (ARV). This study may assist ARV's safety, system, and technology providers during the design, development, and implementation life cycle. The contribution of this work is to highlight and link the learning rules enforced by risk factors when DNN classifiers are expected to provide a near real-time safer Vehicle Navigation Solution (VNS). [ABSTRACT FROM AUTHOR]

Advanced Driver Assistance Systems (ADAS) technologies can assist drivers or be part of automatic driving systems to support the driving process and improve the level of safety and comfort on the road. Traffic Sign Recognition System (TSRS) is one of the most important components of ADAS. Among the challenges with TSRS is being able to recognize road signs with the highest accuracy and the shortest processing time. Accordingly, this paper introduces a new real time methodology recognizing Speed Limit Signs based on a trio of developed modules. Firstly, the Speed Limit Detection (SLD) module uses the Haar Cascade technique to generate a new SL detector in order to localize SL signs within captured frames. Secondly, the Speed Limit Classification (SLC) module, featuring machine learning classifiers alongside a newly developed model called DeepSL, harnesses the power of a CNN architecture to extract intricate features from speed limit sign images, ensuring efficient and precise recognition. In addition, a new Speed Limit Classifiers Fusion (SLCF) module has been developed by combining trained ML classifiers and the DeepSL model by using the Dempster-Shafer theory of belief functions and ensemble learning's voting technique. Through rigorous software and hardware validation processes, the proposed methodology has achieved highly significant F1 scores of 99.98% and 99.96% for DS theory and the voting method, respectively. Furthermore, a prototype encompassing all components demonstrates outstanding reliability and efficacy, with processing times of 150 ms for the Raspberry Pi board and 81.5 ms for the Nano Jetson board, marking a significant advancement in TSRS technology. [ABSTRACT FROM AUTHOR]

Copyright of Journal of Guangdong University of Technology is the property of Journal of Guangdong University of Technology and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Autonomous navigation is the leading technology in current era, in this intelligent traffic light, sign detection, ADAS and obstacle detections were playing major role. Image segmentation is the process of dividing an image into different regions, or semantic classes. This is a challenging problem in autonomous vehicle technology because it requires the vehicle to be able to understand its surroundings to safely navigate. The major challenges in this platform are the accuracy and efficiency of model performance. The proposed method in the abstract uses a convolutional neural network (CNN) to perform image segmentation. CNNs are a type of deep learning model that is well-suited for image processing tasks. The CNN in this paper was trained on a local city dataset, and it was able to achieve a mean intersection over union (IoU) of 73%. IoU is a measure of how well the segmentation results match the ground truth labels. A score of 100% indicates that the segmentation is perfect, while a score of 0% indicates that the segmentation is completely wrong. This means that the method can segment images at a very fast rate, which is important for autonomous vehicles that need to make real-time decisions. Overall, the proposed method is a promising approach for image segmentation in autonomous vehicles. It can achieve high accuracy and speed, and it is easy to implement using Python. The proposed method attains an accuracy of 98.34 %, a Sensitivity of 97.26 % and a sensitivity of 96.37 % had been attained. The method could be used to improve the safety and efficiency of autonomous vehicles by enabling them to better understand their surroundings. [ABSTRACT FROM AUTHOR]

This research paper focuses on the development of a sophisticated cognitive detection system for driver drowsiness utilizing OpenCV and Python, presenting an advanced solution to enhance driver safety. In response to the alarming statistic from Road Transport and Highways India, revealing that 22.8% of accidents result from driver drowsiness, this initiative aims to mitigate the detrimental consequences of such incidents, which lead to loss of lives, vehicle damage, road infrastructure destruction, and financial burdens on individuals. The proposed solution involves a precise driver drowsiness detection system coupled with a multi-modal alert system. Upon detecting signs of drowsiness, the system employs nuanced alert mechanisms, such as controlled subtle jerks, to prompt the driver to regain alertness. Additionally, a red-light warning is projected onto the driver's face to counteract drowsiness effectively. Notably, instead of employing abrupt braking, our system adopts a gradual deceleration strategy to ensure a smoother and safer response, minimizing the risk of sudden movements and potential collisions with the windshield. Furthermore, the integration of cruise control mechanisms contributes to reducing the overall fatality rate. This comprehensive solution not only addresses the critical issue of driver drowsiness but also strives to enhance overall road safety by leveraging advanced technology and strategic alerting methodologies. [ABSTRACT FROM AUTHOR]

