Motion Recognition Based On Binocular Stereo Vision In Indoor Environment

Human-computer interaction,somatosensory games,and video surveillance are all research areas of human motion recognition.They penetrate into every aspect of life.Indoors environment is the environment with the highest frequency of human activities.People spend most of their lives indoors.Therefore,the research on motion recognition in indoor environments is more close to daily life.In the solution of collecting human motion information,wearable equipment is inconvenient to wear and information leakage,while monocular camera is lack of depth information.Therefore,the motion recognition of human body is studied by binocular camera.In human motion recognition based on binocular stereo vision,foreground detection and feature extraction are two important directions.There is background noise interference in the foreground detection of human targets,and the prospect information is prone to errors.For the feature extraction of human movements,the lack of feature subsets can easily lead to confusion.So in the process of motion recognition,we analyze these two directions in detail.In order to obtain the depth map,this paper completes the acquisition of the depth map through the calibration,correction and stereo matching of the camera.At the same time,in the aspect of stereo matching,the Semi-global block matching(SGBM)algorithm with the best comprehensive effect is selected after comparison.Foreground detection can better extract the motion sequence in the image.Visual background extractor(Vibe)algorithm is used to extract the foreground of the depth image.In view of the excessive noise block of Vibe algorithm,this paper makes effective optimization by morphological processing.This paper also further improves the Vibe algorithm with the monocular color information,and obtains a more accurate and clear foreground image.In this paper,the depth information of stereo vision is added and Depth Motion Map-HOG(DMM-HOG)descriptor is used to improve the traditional feature extraction method.After foreground detection and acquisition of human motion foreground,this paper uses dense optical flow method to extract feature points.In this paper,the space-time grid is established on the tracking trajectory of human body sequence,and the traditional descriptors are used to extract features.In this paper,machine learning is used to recognize human actions.Through experiments,we find that the improved feature extraction method significantly reduces the confusion rate and improves the recognition rate of human motion.In summary,based on the analysis of binocular stereo vision,human action recognition is studied.And this paper optimizes and improves the foreground detection algorithm and feature extraction,completes the expected goal of human motion recognition.