Research On People Recognition And Tracking Based On Video Streaming

Human recognition and tracking has extensive application in intelligent video surveillance system, advanced man-machine interactive and animation issues,and also is the Frontier direction in the field of computer vision in recent years. The human bodies’ recognizing and tracking from image sequences fast and accurately is a very important and challenging job. Meanwhile accurate recognition and tracking is also the foundation of behavior’s effective understanding. The thesis using Visual C++6.0, by the help of computer vision library OpenCV, does research on human body recognition and tracking based on video streaming.First, trained a SVM classifier to classify contours. Chosen binary images with human body contours and not as positive and negative samples, extracted eigenvectors from samples with image pixels, and reduced dimensions of the eigenvectors by PCA to train the SVM classifier. Second, detected moving targets by background subtraction and segmented the target contour to be classified by SVM classifier. The results showed this method could classify human body targets fast and accurately in static background. Third, established color histogram model for the recognized human, and tracked the human body using Camshift algorithm. The combination of Kalman and Camshift algorithm optimized the iteration times of Camshift algorithm. Meanwhile, using the size of track window and the similarity of color histograms as the condition whether the target was occluded or not, finally realized the tracking for occluded human targets. Last designed a sample system for people recognition and tracking, realized people’s real-time recognition and tracking in some conditions.