Research On Target Detection And Tracking Algorithm Based On Multi-feature Fusion

Target detection and tracking methods have been proved to be an importantbranch in the field of computer vision. In recent years, as China vigorously promotingthe "Green City", intelligent video surveillance has attracted more and more attention.In the social security administration and social prevention and control,"Green City"comprehensive management information public service platform can monitor thescene for object detection and analysis. When an exception occurs, the platform canachieve automatic alarm and record the information, thus saving a lot of manpowerand material resources to accelerate the construction of urban security system. Inaddition, the target detection and tracking also plays a vital role in military guidanceand robotics research.Firstly, this paper roughly detects foreground objects. For errors caused byillumination changes, shadows and other complex environments in classical motiondetection, based on VIBE background subtraction method, we propose a shadowelimination algorithm combining color and SILTP texture to improve the targetpositioning accuracy of the region, so that the foundation for the subsequent targettracking has been laid; secondly, because of the traditional feature extraction methodis slow and often loses track of the target in complex environments, an improvedhybrid texture combining LBP feature with FAST corner algorithm is proposed torealize the goals of accurately tracking; finally, in the field of target trackingalgorithm, aiming at the disadvantages of the original Meanshift algorithm (thebandwidth of kernel function is unable to adapt to the target fixed scale changes, thesimilarity matching uses only a target position in former frame instead predicting thetarget position, and the characteristic discrimination based on statistical color fallsextremely sharply under similar backgrounds), the paper introduces the motiondetection of interested region in order to remove the interference of background anduses Kalman filters to predict target locations. Compared with the traditionalmean-shift algorithm, while maintaining the amount of computation, the improvedalgorithm has solved the target lost-track due to large block or backgroundinterference. In this paper, on the VS2010platform, OPENCV library is used to realize thesimulation of the proposed algorithm. After verification, the algorithm proposed inthis paper can perform the task of target detection and tracking in real time andachieve a proper multi-target tracking.