Search

Quadrotor UAVs are widely used in military, ecological, environmental, and other fields because of their many advantages, such as flexible movement, vertical takeoff and landing, quadrotor UAVs require the active obstacle avoidance capabilities for invading objects in complex scenarios, while current research is mainly focused on obstacle avoidance under smooth motion conditions, resolve the dynamic high-speed or sudden invading object avoidance problem is difficult. In this paper, we propose a novel active avoidance method for high-speed object invasion based on extreme maneuver controlling. First, design the hybrid control system for the quadrotor UAV to switch between the smooth flight mode and the extreme maneuvering mode respectively. Second, plan the UAV's trajectory and angular rate to ensure high-speed smooth flight and perform extreme maneuvers. Finally, we carried out three comparative experiments and the results demonstrate extreme maneuvering obstacle avoidance strategy is effective.

Bounding box regression is a very effective method to improve the localization accuracy of object detection. Recently, the IoU-based regression losses have been widely used in object detection algorithms. However, we observe that they degenerate seriously in the late training period, leading to slow convergence and inaccurate localization. In this paper, we design a Fast Intersection over Union (FIoU) loss, which can not only keep the advantages but also solve the weakness of IoU-based losses. Furthermore, FIoU can be directly applied to Non-Maximum Suppression (NMS) as a criterion to improve the localization performance. Numerous experiments on two popular benchmark datasets show that our method is superior to other the-state-of-art methods.

As a research hotspot of computer vision, crowd counting methods have achieved success in natural images. But crowd counting in aerial images are rarely explored, and existing methods do not perform well because of the higher resolution, smaller object scale and more complex scene. Therefore, this paper proposes a lightweight dual-task network (LDNet) for crowd counting, which only uses bifurcated structure to overcome these new challenges in aerial images without complicated pipelines. To realize this, a complete but efficient Guidance Branch is proposed to assist Counting Branch in fitting crowd distribution. Furthermore, a scene attention mechanism is used to consider the complex scene information, which are never considered by existing methods. Our LD-Net outperforms existing methods on aerial crowd counting dataset (Visdrone), and gets better or comparable results on natural crowd counting datasets (UCF_CC_50, UCF_QNRF, ShanghaiTech Part A).

3D reconstruction is widely used in various scenarios such as virtual simulation and unmanned driving. Correspondingly, the requirements for its quality have been greatly improved. This paper presents a series of performance evaluation criteria based on the internal distribution of dense point cloud. In the definition of the assessment, the amount of point cloud information, uniform distribution, and the level of structural recurrence are considered. Using the criterion proposed in this paper, the pros and cons of the 3D reconstruction result can be comprehensively evaluated in comparison experiments of improving certain steps or using different methods. The design can further guide directions for optimization research and subsequent application. The experimental results show that adding the proposed standards based on general assessments can effectively reflect the quality of the reconstruction model. The performance evaluation criteria are scientific and practical.

We design a new multi-modal perception fusion system for the whole scene, which can be divided into two parts. The first part is lidar, rgbameral, rgbamera2. The second part is RGB camera 3, RGB camera 4, depth camera. The combination of these three modes provides a close range (0.3-10 m) of environmental perception, and provides the texture and distance information of the environment in the cone-shaped blind area of lidar. Our main work is to design the experimental equipment of the proposed multi-modal perceptual fusion system, and to push forward the optimization theory with the re projection error as the objective function and analyze the multi-modal fusion sensing system. The graph structure is established, and the relative pose transformation between various modes is analyzed. In the back-end optimization part, MCPE (Minimally connected pose estimation) and PSE(Pose and structure estimation)are used to analyze and compare the system. In the experimental part, we use matlab to calibrate the binocular camera by minimizing the error of re projection. The lidar, RGB camera and depth camera are calibrated with the help of the multi-modal calibration tool (binding on ROS) proposed. It provides a solid foundation for the future experiments in perception of self-driving, UAV perception, medical field and military field.

