Reseach And Implementation Of Intelligent Video Analysis System Based On Multi-camera

With the rapid development of science and technology, the requirement of video intelligent analysis system continues to increase. Especially multi-camera surveillance overcomes the shortcomings of the limited range in single-camera surveillance. This thesis mainly study target handoff tracking under non-overlapping regions based on multi-camera and binocular stereo vision measure algorithms. The two algorithms both need the study of feature matching algorithm. So, achieving a more rapid and accurate matching algorithm is the direction of study.The research includes the following steps.Firstly, the existing classical feature extraction algorithms including the Harris, SUSAN, CSS, SIFT and SURF are studied and analyzed. At the same time, the simulation of classic feature extraction algorithm has been made. By comparing and analyzing the result, a final choice has been made that SURF operator is selected as the feature extraction algorithm to achieve binocular stereo vision measure system and target handover- tracking, because of its nature of remaining unchanged under the changes of light, rotation, and scale.Secondly, the feature matching is applied in the target tracking application. The target needs to be identified under multi-camera, which is the realization of handoff tracking of the same target between different fields of vision. The SURF features and color characteristic features are integrated in multi-camera tracking. Simultaneously, the features are learned continuously and volatile SURF feature points are excluded. The goal is found finally in another field of vision by comparing the correlation of SURF features. Then, target handoff tracking is achieved under multi-camera. Mean Shift tracking algorithm is achieved with adaptive scale under single camera. By calculating affine transformation matrix between two frames of video, the scaling factor can be obtained to correct the size of target window, so as to achieve the target tracking with adaptive scale.Finally, the feature matching is applied in the application of binocular stereo vision measure system. Hardware platform of DSP is built with dual cameras and cameras are calibrated to realize object measure on the platform.