Research On Indoor Fall Detection Algorithm Based On Human-object Interaction Detection

In recent years,the proportion of the elderly population has increased rapidly,and the problem of nursing services for the elderly group has become increasingly prominent.Family pension is the main pension mode in China.Indoor accidental fall is an important factor affecting the health of the elderly,the detection of the fall event of the elderly indoors is beneficial to the rescue of the elderly quickly and reduce the secondary injury caused by the fall.In this paper,fall events in indoor situations are studied,and an indoor fall detection algorithm is designed based on the theory of human-object interaction detection to solve the problems of missed detection and false detection caused by ignoring scene information in fall detection.The main work of this paper is as follows:(1)To solve the problem of low accuracy and slow speed of indoor object detection by existing algorithms,an improved target detection algorithm based on YOLOv3 is proposed.Firstly,the feature extraction part of the network is built by using the inverted residual block and attention module,and Mobile Netv3 is used as the feature extraction layer to improve the computing speed of the network.Secondly,the loss function of the network is improved by introducing CIo U,which is used to improve the accuracy of indoor object detection.Then,the appropriate anchor value is calculated by clustering algorithm.Finally,low-light image enhancement algorithm based on haze removal algorithm is used to process low-light image to improve the recognition ability of the network in low-light environment.Experimental results show that m AP of the improved network is improved by 2.29%,and the detection speed is improved to 1.7 times of the previous.(2)In order to reduce the false detection rate in the identification of fall event,and make full use of the information like the location of people and objects in the scene,an image classification fall detection scheme based on the theory of human-object interaction detection was proposed.Firstly,three data set generation methods are formulated,and compared to determine the optimal generation method;Then,build minif?VGG network is used to reduce the detection time of classification network.Finally,the use of weight loss function is used to deal with the problem of unbalanced samples in the data set to improve the detection accuracy of the classification network.The experimental results show that the data generated by the original image information,human posture information and position information has the best classification effect.The accuracy of classification network minif?VGG reached 79.86%,and the F1 score of fall category reached 95.34%,meanwhile the detection speed of the network was improved.(3)The indoor fall detection platform was built according to the actual needs,and the target detection algorithm,attitude estimation algorithm and fall detection algorithm were integrated into the platform,and study the conditions for determining the low-light image to improve the practicability of the platform.The experiment shows that the platform has a good detection effect,and the detection accuracy rate of fall events reaches more than 95%,which is conducive to improving the quality of the elderly care service system.