The increasing sophisticated robot and intelligent system applications require universal visualization platforms which can guarantee the security and efficiency of task process execution in the situation of user-programming and using different kinds of automated equipment. In this paper, we present a universal visualization framework to build up program-driven simulation software of complex robots and intelligent systems by integrating several open-source technical modules, including Ubuntu Linux operation-system, QT Creator IDE environment, ROS robot operation system, OSG(OpenSceneGraph) 3D scene, osgEarth GIS(Geographic Information System)-based 3D scene, and also Python based user-programing robotic script language. Many complex visualization simulation systems of complex tasks in wide area and dynamic scenarios are realized by using this framework. Based on this framework, we built a virtual simulation platform CyberEarth for robotics and Cyber-Physical systems. The typical robotic simulation task, which is a visual coverage task for Multi-Agent/UAV, is also introduced to demonstration the universality of this platform.

In the top-down approaches of multi-person pose estimation, a human detector is adopted first to generate a set of human bounding boxes, then crop these human body and perform a single-person pose estimation model to get the final result. However, some body part of another person on the cropped image will interfere the single-person pose estimation model leading to an inaccuracy result. In order to model the relationship between adjacent keypoints effectively to alleviate this problem, we propose and attention module that could let the model get global receptive field at the shallow layer of the network and pay more attention to the key areas which is more important to pose estimation. Experiment results show that our method achieves 73.9% mAP with 2.4% absolute improvement compared to our baseline on the COCO test-dev dataset.

Device-to-device (D2D) communications underlying cellular networks are a novel approach for providing IoT services, in which D2D communication links and cellular communication links share the same channels. Unfortunately, various kinds of devices are randomly distributed in cellular networks and potentially result in critical interferences. To solve the interference problem, we propose a novel greedy-based channel assignment algorithm in this paper. First, we construct a novel interference graph that describes the interferences between pairs of communication links when they simultaneously share the same channel. Then, a novel channel assignment algorithm is proposed to alleviate system interferences by greedily assigning the best channel to each communication link, which is simple and feasible to implement. Moreover, we convert the channel assignment problem into a robust graph coloring problem and search for a near-optimal solution to minimize the system interferences. Numerical results indicate that the proposed algorithm not only dramatically improves the network capacity but also enhances the fairness among devices.

Timely and credible species recognition of wildlife at their habitats is helpful for ecological monitoring. Thanks to the powerful feature extraction ability of convolutional neural networks(CNN), the performance of species identification has been significantly improved. However the CNN based approach still does not achieve their full potential. The cluttered backgrounds and rich feature changes of wild environment bring great challenges to wildlife recognition. In this paper, we propose a novel approach to improve the anti-occlusion ability of CNN model, which is achieved by training a improved anti-occlusion loss function. The anti-occlusion constraint in our proposed loss function works to reduce the distance of feature expression before and after a sample has been occluded. We validated the effect of occlusion based on public dataset cifar-10, which confirms that its necessary to improve the anti-occlusion ability of CNN model. Based on single-labeled training dataset and generalization dataset, comprehensive comparative evaluation proves that our proposed loss can effectively improve the generalization ability of CNN model during the identification of wildlife.

The traditional electromagnetic nondestructive testing probe structure includes magnets and yokes, and forms an excitation circuit with the wire rope. The wire rope is excited to a near-saturation state. The magnetic sensor is used for qualitative or quantitative detection of leakage flux from defects. The yoke is essential in the traditional excitation circuit. However, the use of a yoke in a detection probe of such an excitation mechanism has a large volume and is heavy. Through simulation, we improved the magnetization direction and installation position of the permanent magnets. Without a yoke, the permanent magnets were placed in the direction of co-extrusion. The wire rope was excited to near saturation in the permanent-magnet part. However, there was no excitation in the center of the two permanent magnets. The magnetic sensor was placed in the center of two permanent magnets to pick up the magnetic-flux-leakage signal for nondestructive testing of the steel wire rope. This method simplifies the permanent-magnet magnetization. The excitation probe with the yoke removed was also reduced in volume and weight. Through experimental testing, it was proven that the proposed shoving-magnetic-field structure and the traditional detection structure have the same detection effect.

