Human Motion Representation And Recognition Based On Video

As the great improvement in operation and storage of computer, computer vision has been gradually applied to achieve complex and advanced functions. Action detection and recognition based on video is one of the most important research direction in the field of computer vison. This subject combine technologies with pattern recognition, article intelligence and image processing,etc. It can been applied on video surveillance, exception detection and human-computer interaction, so it’s valuable to research on this field. The thesis will force on human action recognition based on video.Firstly, Frame subtraction,Optical flow and Background difference have been used to extract the target of the image for movement dection and pretreatment.then processing extraction results by Mathematical morphology and discussing the various edge extraction method for contour detection of moving targetsIn feature extraction, this article introduces PCA method, mesh method and focuses on a improved human skeleton method as extraction feature vector. It has been put forward a extremum unification method and vector quantization sequence by rotated matching. Finally, it analyzed matching results and the correct rate, summarized and compared the advantages and disadvantages of the various feature extraction methods.HMM has been used as classification algorithm to classify human action gesture and using videos form KTH and Weizmann human movement datasets as test samples. The sample sequences are grouping for training and recognition after feature extraction.The experimental results show that the HMM action recognition method based on human star skeleton feature extraction which be put forward in this paper has achieved good matching and recognition results. The algorithm with low-cost and fast speed has good universality and robustness.