Research On Vision Target Tracking In Complex-scene

In recent years, with the development of computer vision technology and expanding demand for practical applications, target tracking technology has become the focus of domestic and foreign researchers because of its applications in areas such as safety monitoring, robot navigation, intelligent transportation and unmanned potential. However, due to the complex application scenarios, random noise, the diversity of targets, target tracking has always been a challenging task, especially for complex moving target tracking scenario which has many theoretical and technical issues that must be resolved. Its further study will have important theoretical and practical significance.In this thesis, target tracking in complex scenarios was studied. Algorithms based on a single fixed camera tracking were discussed in detail, and a complex algorithm of target tracking in moving scenarios was proposed, and a smooth tracking algorithm to keep the targets in the central region of the camera view was designed. The target relay tracking based on multi-camera was studied, and then we raised an algorithm of target relay tracking based on the detection and the SIFT features of targets. The major contributions of this thesis are summarized as follows.In the moving target tracking part, target model and matching method was introduced, the tracking system based on mean-shift which used kernel function-weighted color histogram to characterize the target was studied, This method had good adaptability to target round, transmogrification, and part-occlusion and the calculation was simple, which could be used in real time. However, it had poor performance in tracking targets with maneuverable movement. The tracking method based on mean shift and the Kalman filter was proposed to improve the performance of the current mean shift tracker. Target model could be established by the difference of target color and shape, and then the movement of target would be judged by the changes of matching degree. The Kalman filter prediction result and the mean shift tracking result were linearly weighted by adopting different scale factors. In the meanwhile, the target feature model was updated online to improve the tracking reliability. The experiments showed that the improved algorithm was real-time and robust.In the active tracking of single camera part, the software and the hardware of the system were separately studied. In pan-tilt control parameter calculation part, according to the relative position of the target in the camera scene and structural parameter, the camera moving control model was built, and the control parameter of pan-tilt direction was calculated; In pan-tilt control strategy part, a whole control algorithm was designed to control the head speed by analyzing the position of target in the image and information. According to the control direction and control speed obtained, the head speed and head direction could be changed by changing the pan-tilt control protocol to keep the tracking target in the center of the camera scene. The experiments showed that the algorithm could achieve active smoothing tracking well.In the multi-camera relay tracking part, some key issues and technique in multi-camera relay tracking was studied. Methods for target handoff based on detection and SIFT features of moving object were separately introduced. If the background of the target was simple, the effect was good by detection of moving object; When the environment was complex, a method for Target Handoff based on detection and SIFT features was proposed; The experimental result showed that this algorithm had the stability of SIFT feature as well as the advantages such as simple calculation, high accuracy and good real-time feature of the view line transfer algorithm, target handoff effect is good, it also helped saving the matching time.