Long-term non-interventional observation of wild animals in the natural environment is very necessary for ecological protection. With the development of artificial intelligence, it is possible to effectively utilize the features of wild artiodactyls and realize multi-target tracking and reidentification. In this paper, a re-identification and tracking-by-detection framework is proposed for real-time tele-observation of Artiodactyla. According to the characteristics of artiodactyla, our algorithm designed a three-direction feature extraction and feature matching method to achieve re-identification. The kalman filter is used in cooperation with the detector to confirm the presence of the target. Our framework integrates detector, kalman tracker and KCF tracker, which alleviates the problem of discontinuous detection results caused by the low detection rate of artiodacods in the field environment. In the KCF tracking process, the tracking bounding boxes are corrected with high confidence detection results. The experiments demonstrate the feasibility and effectiveness of the framework.

Road detection is a key problem in the application of autonomous driving and navigation. However, most of the recent approaches only obtain reliable results in certain well-arranged scenes. In this paper, we propose an online-learning road detection method for robust and real-time unstructured road detection in challenging scenes. Firstly, we describe an improved adaptive gamma correction method compensate for nonuniform illumination under rapidly changing illuminate conditions. Then we divide the image with a grid into suitable sized rectangular regions (cells), and some few cells are labeled as train/test samples. Finally, we adopt the online learning process through artificial neural networks so that our method can be adaptive to the varied environment. Experiments on the challenging database demonstrate the real-time capability, adaptability and reliability of our approach to varied illumination and complex scenarios.

In this paper, we propose a multi-camera combined calibration method for co-viewpoint circle-imaging system to obtain panoramic images. In recent years, high-resolution images are applied in wide fields of artificial intelligence to satisfy various scenes, such as tele-observation, 3D environment reconstruction and road detection. But most common methods are based on complicated matching algorithm processing. The result has high precision but also costs much time to deal with images captured by cameras. However, the optical circle-imaging system in this paper only applies simple cropping and splicing to image matrix instead of feature points matching among images captured by different cameras or at different angles. And in order to carry out the calibration and adjust positions of cameras in this system, we design an improved template consisting of three identical chessboards with settled relationship between every two boards. Once the system is constructed, it can give the panorama in a short time with high resolution and accuracy.

This paper addresses the problem of tracking and landing on an arbitrary object of a quadrotor. It is assumed that the vertical and lateral velocity controllers of a quadrotor can be designed independently. We have developed an image based visual servoing method that uses only a downward looking camera for tracking and landing from a quadrotor. Our visual detection algorithm allows the user to specify an arbitrary object to detect its movements in the image plane. In the horizontal plane, we use the lateral velocity controller to keep tracking with the object in the center of the camera image. In the vertical profile, we adopted a log polynomial velocity to guarantee zero closing velocity and acceleration at touchdown. During the tracking phase, the quadrotor tracks the object continuously while maintaining a fixed altitude from the object and keeps it in the center. Once the quadrotor satisfies the landing condition, it moves on to the landing phase at the planned vertical velocity until touchdown. Here, we use the ultrasound sensor running kalman filter to detect the landing condition. Our experiments show that the system is able to track an arbitrary object and land on it successfully.

Recently, deep learning technology has been widely used in ecological environment monitoring, but there are still huge challenges in specific tasks. For example, data sets have increased over the past few years, and in a CPU environment, training is quite time consuming. In order to solve these problems, we bring up a novel and effective method to improve training speed. In this paper, we make use of multi-GPUs to achieve distributed parallel acceleration. Small and shallow networks are not suitable for distributed training, because the computation of each parameter in this network is much higher than that of multi-layer perception or automatic encoder architecture, we use convolutional neural network with parameter sharing as the training network. In this paper, the main adopted method is to compare the performance of training wildlife recognition in single GPU and multi-GPUs environments by using distributed deep learning framework TensorFlow. The experimental results show that multi-GPUs which adopt distributed architecture can significantly accelerate training time consumption than single GPU. The results of this experiment also provides strong support for our follow-up work.

