Dynamic Object Detection And Tracking Of Unmanned Ground Vehicles

Intelligent vehicle is an intelligent agent composed of environmental perception, planning and decision-making and operation control. Its research involves machinery, kinematics and dynamics, electronics, computer, information processing, control and artificial intelligence, and other science and technology field. Intelligent vehicle has broad prospect of application in military field, unmanned and harsh environment work. In the domain of unmanned ground vehicle, dynamic object detection and tracking has always been the research focus.This paper introduces the design and realization of the unmanned ground vehicle software platform which includes device driver layer, perception layer, function modules, algorithm layer and algorithm management layer. The use of multi-thread model effectively improves the efficiency in the use of resources. In order to meet the demand of real-time performance, the system of unmanned ground vehicle will be split into two parts:the server is in charge of the sensor data collection and the output, the client receives environmental perception data and make algorithm decisions. The structure of the distributed system effectively improves the system efficiency.Dynamic object detection and tracking is the key problem in the field of unmanned ground vehicle. Because pedestrian is the main obstacle in the campus, human targets are detected especially in this paper. Traditionally, monocular vision was deployed for searching and tracking human targets in the whole image at a high computation cost. In the process of human target detections, candidates are firstly extracted by lasers, and then confirmed with vision algorithm. Three laser joint data enlarges the scanning area and increases the accuracy of detection. In the vision algorithm, this paper uses C4 detector whose image characteristics is the CENTRIST visual descriptor. The detector could detect human targets with the secondary cascade support vector machine rapidly. Experimental results show the validity and real-time performance of the proposed methods.In tracking algorithm, this paper uses the uniform linear motion model as the motion model of human targets. Kalman Filter is used for the forecasts and the estimates. This paper sets up the color model based on HSV for each detected human target. The discarding of V component could reduce the influence of light changing effectively. In the process of tracking, human candidates are firstly detected by the laser and vision. Then the target's location and its HSV color model are used for the matching between candidate targets and known targets. The experimental results show the efficiency and robustness of the proposed method.