Search

Lateral localization is a very sensible factor for evaluating the performance of autonomous driving. This paper addresses the general reasons of the lateral drifting by analyzing the performance of an accurate localization system at level of 10∼20 cm. The performance drops into 0.4∼3 m in winter because of the snow and wet road surface representations in the LIDAR data. Accordingly, we prove that enhancing the quality of online LIDAR data before calculating the matching score with the map is a key-solution to improve the lateral accuracy. This can be achieved by filtering-out snow, regenerating the road edges at correct positions and sharping-up the lane lines. To achieve these objectives, we propose a machine learning based road-mark reconstruction framework. The map images are converted into edge profiles to represent the road-marks in a series of peaks. Principal Component Analysis (PCA) is used to model the relationships between these peaks and extract the dominant distribution patterns. Based on the leading eigenvectors (eigenroads), the LIDAR edge profiles are safely and efficiently reconstructed during the autonomous driving. The reliability of the proposed framework has been tested using map data generated in 2015 and LIDAR data collected in 2016 and 2017 in snowy and rainy weather conditions. The experimental results have verified the robustness and truthfulness of the proposed framework to significantly improve the lateral accuracy in the LIDAR based localization systems. The localization accuracy has been enhanced to be 96.4% for 15 cm average error in very critical situations of changing the road patterns and the environmental conditions. The reliability of the proposed framework has been tested using map data generated in 2015 and LIDAR data collected in 2016 and 2017 in snowy and rainy weather conditions. The experimental results have verified the robustness and truthfulness of the proposed framework to significantly improve the lateral accuracy in the LIDAR based localization systems. The localization accuracy has been enhanced to be 96.4% for 15 cm average error in very critical situations of changing the road patterns and the environmental conditions.

An image-based self-localization method for automated vehicles is proposed herein. The general self-localization method estimates a vehicle’s location on a map by collating a predefined map with a sensor’s observation values. The same sensor, generally light detection and ranging (LIDAR), is used to acquire map data and observation values. In this study, to develop a low-cost self-localization system, we estimate the vehicle’s location on a LIDAR-created map using images captured by a mono-camera. The similarity distribution between a mono-camera image transformed into a bird’s-eye image and a map is created in advance by template matching the images. Furthermore, a method to estimate a vehicle’s location based on the acquired similarity is proposed. The proposed self-localization method is evaluated on the driving data from urban public roads; it is found that the proposed method improved the robustness of the self-localization system compared with the previous camera-based method.

During automated driving in urban areas, decisions must be made while recognizing the surrounding environment using sensors such as camera, Light Detection and Ranging (LiDAR), millimeter-wave radar (MWR), and the global navigation satellite system (GNSS). The ability to drive under various environmental conditions is an important issue for automated driving on any road. In order to introduce the automated vehicles into the markets, the ability to evaluate various traffic conditions and navigate safely presents serious challenges. Another important challenge is the development of a robust recognition system can account for adverse weather conditions. Sun glare, rain, fog, and snow are adverse weather conditions that can occur in the driving environment. This paper summarizes research focused on automated driving technologies and discuss challenges to identifying adverse weather and other situations that make driving difficult, thus complicating the introduction of automated vehicles to the market. Keywords: Automated vehicle, Self-localization, Surrounding recognition, Path planning, Adverse condition

A robust Graph Slam (GS) framework for generating precise LIDAR maps is still a challenging demand for autonomous vehicle players. This is because of the sparsity of LIDAR 3D point clouds that leads to wrong compensations of relative position errors in x, y and yaw directions. We handle this drawback by proposing a unique GS strategy to process the relative position errors in the image domain. This is achieved by converting the vehicle trajectories into 2D grayscale images using “node strategy” to encode dense environmental representations with proper identifications in Absolute Coordinate System (ACS). Accordingly, Phase Correlation and Fourier Mellin Transform are employed to non-iteratively estimate the relative positions between nodes in the image domain at loop closure areas. These estimations are assigned into a set of edges that constitute the relationships between nodes in the map. Two individual cost functions are designed to utilize these edges in optimizing the nodes' yaw angles and then x, y positions in ACS. The proposed GS framework has been tested in a challenging environment of high buildings, dense trees and longitudinal bridge in Tokyo. The experimental results have verified the robustness to generate precise maps based on Dead Reckoning measurements and outperform GNSS/INS-RTK map in critical road structures.