Wireless Mesh network is a robust and dynamic multi-hop network. Compared with traditional wireless networks, wireless Mesh networks have many advantages, such as good reliability, high bandwidth, low deployment cost, wide coverage, and good scalability. Therefore, wireless MESH network has become the most ideal wireless network solution. The deployment location of Mesh base-station nodes can affect the performance of wireless Mesh networks to a great extent. Therefore, finding an appropriate and effective Mesh router node deployment scheme is the key to improve network performance. At present, Mesh network has been applied in smart city, but it is rarely used in the field. Based on the 3D mountain environment in the Wanglang national nature reserve of Sichuan province, this paper conducts an in-depth study on the deployment algorithm of wireless Mesh network nodes. In this paper: firstly, the deployment problem of wireless Mesh router is introduced. Then the Mesh node deployment problem is abstracted and the corresponding mathematical model is established. Then, under the premise of satisfying the user coverage and connectivity requirements, an algorithm for solving the Mesh node deployment problem is proposed with the minimum network construction cost. Finally, the algorithm is simulated and analyzed on the simulation platform.

This paper focuses on the vision based autonomous landing of quadrotor using fuzzy logic controller with low-cost USB camera as perceptive system. Visual algorithm adopts open source AprilTag detection method with robustness to illumination variation and high accuracy in long range of distance. In order to make the quadrotor flying in continuous and smooth status in position and velocity, we adopt fuzzy PID logic as the controller to calculate desired velocity in longitude, latitude and vertical axis. First, using visual position feedback and the fuzzy PID controller velocity output, the quadrotor aligns its center with the pad center as well as reduce the altitude until lands. The control method has been tested in simulation, and the whole system has been carried out in real outdoor environment.

Ahstract─ Road geometric parameters such as turning radius, roadway width, deflection angle and direction are all the constraints when robots passing through a road. When robots know the geometric parameters of the whole road, the control input could be designed in advance and the trajectory could be planned based on the whole road. In this paper, we present an approach to mathematical formulation of a free road which contains multiple curves. We designed a construction to present a free path with the combination of three parts: straight line, spiral and circular arc. Then add the width of the transverse and construct a free road model oriented to robot. Second, we establish the relationship between the trajectory curvature and the velocity which combined the presented road model and the dynamic model of a four-wheeled differential robot. Third, simulation and actual experiments have been carried out to verify the road model and proved that the model can be used to control the robot.

There are many ecological problems such as the greenhouse effect, air pollution, biodiversity loss. Nature environmental monitoring and wildlife detection by rangers in the wild could be arduous and dangerous. Intelligent ecological environment surveillance system for wildlife and environment factors is necessary. In this paper, we present a new design of the nature ecological intelligent surveillance station with environment factors sensors and robotic observation cameras which can detect wildlife that appear in the field of vision and sequentially track them actively by robotic pan-tilt-zoom cameras according to their rarity. Several such stations form a smart surveillance system to detect wildlife and monitor environment factors in a field. Based on this system, the activities and living habits of wildlife and the climate changes in the forest can be effectively and constantly monitored by the cooperation of multiple devices such as robotic pan-tilt-zoom cameras, wind speed and direction sensors, temperature and humidity sensors, pm2.5 concentration sensors configured in this system. The experiments demonstrate the feasibility and effectiveness.

Recent years, the convolutional neural network have shown to provide excellent results on recognition in different competitions. However, challenges in specific missions still exist. The cluttered backgrounds and rich feature changes of wild environment bring great challenges to the problem of species recognition of wild animals. To address these problems, this paper proposes a novel and effective combination to learn a CNN model. This is achieved by apply simple linear iterative clustering (SLIC)super-pixel segmentation method to unified data dimension during the process of making raw image data (captured by camera-traps)into a dataset. In short, the super-pixel-divided images provides the input of the convolutional neural network. In order to verify the application, we conducted a comprehensive performance comparisons between our SLIC-dataset and generally used Resize-dataset over CNN networks. Results proved that our proposed method performs exceptionally well in low-resolution data when it is crucial to take full advantage of the edge information of original images. In addition, we collected and annotated a standard camera-trap dataset of 14 common wildlife species in China, which contains 16,480 training images and 4,120 testing images.

