Human Action Analysis And Recognition Based On Binocular Vision

The degree of social civilization is closely related with human self. According tothe social psychological perspective, people pay more and more attention onthemselves and their environment, which will encourage people to explore anobservation mode to reflect itself and the surrounding circumstances. Hence, humanmotion analysis and recognition emerges as the times requiring. With the help ofhuman motion analysis and recognition, we can do human motion analysis andsimulation, digital entertainment, medical services, and image precise positioning et al,so it has enormous economic and social value. With the cost of computer hardwaredeclining and the computing power improving, human motion recognition has beendeveloped unprecedentedly. In the computer vision field, as it relates to many subjects,such as the visual computing, machine learning, and digital image processing, theresearch of computer vision will be full of challenges. In recent years, people wereinterested in studying and analyzing the movement of human body from the video,including the identification of simple movements, such as the person’s posture, facialexpressions, and so on. However, although such research method is simple, yet it isnot accurate in the analysis process. Meanwhile, another existed research methodsshowed space and time complexity.This article is to address the problem, and proposed a new approach-based onthe eyes of human action recognition. The current study and binocular researchfocuses on the spatial information recovery. With the help of the eyes of advantage,starting from the disparity map feature extraction characterizes the relevant humanaction. This feature not only contain local information also contains globalinformation to be able to better describe the behavior has a strong anti-jamming andanti-noise ability, strong robustness. Finally, the use of machine learning algorithms:Naive Bayes classifier, principal component analysis and support vector machine tobuild a human action recognition model. The main work of this paper is as follows:(1)Occasions and the application of technology, research and analysis of thedisparity map, summarize and compare some of today’s popular disparity mapgeneration technology. This paper mainly relates to the analysis of the three stereomatching algorithms: BM, SGBM and GC. Through comparing of matching effect,the compromise choice SGBM is the matching algorithm.(2)On feature extraction, focusing on the disparity map processing. In somecases, the generated disparity map grayscale unevenly distributed, the paper takenhistogram equalization method to handle, so that the gray uniform distribution. At thesame time after the equalization process, as it features the disparity map grayhistogram, can also be a good characterization of the human action.(3)Extracted based on the characteristics of the disparity map to create aselection of the more popular machine learning algorithm-SVM recognition model. In the process of building the model, used LIBSVM package, quick and easy solution tothe classification of human action recognition. It also identified four common choicesof the kernel function of the linear, polynomial, radial basis and S-shaped function.