Multi-target Tracking Research Based On Support Vector Machines

Multi-target tracking based on multi-sensor information fusion is the leading technology in target tracking research now, which can combine multi-sensor information to increase the precision of target state estimation, and has more advantages than the case using single sensor.Due to its broad perspective of applications in military and civil industries, study on multi-target tracking based on multi-sensor information fusion has been paid more attention by researchers in the world. One of the key problems in the study is data association under the environment with dense clutter and high-level maneuvering targets. In this paper, the following facets will be studied:1. A new data association algorithm based on multiple features and support vector machines for multi-target tracking is presented. In the new algorithm, the filtering measurement innovation is considered to be the input of the support vector machines for judging the association between each target measurement and track. According to the radar tracking system, the accurate tracking for multi-targets in heavily cluttered environment is implemented.2. For multi-target tracking system based on multi-sensor information fusion, a track-track data association algorithm based on support vector machines is proposed. In this algorithm, the difference between each tracking of different sensor are calculated, and as inputs are sent into the support vector machines to for judging the association between track and track.3. For the radar/infrared image fused tracking system, a extend Kalman filter combines with support vector machines to form a closed loop. The features related to the target maneuver are extracted from radar and infrared sensor. and as inputs are sent into the Support Vector Machines firstly, and then, the target's maneuver inputs are estimated secondly, so that, the accurate tracking is achieved finally.