Between June 2018 and December 2023, Tesla released quarterly safety reports citing average miles between crashes for Tesla vehicles. Prior to March 2021, crash rates were categorized as (1) with their SAE Level 2 automated driving system Autopilot engaged, (2) without Autopilot but with active safety features such as automatic emergency braking, and (3) without Autopilot and without active safety features. In January 2023, Tesla revised past reports to reflect their new categories of with and without Autopilot engaged, in addition to making small adjustments based on recently discovered double counting of reports and excluding previously recorded crashes that did not meet their thresholds of airbag or active safety restraint activation. The revisions are heavily biased towards no-active-safety-features—a surprising result given prior research showing that drivers predominantly keep most active safety features enabled. As Tesla's safety reports represent the only national source of Level 2 advanced driver assistance system crash rates, clarification of their methods is essential for researchers and regulators. This note describes the changes and considers possible explanations for the discrepancies.

Safe autonomous vehicle (AV) operations depend on an accurate perception of the driving environment, which necessitates the use of a variety of sensors. Computational algorithms must then process all of this sensor data, which typically results in a high on-vehicle computational load. For example, existing lane markings are designed for human drivers, can fade over time, and can be contradictory in construction zones, which require specialized sensing and computational processing in an AV. But, this standard process can be avoided if the lane information is simply transmitted directly to the AV. High definition maps and road side units (RSUs) can be used for direct data transmission to the AV, but can be prohibitively expensive to establish and maintain. Additionally, to ensure robust and safe AV operations, more redundancy is beneficial. A cost-effective and passive solution is essential to address this need effectively. In this research, we propose a new infrastructure information source (IIS), chip-enabled raised pavement markers (CERPMs), which provide environmental data to the AV while also decreasing the AV compute load and the associated increase in vehicle energy use. CERPMs are installed in place of traditional ubiquitous raised pavement markers along road lane lines to transmit geospatial information along with the speed limit using long range wide area network (LoRaWAN) protocol directly to nearby vehicles. This information is then compared to the Mobileye commercial off-the-shelf traditional system that uses computer vision processing of lane markings. Our perception subsystem processes the raw data from both CEPRMs and Mobileye to generate a viable path required for a lane centering (LC) application. To evaluate the detection performance of both systems, we consider three test routes with varying conditions. Our results show that the Mobileye system failed to detect lane markings when the road curvature exceeded ± 0.016 m−1. For the steep curvature test scenario, it could only detect lane markings on both sides of the road for just 6.7% of the given test route. On the other hand, the CERPMs transmit the programmed geospatial information to the perception subsystem on the vehicle to generate a reference trajectory required for vehicle control. The CERPMs successfully generated the reference trajectory for vehicle control in all test scenarios. Moreover, the CERPMs can be detected up to 340 m from the vehicle's position. Our overall conclusion is that CERPM technology is viable and that it has the potential to address the operational robustness and energy efficiency concerns plaguing the current generation of AVs. [ABSTRACT FROM AUTHOR]

Advanced Driver Assistance Systems (ADAS) are advanced technologies that assist drivers with vehicle operation and navigation. Recent improvements and brisk expansion in the ADAS market, as well as an increase in the frequency of incidents such as sensor spoofing, communication interruption etc., in autonomous vehicles (AVs), have raised the need to research ADAS security technology. The security issues raised by incorporating these technologies into automobiles must be addressed to protect the privacy and safety of passengers and other road users. As a result, the purpose of this research is to investigate the security issues that arise from the integration of ADAS technologies. Addressing these challenges holds the potential to establish a foundation for enhanced safety and dependability within transportation networks amidst the ongoing advancements in vehicle technology. This paper starts by describing the vulnerabilities, threats, assaults, and defense mechanisms of the ADAS. It then delves into the attacks and countermeasures in three categories, namely VANET, Hardware, and Adversarial attacks. VANET attacks encompass threats targeting Vehicular Ad Hoc Networks, aiming to disrupt communication among vehicles or between vehicles and infrastructure. Hardware attacks focus on vulnerabilities within the physical components of ADAS, including sensors, processors, or communication modules. Adversarial attacks involve deliberate manipulations or perturbations introduced into machine learning models or algorithms utilized within ADAS. These attacks aim to deceive or undermine the functionality of AI-based systems, causing misclassification, compromising system integrity, and posing risks to user safety by exploiting vulnerabilities in the AI decision-making process. Finally, this study highlights potential areas for future research, such as the utilization of artificial intelligence (AI), the necessity of industry-wide standardization, and recommends specific future work tailored to each attack described in the corresponding sections.

