Research On Indoor Fall Detection In Visual Shielding Video Surveillance

With the continuous improvement of economic life and medical treatment in nowadays society,the aging of the population has become an increasingly serious social problem.As coordination declines with age,old people are more likely to fall than young people,leading to various injuries,strokes and other health problems.And many of the elderly live alone,making it difficult for them to immediately seek help if they fall unexpectedly.Therefore,the research on the subject of fall detection is of great significance.Among these studies,the fall detection system based on surveillance video has become a research hotspot in the field of smart home due to its advantages such as low equipment cost and strong real-time performance.However,current computer vision-based fall detection methods often have the problem of privacy invasion.Camera equipment installed in indoor homes constantly monitor the daily lives of the elderly living alone in real time.Although the life safety of the elderly needs video surveillance to provide protection,they also have the right to enjoy privacy protection as independent individuals.Users’ demand for privacy protection has largely hindered the practical application of this technology in real life.Based on this problem,this paper proposes a visual shielding method based on compressed sensing.Based on the analysis of compressed video data,key technologies such as moving object detection,feature extraction and classification algorithms that are suitable for compressed data are mainly studied.The research work of this paper mainly includes the following parts:(1)This paper analyzes the research status of fall detection.Especially,the fall detection technology based on video surveillance is mainly investigated.Several typical feature extraction algorithms and classification algorithms are also introduced.(2)The compressed sensing theory is introduced.According to the idea of image block compressed sensing,the measurement matrix is used to perform the same compressed sampling on video sequence images,and the single-layer perception is further extended to multi-layer perception.This process greatly reduces the number of pixels in the original image,eliminates video visual information while reducing the amount of data processing,and realizes the process of visual shielding coding,thus solving the privacy violations involved in video surveillance.(3)A feature extraction algorithm and a classifier for visual shielding data are proposed.After single-layer or multi-layer compressed sensing of video data,the image quality deteriorates sharply,and the appearance details of the original image are eliminated.However,because of the same projection processing,the motion characteristics of pixels between adjacent video sequences are preserved.Considering the above characteristics of compressed data,the feature extraction algorithm in this paper studies on dense trajectories feature algorithms.This feature fuses the spatiotemporal information of the sampling points and has the advantage of robustness.In addition,in view of the dichotomy problem of fall detection,the feature vectors of compressed video are sparsely recognized by a classifier based on the intra-class dictionary learning method.(4)In order to further improve the performance of the above algorithm,moving object detection method is used to implement the fall detection of the compressed video.First,the low-rank sparse decomposition model is used to extract the foreground in the visual shielding video,and then the subsequent steps such as feature extraction and action recognition are implemented.Experimental results on three fall databases demonstrate the performance superiority of the improved method.