Research On Human Activity Analysis Based On Binocular Vision

Human action analysis is a hot topic in the field of computer vision.It has a lot of applications in video surveillance,perceptual interface,motion analysis and virtual reality,etc.However,there are still some bottlenecks,which restrict the popularization of this technology,such as the problems of occlusion,ambiguity,backgrounds,and non-rigid bodies.The stereo matching and depth information acquisition based on binocular vision and the analysis of human behavior with the Convolution Neural Network problem in binocular vision are researched,we put forward some solutions and improvement measures.The main works and innovations of this paper could be summarized as follows:1.In the research of the stereo matching and the acquisition of depth information in the binocular vision,a stereo matching algorithm combining SURF based on the feature of edges and region matching was proposed.The algorithm was designed to reduce the effects of occlusion and ambiguity,and improve the accuracy of the behavior analysis algorithm with the three-dimensional depth information.The method included four parts:the calibration in binocular vision system,the moving object detection,the SURF stereo matching and region matching optimization,and the acquisition of three-dimensional depth information.After the calibration in the binocular vision by plane pattern two-step method,the moving human object was extracted by the improved background subtraction of Gaussian mixture model.In the process of stereo matching,firstly,the SURF matching of the acquired human body edge information was performed,and then the matching result was optimized with the region matching algorithm based on the limit constraint,improving the accuracy of the matching of human feature points.Finally,the three-dimensional depth information was obtained with the obtained matching point.The experiment results demonstrated that the algorithm can obtain the three-dimensional spatial coordinates of the human body accurately,and overcome the interference of occlusion and ambiguity effectively.2.In the research of human behavior analysis based on binocular vision,the human behavior analysis algorithm with convolutional neural networks(CNN)based on small sample was proposed.The Convolution Neural Network was divided into feature extraction layer and feature mapping layer.In the feature extraction layer,CNN neurons were used to extract local features.Then,the network layer made by multiple feature mapping layers was used to calculate,which made the feature extraction more accurate and reliable.According to the human behavior analysis algorithm with Convolutional Neural Networks(CNN)based on small sample,the images collected by the left and right camera under binocular vision were classified and identified using CNN method,and then the recognition results were carried the weighed fusion processing,finally the higher matching degree was obtained by regulating parameters.The experimental results showed that this algorithm was able to identify the single action and interactive action accurately,and improve the recognition rate of the human behavior analysis effectively.