The continuous evolution of autonomous vehicles technologies has significantly elevated the capabilities of intelligent transportation and road safety. Among these advancements, driving automation systems have played a vital role within vehicles, encompassing a diversity of functionality extending from systems assisting the driver called Advanced Driver Assistance Systems to Automated Driving Systems offering full control over various driving functions. Among these systems, Traffic Sign Recognition (TSR) system plays a significant role in terms of functionality and development. This survey focuses on the development of TSR systems including both detection and classification methods considering diverse techniques based on color, shape, machine learning and deep learning algorithms and offers insights into the various approaches. Hence a comparative synthesis is established to discuss thorough of them along. Furthermore, this paper presents a comprehensive study for TSR systems using two hardware platforms: Raspberry Pi and Nano Jetson to analyze their performances and provide insights into the most suitable hardware architectures for deploying efficient and reliable TSR systems. This paper concludes by advocating for a synergistic combination of the two best methods of detection and classification of road signs in terms of accuracy and processing time to build a new real time TSR methodology that effectively address this challenge and ultimately improve TSR performance implemented on the best target.

At present Deep Neural Networks (DNN) have a dominant role in the AI-driven Autonomous driving approaches. This paper focuses on the potential safety risks of deploying DNN classifiers in Advanced Driver Assistance System (ADAS) systems. In our experience, many theoretically sound AI-driven solutions tested and deployed in ADAS have shown serious safety flaws in practice. A brief review of practice and theory of automotive safety standards and related body of knowledge is presented. It is followed by a comparative analysis between DNN classifiers and safety standards developed in the automotive industry. The output of the study provides advice and recommendations for filling the current gaps within the complex and interrelated factors pertaining to the safety of Autonomous Road Vehicles (ARV). This study may assist ARV’s safety, system, and technology providers during the design, development, and implementation life cycle. The contribution of this work is to highlight and link the learning rules enforced by risk factors when DNN classifiers are expected to provide a near real-time safer Vehicle Navigation Solution (VNS).

• Advanced Driver Assistance Systems (ADAS)-equipped vehicles have potential to increase mobility and safety of older drivers; • Focus groups explored perceptions of ADAS in older drivers with and without age-related declines; • Findings highlighted the need for warning system alerts to be flexible, and complexity of vehicle interfaces reduced to meet older drivers' needs; • Customising vehicles to better suit older drivers needs will improve their safety and increase mobility and independence. Perceptions of Advanced Driver Assistance Systems (ADAS) were explored in two semi-structured face-to-face focus group studies of 42 older drivers (aged 65 years and older) with and without age-related declines. Study 1 explored perceptions regarding ADAS, focusing on visual, auditory, physical, and cognitive factors. Study 2 extended this by additionally exploring perceptions following exposure to videos and stationary vehicle demonstrations of an ADAS. Participants had a range of visual, hearing, memory, and health characteristics which impacted on their daily life. In both studies, some participants had insights regarding various ADAS technologies prior to the study, but many were unfamiliar with these systems. Nevertheless, overall, participants reported that ADAS would assist them to drive as they age and increase their mobility and independence. There were comments regarding the benefits of warning alerts, although the potential for them to be distracting was also highlighted. Participants with vision impairment preferred audio alerts and participants with hearing impairment preferred visual display alerts. Findings highlighted the potential for ADAS to assist those with age-related declines and the need to increase the flexibility of warning system alerts to suit the varying requirements of older drivers, as well as to reduce the complexity of vehicle interfaces. Collectively, these strategies would maximize the benefits of these vehicles to increase the mobility, independence, and quality of life of older drivers with and without age-related declines. [ABSTRACT FROM AUTHOR]