In order to achieve automated driving on public roads, image-based recognitions such as object detection and traffic light detection are significant technologies to understand surrounding road situations. However, the automated vehicle might be faced severe situations for sensing due to heavy sunshine. If the image captured by the onboard camera is overexposed, the image information will be lost and there is a risk of false detection. In particular, in traffic light detection where the acquisition of color information is essential, if the front traffic light and the sun overlap during intersection driving, it will not be possible to make an appropriate intersection approach judgment. Therefore, this paper developed the method to recognize sun-glare regions in the image using visual explanations of Convolutional Neural Network (CNN). The CNN outputs an attention map using the Grad-CAM method, and then the global direction of the sun-glare region can be estimated by time-series processing. The developed method contributes to implementing robust image recognitions by estimating the direction in which visibility is reduced by sunlight such as direct sunlight and reflected light from buildings.

Automatic map quality assessment is a very important process to bring the mapping modules into levels four and five of autonomous driving. In this paper, we propose a robust framework to check the map quality on behalf of human beings with indicating the possible ghost areas without using ground truth. The essence is to conduct the assessment process in the image domain instead of the point cloud plane. Therefore, the road is described by a set of nodes and each node represents a considerable road texture in Absolute Coordinate System using LIDAR reflectivity. This converts the vehicle trajectory into grayscale images with encoding stationary landmarks and road shapes. In addition, the global position errors are converted into relative position errors between the nodes and transformed into ghosting effects in the image domain. Accordingly, a mechanism to evaluate the map quality at the revisited areas is proposed based on sharpness, luminance and structure factors of the road surface. The framework has been tested in challenging environments including open-sky areas, the world’s second-longest tunnel and courses of dense trees and high buildings. The experimental results have verified the novelty and reliability of the proposed strategy to provide very trustful map quality assessment by relying on map images only. Moreover, the system is scalable to compare the maps and significantly indicates the outperformance in terms of accuracy and quality.

This paper reports an image-based localization for automated vehicle. The proposed method utilizes a mono-camera and an inertial measurement unit to estimate the vehicle pose. Self-localization is implemented by a map matching technique between the reference digital map and sensor observations. In general, the same types of sensors are used for map data and observations. However, this study is focused on the mono-camera based method using Lidar-based map for the purpose of a low-cost implementation. Image template matching is applied to provide a correlation distribution between the captured image and the predefined orthogonal map. A probability of the vehicle pose is then updated using the obtained correlation. The experiments were carried out for real driving data on an urban road. The results have verified that the proposed method estimates the vehicle position in 0.11[m] positioning errors on real-time.

The automated driving is an emerging technology in which a car performs recognition, decision making, and control. The decision-making system consists of route planning and trajectory planning. The route planning optimizes the shortest path to the destination like an automotive navigation system. According to static and dynamic obstacles around the vehicle, the trajectory planning generates lateral and longitudinal profiles for vehicle maneuver to drive the given path. This study is focused on the trajectory planning for vehicle maneuver in urban traffic scenes. This paper proposes a trajectory generation method that extends the existing method to generate more natural behavior with small acceleration and deceleration. This paper introduces an intermediate behavior to gradually switch from the velocity keeping to the distance keeping. The proposed method can generate smooth trajectory with small acceleration/deceleration. Numerical experiments show that the vehicle generates smooth behaviors according to surrounding vehicles.

金沢大学高度モビリティ研究所, Autonomous vehicles are being developed rapidly in recent years. In advance implementation stages, many particular problems must be solved to bring this technology into the market place. This paper focuses on the problem of driving in snow and wet road surface environments. First, the quality of laser imaging detection and ranging (LIDAR) reflectivity decreases on wet road surfaces. Therefore, an accumulation strategy is designed to increase the density of online LIDAR images. In order to enhance the texture of the accumulated images, principal component analysis is used to understand the geometrical structures and texture patterns in the map images. The LIDAR images are then reconstructed using the leading principal components with respect to the variance distribution accounted by each eigenvector. Second, the appearance of snow lines deforms the expected road context in LIDAR images. Accordingly, the edge profiles of the LIDAR and map images are extracted to encode the lane lines and roadside edges. Edge matching between the two profiles is then calculated to improve localization in the lateral direction. The proposed method has been tested and evaluated using real data that are collected during the winter of 2016-2017 in Suzu and Kanazawa, Japan. The experimental results show that the proposed method increases the robustness of autonomous driving on wet road surfaces, provides a stable performance in laterally localizing the vehicle in the presence of snow lines, and significantly reduces the overall localization error at a speed of 60 km/h. © 2005-2012 IEEE.

