Object Tracking and Collision Avoidance Using Particle Filter and Vector Field Histogram Methods

Abstract

<abstract> <bold>Abstract.</bold> Object tracking and collision avoidance are important topics in various applications of autonomous vehicles and robotics. In previous studies, we have developed a real-time stereo vision system equipped with efficient algorithms for obstacle detection and recognition. To detect and locate obstacles, the estimated three-dimensional information of each pixel is projected onto a non-linear top-view map, and the blob segmentation is used to find the obstacle candidates in the top-view map. Detected obstacles can be further recognized based on their geometrical features and histogram of oriented gradient (HOG) feature using support vector machine method. The 3D information of obstacles and the relationship of obstacles are subsequently applied to track obstacle paths and to estimate the motion model. In this study, particle filter approach was applied to offer stable tracking of detected obstacles in different frames. This method was compared with Kalman filter approach under various experimental conditions. Finally, by combining various information including color and 3D information as well as motion models, the collision avoidance algorithm was developed based on the vector field histogram approach. The output of the collision avoidance algorithm includes the direction and predicted path of the vehicle and warning signals for the driver. The performance of the collision avoidance algorithm was analyzed and compared with the method based on the A-star path-finding algorithm. Experiments were carried out under different scenarios of driving conditions and obstacle states to compare the overall performance of the system. The proposed object tracking and collision avoidance method will help in enhancing the safety of operations of agricultural vehicles.