While Advanced Driver Assistance Systems (ADAS) promise safety benefits to drivers, there is evidence to suggest that drivers are unaware or uninformed about their vehicles’ systems and thus have poor mental models about the systems. Previous studies suggest that training improves drivers’ mental models, although some studies report limited impacts. This study investigated the relationship between training and drivers’ mental models about Adaptive Cruise Control (ACC), compared the impact of two different training approaches on drivers’ mental models, and examined the relationship between driver knowledge and trust regarding ADAS technologies. This study was conducted online, and participants were randomly and equally assigned to one of three training groups – owner’s manual (text-based); state diagram visualization; or sham (control). Surveys measured their trust and mental models about ACC before and after training. The results found that the text-based group outperformed the visualization group and the control group in terms of post-training overall mental model scores, but these differences were not statistically significant. No correlation between post-training mental model scores and overall trust scores was found. This study provides evidence that training improves users’ mental models about technology and finds that different training platforms or paradigms may affect learning differently.

Vulnerable road users (VRUs) are frequently involved in road traffic crashes worldwide, and the most serious consequences often occur in collisions with motor vehicles. Countermeasures such as advanced driver assistance systems (ADASs) for have been developed and implemented in vehicles to reduce these crashes. Estimations of the effectiveness of these systems are so far dominated by prospective studies; only a few analyze real-world crash data. The objective of this study was to estimate the crash-reducing effects of the VRU ADAS systems in car-to-pedestrian and car-to-bicycle crashes involving Volvo cars in Sweden using real-world crash data. Car-to-pedestrian and car-to-bicycle collisions across Sweden from 2015–2020 (in which the car was moving forward) were identified using insurance claims at If P&C insurance. Volvo car models with ADAS VRU and comparable Volvo models without the system were selected. Poisson regression was used to estimate the effect of the VRU ADAS on the crash rate, calculated per insured vehicle-years. For the second generation of VRU ADAS, the straight crossing path (SCP) and parallel (Longitudinal) crash rate was found to be 12% less in car-to-pedestrian/bicycle crashes—23% less in car-to-pedestrian crashes and 6% less in car-to-bicycle crashes. However, no results were statistically significant. There was no difference in crash rates when evaluating all crash situations where the car was moving forward. The limitation of this study is the low number of crashes. Reducing crashes and injuries in vehicle-to-VRU conflicts is of utmost importance for fulfilling the aims highlighted in Agenda 2030. Improving ADAS to better avoid collisions with pedestrians and cyclists has the potential to greatly improve traffic safety; future generations of the system are expected to increase VRU safety benefits. Measuring ADAS effects in real-world traffic and interpreting and relating the results are essential to ensure that future systems improve VRU safety.

Advanced driver assistance systems are an important step on the way towards the autonomous driving. However, there are new challenges in the release of increasingly complex systems. For the testing of those systems many test kilometers are necessary to represent sufficient diversity. Hence, the virtual testing of driver assistance systems brings new opportunities. In virtual environments, it is possible to run a much higher distance in a short time. Simultaneously, the complexity of the environment and the test scenarios are individually adjustable. It is possible to test scenarios that are not feasible in a real environment due to a risk of injury. A big challenge is the physical correct implementation of real vehicles and their components into the Virtual Reality. To enable a realistic virtual testing the vehicles surrounding sensors need to be modeled adequately. Thus, this paper presents an approach for the implementation of a Lidar model into a Virtual Reality. A physical Lidar model is combined with a real-time capable vehicle dynamics model to investigate the influence of vehicle movements to the sensor measurements. The models are implemented into a highly realistic virtual city environment. Finally, a test campaign shows the influence of the Lidars physics and the vehicle dynamics on the detection results. [ABSTRACT FROM AUTHOR]

