2. Key Laboratory of Opto-Electronics Information Technology of Ministry of Education, Tianjin University, Tianjin 300072, China
Overview: Intelligent vehicle refers to the new type of car which integrates a variety of technologies, including environmental perception, path planning, decision-making, controlling, etc., which carries advanced vehicle sensor, controller, actuator and other devices, can realize the car with X (people, vehicles, road, cloud, etc.) of information exchange and sharing to achieve safety, high efficiency, energy saving, and ultimately. Environmental perception is the technology which detecting vehicle environment information relies on the on-board sensors including vehicle vision sensors, LiDAR, millimeter wave radar, global positioning system (GPS), INS system and ultrasonic wave radar. In order to ensure the accuracy and stability of environmental perception of intelligent vehicle, it is necessary to use intelligent vehicle on-board sensors to detect and track the objects in the passable area. This paper puts forward a kind of object detection and tracking algorithm based on the LiDAR and camera information fusion. Firstly, this algorithm uses the LiDAR point cloud data clustering method to detect the objects in the passable area and project them onto the picture to determine the tracking objects. The LiDAR point cloud data clustering method contains filtering of original point cloud data, ground detection, passable area extraction based on point cloud data reflectivity and data clustering based on DBSCAN algorithm. After the object has been determined, this algorithm uses color information to track the object in the image sequence. Since object tracking algorithm based on image is easily influenced by light, shade and background interference, this algorithm uses LiDAR point cloud to modify tracking results in the process of tracking. The tracking strategy is: first, place N initial particles uniformly at the target position; second, calculate the similarity between the current moment particles and the previous moment particles according to the Bhattacharyya coefficient; third, resample particles according to similarity; finally, since LiDAR point cloud can be projected onto picture, calculate the object position by combining the particles and the point cloud through the algorithm. At the end of paper, this paper uses KITTI data set to test and verify the algorithm. KITTI dataset is established by Germany Karlsruhe Institute of Technology and Technology Research Institute in the United States, which is currently the largest data of computer vision algorithm for automatic driving scenarios evaluation. The experiment used a computer with 4 GB memory as the experimental platform and programmed on MATLAB 2017b. In this paper, particle filter, unscented Kalman filter (UKF) and DCO-X algorithm are used as comparison algorithms to verify the effectiveness of the algorithm. Experiments show that the algorithm has a good effect in object tracking evaluation standard of X direction, Y direction errors and center position error, regional overlap and the success rate.