The Algorithms Research Of Multi-target Tracking System Based On DSP

Multi-target recognizing and tracking technique is widely applied to military field such as industries, navigation, especially weapon control and guiding system.This thesis studied the algorithm of multi-target recognizing, tracking technique and has got the experimental result. On the premise of meeting the requirement of real-time processing, this system conquers environment and noise interference, and can track the given targets continuously and steadily.The whole system is classified into two stages: multi-target capture and multi-target tracking. Multi-target capture is also classified into two parts: multi-target search and multi-target detect. The operation of the first part is to remove quiescent background from the image. This thesis combines frame-difference and frame-accumulation, adopts a method named improved accumulative difference image (ADI) basis of adaptive quiescent substraction, and scans all the mobile objects. In the second part, it provided a two-step detecting method: from roughness to precision. First it finds roughly all the possible mobile targets'places with image mark algorithm, then it uses one-dimensional maximal between-class variance image segmentation associated with region growing algorithm to find their accurate locations.In the stage of multi-target tracking, the strategy of multi-target tracking is adaptive multi-digital-tracking window tracking. Each target is adaptively tracked by a digital window, all the windows are controlled with PRI strategy. The tracking algorithm is prediction , it predicts objects'trail, speed and the window's location with prediction algorithm. The thesis simulates and compares three algorithms with Matlab: Linear Approximation, standard Kalman Filter and Sage-Husa Adaptive Filter, analyses the principle of selecting optimum algorithm according to environment and system real time. This thesis's algorithms have also dealt with some special instances such as some objects are hided, lost, new object appears and so on.In multi-target capture and tracking process, target recognition is necessary. This thesis uses object's feature information matching and template matching method, real-time update feature information and template according to the threshold set. It can recognize targets of each working state adaptively. The software system use linked list combined with structure of object feature information to program, effectively resolved these problems such as many objects, mixed information, difficult looking up and managing.This thesis successfully validated multi-target recognizing and tracking algorithm in the computer and DSP system. The result indicates that the system can exactly capture many objects within a certain range of error; and can continuously tract them within a certain range of objects'speed. It proves that these algorithms have feasibility.