Computer Vision Based Indoor Fall Detection

The trend of global aging is becoming increasingly serious and the change of people’s modern lifestyle has increased in the number of empty nest. The health of the elderly has become a hot social concern. Fall is the main reason that causes injury and has a great impact on the elderly’s daily life in physical and psychological healthy. By taking the advantage of advanced sensor technology, image processing technology and computer, fall can be detected automatically, which can not only provide timely treatment for the elderly injured by falling but also reduce the possibility of death due to the delay of treatment.There are some deficiencies in the current fall detection methods, ambient device based fall detection is vulnerable to the environment and cause high error rate, while wearable device based fall detection make an influence on human’s activity. Computer vision based fall detection has the advantage of high accuracy and less human intervention. With the widely application of video surveillance, video-based fall detection has good development prospects. A computer video based fall detection algorithm is proposed in this thesis.A single Gaussian background model is used. The latest Gauss distribution of each pixel is obtained and its online mean is updated for the segmentation of foreground. The single Gaussian model has the advantage of low computational requirements and fast computing speed. By extracting and analyzing the aspect ratio, the effective area ratio and the silhouette area of the foreground, due to the invariant in the camera direction, two different kinds of foreground silhouette area are selected as the feature of fall detection.SVM is used for the classification of fall and other activities. Experiments are carried on the publicly available dataset from the university of Canada, The results indicate 93.7% of the fall detection accuracy and 6.64% of false rate and 4.8% of missing rate. The main reason for the false judgement is that some normal activities such as lie on the couch are similar to the definition of fall detection, The miss of fall occurs when people lie in a chair for a while followed by a falling on the ground and stay a short peiod of time.