Copyright of Automotive Digest is the property of Automotive Digest Editorial Office and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Safe autonomous vehicle (AV) operations depend on an accurate perception of the driving environment, which necessitates the use of a variety of sensors. Computational algorithms must then process all of this sensor data, which typically results in a high on-vehicle computational load. For example, existing lane markings are designed for human drivers, can fade over time, and can be contradictory in construction zones, which require specialized sensing and computational processing in an AV. But, this standard process can be avoided if the lane information is simply transmitted directly to the AV. High definition maps and road side units (RSUs) can be used for direct data transmission to the AV, but can be prohibitively expensive to establish and maintain. Additionally, to ensure robust and safe AV operations, more redundancy is beneficial. A cost-effective and passive solution is essential to address this need effectively. In this research, we propose a new infrastructure information source (IIS), chip-enabled raised pavement markers (CERPMs), which provide environmental data to the AV while also decreasing the AV compute load and the associated increase in vehicle energy use. CERPMs are installed in place of traditional ubiquitous raised pavement markers along road lane lines to transmit geospatial information along with the speed limit using long range wide area network (LoRaWAN) protocol directly to nearby vehicles. This information is then compared to the Mobileye commercial off-the-shelf traditional system that uses computer vision processing of lane markings. Our perception subsystem processes the raw data from both CEPRMs and Mobileye to generate a viable path required for a lane centering (LC) application. To evaluate the detection performance of both systems, we consider three test routes with varying conditions. Our results show that the Mobileye system failed to detect lane markings when the road curvature exceeded ±0.016 m−1. For the steep curvature test scenario, it could only detect lane markings on both sides of the road for just 6.7% of the given test route. On the other hand, the CERPMs transmit the programmed geospatial information to the perception subsystem on the vehicle to generate a reference trajectory required for vehicle control. The CERPMs successfully generated the reference trajectory for vehicle control in all test scenarios. Moreover, the CERPMs can be detected up to 340 m from the vehicle’s position. Our overall conclusion is that CERPM technology is viable and that it has the potential to address the operational robustness and energy efficiency concerns plaguing the current generation of AVs.

The transportation system undergoes severe impacts due to potholes and the presence of stray animals on the roads resulting in accidents and fatal injuries. The utilization of intelligent transportation systems would reduce accidents and impart safety to the overall transportation network. This research aims to impart transportation safety through a real-time alert warning system for avoiding accidents due to potholes and the presence of stray animals. The study incorporates real-time detection of transportation entities like vehicles, animals, and pedestrians through a YOLO v3 computer vision algorithm processed on the GPU environment for a higher frame rate. The potholes and animal hotspots are mapped to form a geospatial database on which the buffer tool of geographic information system (GIS) is applied. The buffer zone was implemented on the geospatial layer to alert the driver in real-time, while the vehicle approaches the buffer zone. The system yields high precision of 0.976 mean average precision (mAP) score of entity detection and the real-time alert warning alerts the driver to ensure transportation safety while avoiding any possible accidents or fatal crashes. [ABSTRACT FROM AUTHOR]

This review paper examines the dynamics of snow accumulation on vehicle surfaces and its impacts on vehicle performance and safety. It focuses on the use of computational fluid dynamics (CFD) to model snow ingress in vehicle air intakes and its interactions with sensors in advanced driver assistance systems (ADAS). Central to these studies is the coefficient of restitution (COR), which measures the elastic properties of snow upon collision with vehicle surfaces. The paper provides an overview of various turbulence models, such as large eddy simulation (LES) and Reynolds-averaged Navier-Stokes (RANS), assessing their effectiveness in simulating the aerodynamic fields that affect snow trajectories. It critically reviews existing models and highlights recent experimental studies related to COR of snowflakes based on different impact velocities and particle conditions. The discussion on the practical implementation of these findings in automotive design underscores their importance in enhancing vehicle safety and reliability under extreme weather conditions. [Display omitted] [ABSTRACT FROM AUTHOR]

This paper proposes a test scenario design method that reflects the longitudinal characteristics of reality for effective verification of advanced driver assistance systems (ADAS) and autonomous driving systems (ADS). Since the target systems interact with the external environment differently from the existing vehicle control system, realistic and various verification scenarios are required for verification. The proposed method consists of a vehicle model for simulating the vehicle behavior and a driver model to actively respond to the driving environment. In particular, the driver model used a model predictive control (MPC) algorithm to reflect the characteristic of human drivers. The longitudinal driving characteristics of human drivers were derived through a large-scale driving database analysis and considered in the driver model. The proposed method was compared with an existing car-following model using computer simulations. It was confirmed that its longitudinal driving behavior is similar to that of human drivers and that various scenarios can be designed by changing the model parameters. [ABSTRACT FROM AUTHOR]