Patrol and Surveillance could be dangerous and arduous in ecological environment protection. In this paper, we presents a new design of unmanned all-terrain vehicle (U-ATV) system based on open source framework. According to bottom-up design principles, this system consists four layers: robot body layer, data service layer, cloud service layer, user interaction layer. The open source based linux OS is deployed in both robot system and cloud system. The U-ATV robot is driven by ROS kinematic engine. The perception information from many kinds of sensing devices is collected by C++/Python interfaces. The user interaction is realized by QT based cross-platform IDE software. The experiments demonstrate the feasibility and effectiveness of this design based on a hard-in-Ioop robotic driving and operation testing platform.

The use of visual methods to explore the terrain in the wild has drawn more and more attention. Structure from motion techniques have been extensively employed to 3D topographic reconstruction from videos. Conventional 3D reconstruction of the large scene area requires the aircraft to carry the camera to photograph, but need a lot of time. As a result, it needs to plan the keyframes and matching. In this paper, a keyframe matching topological graph selection strategy is proposed for effective 3D reconstruction. First, plan the scene traversal photographic mode by Terrain elevation information. A photography constraint method, which conforms to the orthographic and equidistant projection, is proposed. After that, the selection of the key frame is performed using the planning information and the DEM information. Finally, establish the matching relationships between the frames. We verify the efficacy of our method on a mountain terrain, and demonstrate that our method is able to greatly reduce the reconstruction time and improve the reconstruction accuracy.

The four-wheel differentially driven mobile robot adjusts its position and orientation by controlling the rotating speeds of four wheels. Proper four wheel speed coordination is important to reach the desired speed and trajectory , and will not cause damage to the mechanical structure. In this paper, considering the robot structure parameters and path geometric constraints, we establish the relationship between turning radius, the robot's velocity at the center of gravity (COG) and the rotating speeds of four wheels. In addition, the relationship between the yaw rate and the slip angle and the above three variables is alse established. Those relations are all based on the theory of instantaneous center of rotation (ICR). Simulative results verify them correctness and will not do damage to mechanical structure.

Indoor localization system based Wi-Fi received signal strength (RSS) has gained popularity in recent years, as wireless local area networks and Wi-Fi enabled mobile devices are pervasive penetration. Unfortunately, the Wi-Fi RSS measurements are susceptible by device heterogeneity, multipath and signal noise, etc. To remedy these problems, we propose a robust Wi-Fi fingerprint-based indoor localization system. The proposed algorithm extract a robust positioning feature from Wi-Fi signals in both 2.4 GHz band and 5 GHz band by kernel principal component analysis (KPCA). Furthermore, we utilize Wi-Fi signal selection algorithm and coarse localization scheme for increasing localization accuracy and reducing the computational burden. Finally, the weighted k nearest neighbor method (WKNN) is used to obtain the estimated location. The proposed system implemented in a realistic indoor Wi-Fi environment, and results indicate that it is efficient in improving the positioning performance.

Coverage path planning is the operation of finding a path that covers all the points of a specific area. Thanks to the recent advances of hardware technology, Unmanned Aerial Vehicles (UAVs) have multiple use at present besides obvious military applications, such as nature environment 3D reconstruction. However most of the research focus on finding the coverage path in an planar coverage surface, without considering the photogrammetric constraints, such as constant distance for consistency resolution and normal orientation for orthophotograph. This paper present a modified back-and-forth 3D coverage path planning method for quadrotor with gimbal camera under photogrammetric constraints that can reduce the consumption of time and energy. First, the 3D terrain is modeled by quasi uniform B-spline surface for a mesh representation. Then, the center of the camera footprint on the terrain surface and the air points can be generated by photogrammetric constraints. Finally, a modified back-and-forth coverage path planning method has been used to calculate the scan direction and coverage path. The experiment results have showed that our method can cover all the waypoints in the air that satisfy all the photogrammetric constraints.

