The Design Of The Fall Detection System Based On Computer Vision

Fall accidents threaten people's healthy life,especially in the time that the aging population problem is increasingly prominent.To solve this issue,the researchers propose three technical solutions:wearable device based approach,ambient device based approach and computer vision,while the technique of computer vision has more advantages to detect fall in the environment of indoor room.Therefore this thesis proposes an algorithm of fall detection based on computer vision,and designs the embedded system the main function of which is fall detection and video surveillance.The algorithm of the fall detection acquires the raw video through a single camera,then it recognizes the moving target by the method of Gaussian Mixture Model and extracts some feature variables to describe the target movement.Finally the feature variables are calculated by the fall judgment mathematical model to obtain the detection results.There are some features in the algorithm design.First,it extracts two kinds of features,the dynamic features and state features,by means of two angle of moving object itself and scene environment.Second,the fall judgment mathematical model combines threshold method and machine learning method to improve the performance of the algorithm.What is more,this thesis discusses the case of occlusion handling and optimizes the object's bounding box in the stage of image preprocessing.The system is built in the model of client/server via WIFI.The client is realized by the embedded system with ARM&DSP heterogeneous dual-core processor to achieve the function of video capture,fall detection,video compression and network transmission;the server is a PC which is able to link network,alarm the fall accident and reset,decode,play,and save the video.Under this mechanism,the system can detect fall automatically and send the alarm signal to the caregivers to monitor the room on the net remotely.The results of experiments demonstrated that the accuracy rates in the condition of no occlusion,pervious to light the objects occlusion and light changing were 94.2%,91.5%,and 91.0%,which assisted the caregivers to rescue the elderly timely and effectively.