Thanks to the rapid development of computer vision and deep learning technologies, advanced driver assistance systems (ADAS) have recently become widespread. These systems aim to increase driving safety and reduce the number of traffic accidents. Modern cars usually have ADAS systems integrated into their electronics, but other vehicles do not have such an integrated system. This paper presents a portable and image-based ADAS system for real-time detection of traffic signs, vehicles, and pedestrians. To realize real-time detection, the developed system uses the YOLO v5 algorithm. This single-stage detector is very popular as it has high detection speed and accuracy. The model was trained on the Tesla P100 graphics processing unit (GPU) with nearly 2500 images and 8 hours using GTSRB and study-specific dataset to analyze the developed system. Then, the implementation metrics (F1 score, P, R, PR curves) were calculated to evaluate the training and testing performances of the model. In addition, the model was compared in low-power, high-performance embedded platforms and in a computer to measure the real-time performance. The comparison results for (Jetson Xavier AGX, Jetson Xavier Nx, Jetson Nano and Test PC) achieved a speed of (43.59, 23.17, 6.4 and 3.41) frames per second for real-time detection, respectively. Considering the excellent accuracy and high speed, this study will guide researchers in demonstrating the efficiency and suitability of real-time road object detection with YOLO v5 on mobile platforms.

Future driver assistance and autonomous driving systems require high-resolution 4D imaging radars that provide detailed and robust information about the vehicle's surroundings, even in poor weather or lighting conditions. In this work, a novel high-resolution radar system with 1728 virtual channels is presented, exceeding the state-of-the-art channel count for automotive radar sensors by a factor of 9. To realize the system, a new mixed feedthrough and distribution network topology is employed for the distribution of the ramp oscillator signal. A multilayer printed circuit board is designed and fabricated with all components assembled on the back side, while the radio frequency signal distribution is on a buried layer and only the antennas are on the front side. The array is optimized to enable both multipleinput multiple-output operation and transmit beamforming. A sparse array with both transmit and receive antennas close to the transceivers is realized to form a 2D array with a large unambiguous region of $130 \mathrm{^{\circ }}$ × $75 \mathrm{^{\circ }}$ with a maximal sidelobe level of $-15 \,\mathrm{{\rm dB}}$. The array features a $3 \,\mathrm{{\rm dB}}$ beamwidth of $0.78 \mathrm{^{\circ }}$ × $3.6 \mathrm{^{\circ }}$ in azimuth and elevation, respectively. Radar measurements in an anechoic chamber show that even the individual peaks of the absorber in the chamber can be detected and separated in the range-angle cut of the 4D radar image. The performance is validated by measurements of a parking lot, where cars, a pedestrian, a fence, and a street lamp can be detected, separated, and estimated correctly in size and position.

This paper proposes a deep learning based object detection method to locate a distant region in an image in real-time. It concentrates on distant objects from a vehicular front camcorder perspective, trying to solve one of the common problems in Advanced Driver Assistance Systems (ADAS) applications, which is, to detect the smaller and faraway objects with the same confidence as those with the bigger and closer objects. This paper presents an efficient multi-scale object detection network, termed as ConcentrateNet to detect a vanishing point and concentrate on the near-distant region. Initially, the object detection model inferencing will produce a larger scale of receptive field detection results and predict a potentially vanishing point location, that is, the farthest location in the frame. Then, the image is cropped near the vanishing point location and processed with the object detection model for second inferencing to obtain distant object detection results. Finally, the two-inferencing results are merged with a specific Non-Maximum Suppression (NMS) method. The proposed network architecture can be employed in most of the object detection models as the proposed model is implemented in some of the state-of-the-art object detection models to check feasibility. Compared with original models using higher resolution input size, ConcentrateNet architecture models use lower resolution input size, with less model complexity, achieving significant precision and recall improvements. Moreover, the proposed ConcentrateNet architecture model is successfully ported onto a low-powered embedded system, NVIDIA Jetson AGX Xavier, suiting the real-time autonomous machines. [ABSTRACT FROM AUTHOR]