WiFi indoor localization has attracted much attention owing to the pervasive penetration of wireless local area networks (WLANs) and WiFi enabled mobile devices. Traditional WiFi indoor localization systems rely on received signal strength (RSS), which is instability and low space distinguish ability. Recently, channel state information (CSI) has been adopted instead of RSS and proven to be an efficient method. However, CSI raw data is sensitive to environment change. In this paper, we propose an improved WiFi indoor localization system combining CSI and RSS. Firstly, this study applies a time domain filtering method to reduce CSI data dynamic range. Then, we use the coherence bandwidth for CSI dimensional reduction. Afterwards, we extract a robust positioning feature from CSI and RSS. Finally, an improved weighted k-nearest neighbor algorithm based on kernel methods is used to estimate the location. The experiments demonstrate the effectiveness of the proposed system.

With the rapidly increasing number of users and various internet of things applications, the capacity of current air interface of cellular networks has quickly become insufficient. Heterogeneous cellular network based IoT communications is an economical and effective way to increase network capacity and extend coverage area, which has been considered as a natural choice to address this problem. In heterogeneous cellular networks, the layout of homogeneous cellular networks (i.e. homogeneous networks) has been changed by adding low-power nodes into macrocell. As a result, a new problem arises: mobile association among various serving nodes. In order to achieve excellent system performance, this problem should be well addressed. In this paper, we proposed a novel association approach which takes backhaul link, quality of service and resource consumption into consideration. A utility framework which is a function of spectrum efficiency and cell load is formulated. The approach not only improves spectrum efficiency in relay enhanced network but also maintains the load balancing between cells. The simulation results show that the proposed approach can provide higher system capacity and better load balancing.

In recent years, the correlation filter-based trackers (CFTs) have shown to provide excellent results in different competitions and benchmarks, but there is still a need to improve the robustness of CFTs. Compared with the traditional kernel correlation filter tracker, the approach we present in this paper makes some significant improvements. The strong features including HOG and Color-naming are integrated to maintain a more powerful object representation, and PCA is applied to boost the computation speed. To deal with scale variation, a multiple scale adaptive method is adapted. Model updater is modified by considering all previous frames to update the discriminative classifier coefficients. Extensive evaluations are performed on 51 challenging benchmark sequences. The experiments results show that our approach outperforms state-of-the-art tracking methods. Additionally, the proposed approach is implemented in our practical project, and the results also prove the availability of our approach.

In this paper, we present a fast method of key frame selection and an accurate feature point matching method for 3D reconstruction through the control of quadrotor. The quadrotor is controlled to fly at a fixed altitude and the gimbal camera has always been a downward direction. Therefore, the additional information of the quadrotor could be easily added to the 3D reconstruction process. As a result, we find a fast method to select key frames and an optimized matching method through the geometric constraints of the flight path and direction. The method is mainly to improve the efficiency of key frame selection and to get more accurate matching points. The advantage of our approach is to combine the two processes between video capture and video processing by adding flight control to the 3D scene reconstruction method. Therefore, it can use less time to complete the entire reconstruction task. The proposed approach is tested by quadrotor platform. Experiment results show that our method can greatly reduce the key frame selection time and get more matching points at the same accuracy.

