Research On Pedestrian Detection Technology In Smart Home Surveillance Video

Pedestrian detection provides important technology support for robots,assisted driving and automatic driving,video surveillance and prejudgment of human behavior in video,and its extensive application makes it a very important branch of computer science.Recently,the concept of home automation has been popular in our daily life and has been widely utilized.Monitoring systems are common in home automation which insures the security of homes.Pedestrian detection is the main problem in video surveillance,whose algorithm is worth much attention.The development of pedestrian detection has experienced three main phases:The first one is image processing-based,including optical flow method,frame difference method and background difference method.Afterwards,in 2005,the gradient direction histogram is a big breakthrough in object detection area.HOG-based feature extraction makes pedestrians within the class have small differences in features and large differences from other classes,which is the basis of subsequent development in pedestrian detection technology.The last method is deep learning-based method.In recent years,with the advance of computer hardware,a large number of computational problems in deep learning can be solved,which leads to extensive application of deep learning in image processing.Faster R-CNN is one of the most prominent which comparative accuracy and speed.Several pedestrian detection methods are compared in this dissertation,and the problems in these methods are analyzed.Finally,Faster-CNN-based object detection method is applied in the proposed pedestrian detection algorithm.The major research results are as follows:(1)Use background modeling,frame difference method and HOG-based method to detect people in video respectively,and summarize the problems existing in every method;(2)Apply the Faster R-CNN algorithm to the detection of pedestrian in smart homes,construct the whole detection system.Besides,the speed of detection is accelerated by GPU and almost meets the real-time requirement.(3)The pedestrian detection system in home automation can achieve some targets 1)pedestrians can be detected by Faster R-CNN algorithm and the picture with pedestrians can be saved;2)these images are sent to web server;3)the images sent to web server are pushed to phones to safeguard the security of smart homes.According to experiences,this system can work well in complex scenes and meet real-time requirements.