Traffic light recognition is an indispensable elemental technology for automated driving in urban areas. In this study, we propose an algorithm that recognizes traffic lights and arrow lights by image processing using the digital map and precise vehicle pose which is estimated by a localization module. The use of a digital map allows the determination of a region-of-interest in an image to reduce the computational cost and false detection. In addition, this study develops an algorithm to recognize arrow lights using relative positions of traffic lights, and the arrow light is used as prior spatial information. This allows for the recognition of distant arrow lights that are difficult for humans to see clearly. Experiments were conducted to evaluate the recognition performance of the proposed method and to verify if it matches the performance required for automated driving. Quantitative evaluations indicate that the proposed method achieved 91.8% and 56.7% of the average f-value for traffic lights and arrow lights, respectively. It was confirmed that the arrow-light detection could recognize small arrow objects even if their size was smaller than 10 pixels. The verification experiments indicate that the performance of the proposed method meets the necessary requirements for smooth acceleration or deceleration at intersections in automated driving.

Accurate state recognition of traffic lights is important for a safe urban drive of autonomous vehicles. A traffic light with lower luminance than its ambient is difficult to recognize particularly in a distant. This paper proposes a strategy that improves accuracy of signal state recognition of distant traffic lights with low luminance. Digital map is used to get candidate regions of traffic lights on a captured image. Each candidate is cropped and resized to equalize the size of lamp area on the image. On the resized images, brightness channel image is modified by enhancing pixels that have hue value in the range of traffic signal color. Then, high value pixels on the SV m image, which is the muitilpied image of the satulation channel and the modified brightness channel, is regarded as candidates of lamp area. This strategy attempts to emphasize traffic signal lights darker than its ambient. Experiments in real driving data showed the performance of the proposed method.

This paper presents the 76GHz MWR (Millimeter-Wave Radar)-based self-localization method for automated driving during snowfall. Previously, there were many LIDAR (Light Detection and Ranging)-based localization techniques proposed for high measurement accuracy and robustness to changes of day and night. However, they did not provide effective approaches for snow conditions because of sensing noise (i.e., environmental resistance) created by snowfall. Therefore, this paper developed a MWR-based map generation and a real-time localization method by modeling the uncertainties of error propagation. Quantitative evaluations are performed on driving data with and without snow conditions, using a LIDAR-based method as the baseline. Experimental results show that a lateral root mean square error of about 0. 25m can be obtained, regardless of the presence or absence of snowfall. Thus, it can be investigated that a potential performance of radar-based localization.

This paper presents a strategy to estimate the vehicle altitude using predefined elevation maps. The map images in multilayers roads must be distinguished and separated with respect to the driving segments during the autonomous driving. This can be achieved by continuously estimating the altitude value and retiring the map images accordingly. The main influence is to increase the lateral and longitudinal Iocalization based on matching the LIDAR and retrieving map images. Accordingly, a framework is designed and evaluated in complicated environments of common driving areas between bridges, tunnels and ground roads. The experimental results have verified that the proposed strategy is very reliable and provides precise altitude estimation with averaged accuracy of 10cm.

This paper highlights the importance of incorporating holistic and feature based localization systems in autonomous driving. The intensity based localization system is represented by calculating the matching score between LIDAR and map images whereas the feature based system is integrated by extracting the lateral edges with respect to the vehicle heading angle. An edge matching technique is then applied to estimate the lateral position based on the common features between the map and LIDAR images. The experimental results have verified that the estimation of the lateral and longitudinal poses has become more robust by combining the image and edge matching results against the changes of weather and environmental conditions.

Object classification is an important issue in order to bring autonomous vehicle into reality. In this paper, real-time and robust classification based on Real AdaBoost algorithm is researched and improved. Various effective features of road objects are computed using LIDAR 3D point clouds. The improved classifier provides an accuracy of over 90 (%) in a range of 50 (m) and classifies objects into car, pedestrian, bicyclist and background. Moreover, processing time of classifying an object consumes only 0.07∗10−3 (sec) that enables this method to be used for autonomous driving on urban roads.

This paper reports an image-based localization for automated vehicle. The proposed method utilizes a mono-camera and a low-cost velocity and inertial sensor to estimate the vehicle pose. Image template matching is applied to provide a correlation distribution between the captured image and a vector structured digital map. A probability of the vehicle pose is then updated using the obtained correlation. The experiments were carried out for real driving data on an expressway. We evaluated the proposed method for daytime and nighttime driving data under different lighting conditions. The results have verified that the proposed method estimates reasonable positioning errors on real-time comparing with a low-cost GNSS/INS.