Modern surveillance systems commonly use pan-tilt-zoom camera to monitor the wildlife. With the expansion of monitored area, it is a fundamental problem to acquire clear images of the interested target by controlling camera. In this paper, we propose a kinematics model of a pan-tilt-zoom robotic camera to address this problem. The pan-tilt-zoom camera can be regarded as a special Tele-Observation robot with three degrees of freedom: pan, tilt and zoom. The field of view of the pan-tilt-zoom camera is taken as the end-effector of the Tele-Observation robot. A model of the Tele-Observation robot is proposed and the model parameters are estimated. The zoom model that reflects how focal length vary as a function of zoom scale is introduced. This model has higher accuracy of target's spatial location compared with existing method, especially on the aspect of depth information. The experiments demonstrate the feasibility and effectiveness of our new model. The average relative error of the field experiment is 0.49%, which can satisfy the requirement for wild surveillance.

In wide environment surveillance systems, panoramas are widely used to display environmental information. Because of the diversity of environment, refreshing the environment uniformly is unscientific. This paper presents a hybrid performance evaluation criterion based on information entropy. In the definition of this criterion, the effective degree of information, the importance of information and refresh demand degree are taken into account. Using the performance evaluation criterion proposed in this paper, a refresh path can be found to refresh the panorama with different importance harmoniously and keep the panorama with higher information entropy. It is beneficial for surveillance systems to find the interested target. The experiment results show that the probability of target acquisition is increased by 24.7% and the mean acquisition time is shortened by 37.7%. Compared with the typical zigzag path method, the performance evaluation criterion is scientific and practical.

Protein-DNA complexes play crucial roles in gene regulation. The prediction of the residues involved in protein-DNA interactions is critical for understanding gene regulation. Although many methods have been proposed, most of them overlooked motif features. Motif features are sub sequences and are important for the recognition between a protein and DNA. In order to efficiently use motif features for the prediction of DNA-binding residues, we first apply the Convolutional Neural Network (CNN) method to capture the motif features from the sequences around the target residues. CNN modeling consists of a set of learnable motif detectors that can capture the important motif features by scanning the sequences around the target residues. Then we use a neural network classifier, referred to as CNNsite, by combining the captured motif features, sequence features and evolutionary features to predict binding residues from sequences. The datasets PDNA-62 and PDNA-224 are used to evaluate the performance of CNNsite by five-fold cross-validation. Performance evaluation shows that the motif features performs better than sequence features and evolutionary features with at least 6.73% on ST, 0.097 on MCC and 0.069 on AUC. When comparing with previously published methods, CNNsite performs better with at least 0.019 on MCC, 4.37% on ST and 0.040 on AUC. CNNsite is also evaluated on an independent dataset TS-72 and CNNsite outperforms the previous methods by at least 0.012 on AUC. The discriminant powers of the motif features of size from 2 to 6 residues show that many motif features with large discriminant power are composed by the residues that play important roles in the DNA-protein interactions. The standalone version of the CNNsite is available at http://hlt.hitsz.edu.cn:8080/CNNsite/.

Network flow model (NF) with the pairwise property has been paid more attention by researchers in recent years. However, the pairwise will bring a detrimental tracking performance under the occlusion case, such as the phenomenon of ID switches (IDs) and frequent appearing of fragments. Referring to the pairwise, we propose the robust network flow (RNF) model, which is based on a prior that every candidate detection has a correlative degree with the current tracking path. Based on the correlative degree, the cost of the pairwise nodes is recomputed and the best candidate detection can be found. The experimental results on four challenging video sequences show that our method is able to obtain the superior performance under the occlusion case.

Single object tracking can be focused on two phases under the particle filter framework: one is sparse representation, which can be regarded as a matching evaluation; the other is template update, which can be regarded as the appearance changes of the target. Template update is the most direct and basic phase to ensure a high quality tracking. However, most template update schemes can not capture the latest appearance of the target, thereby leading to a low quality tracking. In this paper, we propose a new template update scheme, which can obtain the latest trends of the target. The experimental results on popular benchmark video sequences show that the proposed template update scheme is feasible and effective.