This paper presents a command governor (CG) based adaptive cruise controller (ACC) that is applied in simulation to normal driving scenarios and emergency stopping scenarios. The vehicle-following case study used in this paper involves a heavy-duty ego vehicle and a light-duty non-connected lead vehicle (i.e., the ego vehicle does not communicate with the lead vehicle and can only infer the lead vehicles’ position and velocity states through its own sensors). Typically, to ensure constraints in the presence of disturbances, receding horizon based ACCs will assume some known worst-case behavior of the lead vehicle. In the presence of a stochastic, non-connected lead vehicle, however, achieving such a guarantee requires a worst-case assumption on the behavior of the lead vehicle for all future time. In this work, the CG assumes a lead vehicle velocity profile that will be achieved with a prescribed level of certainty, based on a stochastic characterization of lead vehicle behavior that has been informed by actual on-road data. The CG ensures safe following distance under this probabilistic lead vehicle assumption. Here, “safe following distance” is based on the ego vehicle’s ability to come to a stop without collision if the lead vehicle were to suddenly brake at maximum deceleration after proceeding at a velocity profile that is prescribed based on a statistical lower bound on lead vehicle velocity. Ultimately, the CG ensures that the worst-case safe following distance is satisfied with a prescribed probability, thereby paralleling chance-constrained CG formulations. Simulation results for a heavy-duty truck indicate that the CG-based ACC outperforms a PID-ACC in terms of fuel economy and drivability. Additionally, the CG-ACC approach was able to ensure rear-end collision avoidance in emergency stopping simulations. [ABSTRACT FROM AUTHOR]

To demonstrate the expected performance of an advanced driver assistance system (ADAS) in an intelligent or a highly automated vehicle test approaches should include a combination of simulations, track tests, and road tests. The main objective of our work was to propose a new evaluation approach conducted by an external entity where the vehicle is treated as a black box. This approach allowed for the identification of a set of worst-case scenarios for a given ADAS application and combined the three test approaches mentioned earlier. Our proposed evaluation approach is broken down into three parts: (1) Scenarios synthesis, sampling strategy, and simulations, (2) Risk assessment and classification, and (3) Validation. In this article, we present our proposed approach for the validation portion. The validation portion can be further broken down into three parts. The first step included a description of the studied autonomous emergency braking (AEB), physical track tests, and of the different machines and ensemble learning techniques employed to create the predictive model. The second part utilized the Field Operational Tests database (SPMD), to implement the new sampling strategy based on the original, modified, and multivariate Metropolis-Hastings algorithm. The third part focused on collecting the prediction results, then assess the risk of each test, to classify them using a non-supervised technique (k-Means clustering). This allowed us to build a set of worst-case scenarios to make a final selection. Finally, two web applications were developed and deployed. [ABSTRACT FROM AUTHOR]

For automated vehicles (AVs) to navigate safely, they must be able to anticipate and predict the behavior of pedestrians. This is particularly critical in urban driving environments where risks of collisions are high. However, a major challenge is that pedestrian behavior is inherently multimodal in nature, i.e., pedestrians can plausibly take multiple paths. This is because, in large part, pedestrian behaviors are driven by unique intentions and decisions made by each pedestrian walking along a particular sidewalk or crosswalk. As described in this paper, we developed a hybrid automaton model of multimodal pedestrian behavior called Multimodal Hybrid Pedestrian (MHP). We account for multimodal pedestrian behavior by identifying pedestrian decision-making points and developing decision-making models to predict pedestrian behaviors in a probabilistic hybrid automaton framework. The resulting MHP model is more likely to predict the ground truth trajectory compared to two baseline models-a baseline hybrid automaton model and a constant velocity model. The MHP model is applicable to a wide variety of urban scenarios-midblocks, intersections, one-way, and two-way streets, etc., and the probabilistic predictions from the model can be utilized for AV motion planning.

Overtaking a cyclist on a two-lane rural road with oncoming traffic is a challenging task for any driver. Failing this task can lead to severe injuries or even death because of the potentially high impact speed in a possible collision. To avoid a rear-end collision with the cyclist, drivers need to make a timely and accurate decision about whether to steer and overtake the cyclist, or brake and let the oncoming traffic pass first. If this decision is delayed, for instance because the driver is distracted, neither braking nor steering may eventually keep the driver from crashing—at that point, rear-ending a cyclist may be the safest alternative for the driver. Active safety systems such as forward collision warning that help drivers being alert and avoiding collisions may be enhanced with driver models to reduce activations perceived as false positive. In this study, we developed a driver model based on logistic regression using data from a test-track experiment. The model can predict the probability and confidence of drivers braking and steering while approaching a cyclist during an overtaking, and therefore this model may improve collision warning systems. In both an in-sample and out-of-sample evaluation, the model identified drivers’ intent to overtake with high accuracy (0.99 and 0.90, respectively). The model can be integrated into a warning system that leverages the deviance of the actual driver behavior from the behavior predicted by the model to allow timely warnings without compromising driver acceptance. [ABSTRACT FROM AUTHOR]

