Research on Human Behavior Recognition in Static Images Based on Attitude Adaptive
Download as PDF
Haijing Zhou, Shaofeng Han
human behavior recognition in static images is one of the hot research directions in the field of image processing. Correct recognition of human behavior in static images is helpful for image classification, retrieval, video monitoring and human tracking applications. In this paper, firstly, the adaptive gaussian mixture background modeling and morphological method are used to detect the edge using canny operator to realize the feature extraction of the target human body contour. Then, the human body is divided into pose parts by combining fourier descriptors based on centroid edge distance, k-means clustering algorithm and svm classifier. Image regions with similar structures are found by distance measurement method as positive data samples, and the classification model is obtained by training. Experimental results show that svm classification is more accurate and does not require repeated iterative training. It can effectively identify complex human motion behaviors in video and improve the recognition rate of behaviors.
Attitude Recognition; Fourier Descriptors; Support Vector Machine; Feature Extraction