Patch strategy is widely adopted in visual tracking to address partial occlusions. However, most patch-based tracking methods either assume all patches sharing the same importance or exploit simple prior for computing the importance of each patch, which may depress the tracking performance when the target object is non-rigid or the background information is included in the initial bounding box. To this end, an importance-aware appearance model with respect to the target patches and background patches is built, which is able to adaptively evaluate the importance of each target/background patch by means of the local self-similarity. In addition, we propose a novel bi-directional multi-voting scheme, which integrates a multi-voting scheme and the two-side agreement scheme, to produce a reliable target-background confidence map. Combining the importance-aware appearance model and the bi-directional multi-voting scheme, a robust patch-based tracking method is proposed. Experimental results demonstrate that the proposed tracking method outperforms other state-of-the-art methods on a set of challenging tracking tasks.

Under the background of today's concept "Internet +", the marketing of electric power enterprises develop a new trend and the reform of power marketing is in urgent need. In order to adapt to the development of times and meet the needs of business growth, this paper puts forward the concept of Power Integrated Cyber-marketing, and further constructs the Power Integrated Cyber-marketing application model to integrate the idea of power marketing information collection, processing and application during the whole process. Finally, an application example is used to make the empirical analysis.

Modern nature surveillance systems are highly automated and require distributed architectures in order to increase understanding of the environment by integrating data from various monitors and sensors, which are mounted not only on fixed stations, but also on robotic actuators. In this paper, we propose a distributed architecture of robotic surveillance cyber-physical system (RS-CPS) in the nature environment, which is designed as integration of computation and physical processes. The RS-CPS system consists of five levels including smart connection, data-to-information conversion, cyber communication, cognition, and configuration. Based on this architecture, the nature risk such as wildfires, the activities of wild animals and the climate changes in plants can be constantly monitored by the cooperation of multiple agents such as autonomous UAVs, robotic pan-tilt-zoom cameras, forest infrared cameras, songbird acoustic recorders, GPS tracking collars and ecological factor sensors located in field. The experiments demonstrate the feasibility and effectiveness of the new architecture.

In this paper, a salient region detection algorithm based on priors which estimate the likely position of background which we make it more accurate, is proposed. Instead of considering the contrast between each element (pixel or region) and their surrounding regions, we consider background cues in different ways such as PCA and sparse representation. Then the saliency value of each element (superpixel) is obtained by computing the similarity with background cues. Considering such a problem which the background template based on prior probably includes the interference of foreground, so we use graph-based manifold ranking method to eliminate the interference of foreground in background template. In a detailed experimental evaluation on large benchmark database, the results shows the proposed algorithm perform well when compare with other state-of-the-art methods and can effectively eliminate the interference of foreground in initial background templates.

We deal with Pursuit-evasion games with a high speed evader which has superiority in velocity over a group of heterogeneous pursuers in this paper. Heterogeneity in the group of pursuers is expressed as heterogeneity in the individual maximum speeds. We introduce Apollonius circles generated by every pursuer and the evader to analyse the criteria for a successful capture. Collective robots pursuit problem under unknown environment is investigated from the view of behavior-based control method, and Motor Schema-based reactive control architecture is adopted. Three basic behaviors using for pursuers are designed, namely Movetogoal, Avoidobatacle and Hunting, wherein hunting behavior is realized through a kind of reinforcement learning algorithm called Q-Learning. Pursuit of the evader is based on synthesized behavior, generated through summarizing the outputs of all behaviors weighed. Results of simulation experiments validate the effectiveness of our method.

With the emerging of cloud computing, a growing body of work is focused on cloud robotics. In this paper, robots and the other devices in home are regarded as service supplier or service consumer and are packaged as RaaS (Robot as a Service) units. We design the RaaS Unit Module and propose a proxy-based cloud infrastructure for the home service robots. In addition, by modeling the disconnection times as a Poisson Process, a simple offloading strategy about whether it is suitable to offload a task to cloud is analyzed. To verify the feasibility of this infrastructure, a specific instruction set is developed to call cloud robot's service via Lua Programming Language. By using iFLY voice cloud which is a third application, speech recognition is implemented to allow users to call robot service by voice input, which illustrates that this cloud robot can utilize cloud resources.