Active pedestrian collision avoidance (APCA) systems can significantly reduce road injuries and thus have attracted considerable attention from both the automobile and the transportation industries. Numerous studies have focused on APCA; however, it remains challenging to model variable and complex scenarios in a safe, efficient and low-cost way. For this purpose, this paper proposes a novel multipedestrian collision risk assessment framework comprising a motion prediction module, a collision checking module and a collision risk assessment module. First, the motion of the ego vehicle (EV) is predicted through a constant-acceleration (V-CA) model and a constant-turn-rate and constant-velocity (V-CTRV) model in different scenarios. Next, the pedestrian motion is estimated using Kalman filter (KF) approaches based on a constant-velocity (PCV) model and a constant-acceleration (P-CA) model. Then, the potential collision area (PCA) is defined according to the predicted motions of the EV and pedestrian, and the time-to collision (TTC) is selected to conduct collision checking. Finally, the most dangerous pedestrian is identified as the target with the minimum TTC in a multipedestrian scenario, and field tests are conducted on an autonomous vehicle platform. Comparative simulations indicate that the P-CA-based KF is more accurate and robust in variable-velocity scenarios than the P-CV-based KF. Furthermore, multipedestrian scenario simulations validate the effectiveness of the proposed framework. [ABSTRACT FROM AUTHOR]

This paper proposes a test scenario design method that reflects the longitudinal characteristics of reality for effective verification of advanced driver assistance systems (ADAS) and autonomous driving systems (ADS). Since the target systems interact with the external environment differently from the existing vehicle control system, realistic and various verification scenarios are required for verification. The proposed method consists of a vehicle model for simulating the vehicle behavior and a driver model to actively respond to the driving environment. In particular, the driver model used a model predictive control (MPC) algorithm to reflect the characteristic of human drivers. The longitudinal driving characteristics of human drivers were derived through a large-scale driving database analysis and considered in the driver model. The proposed method was compared with an existing car-following model using computer simulations. It was confirmed that its longitudinal driving behavior is similar to that of human drivers and that various scenarios can be designed by changing the model parameters.

• Hazard warning modalities and timing thresholds were evaluated for older drivers with impaired vision. • Early warnings improved hazard responses and reduced collisions for older drivers with impaired vision. • Late warnings did not improve hazard responses or collision rates for older drivers with impaired vision. • Non-directional-tactile warnings were the most effective in reducing collisions for older drivers with impaired vision. • Tailored development of hazard warnings can provide benefits for older drivers with impaired vision. We examined collision warning systems with different modalities and timing thresholds, assessing their impact on responses to pedestrian hazards by drivers with impaired contrast sensitivity (ICS). Seventeen ICS (70–84 y, median CS 1.35 log units) and 17 normal vision (NV: 68–73 y, median CS 1.95) participants completed 6 city drives in a simulator with 3 bimodal warnings: visual-auditory, visual-directional-tactile, and visual-non-directional-tactile. Each modality had one drive with early and one with late warnings, triggered at 3.5 s and 2 s time-to-collision, respectively. ICS participants triggered more early (43 vs 37 %) and late warnings (12 vs 6 %) than NV participants and had more collisions (3 vs 0 %). Early warnings reduced time to fixate hazards (late 1.9 vs early 1.2 s, p < 0.001), brake response times (2.8 vs 1.8 s, p < 0.001) and collision rates (1.2 vs 0.02 %). With late warnings, ICS participants took 0.7 s longer to brake than NV (p < 0.001) and had an 11 % collision rate (vs 0.7 % with early warnings). Non-directional-tactile warnings yielded the lowest collision rates for ICS participants (4 vs auditory 12 vs directional-tactile 15.2 %) in late warning scenarios. All ICS participants preferred early warnings. While early warnings improved hazard responses and reduced collisions for ICS participants, late warnings did not, resulting in high collision rates. In contrast, both early and late warnings were helpful for NV drivers. Non-directional-tactile warnings were the most effective in reducing collisions. The findings provide insights relevant to the development of hazard warnings tailored for drivers with impaired vision. [ABSTRACT FROM AUTHOR]