Real-time fall detection method for embedded platform based on improved pose estimation algorithm

Abstract

The invention is an embedded platform real-time fall detection method based on an improved posture estimation algorithm. The method uses depth separable convolution, an attention mechanism and an inverse residual structure to build a posture estimation network, which is used in fall detection, so that the posture estimation The accuracy of the network is further improved, and the amount of parameters and calculations are greatly reduced. Calculate the distance of each joint point of the human body between different video frames to track the human body, and use the front and rear video frames to calculate the acceleration of the joint points of the human body. The relative position of the points can be used to determine whether a fall has occurred, making the attitude estimation network more suitable for deployment on embedded platforms. Deploying on the TX2 embedded platform can achieve real-time effects. The method of the present invention uses joint point coordinates and skeleton information of multiple human bodies obtained in the front and rear frames to track the human body, and the tracking of multiple people makes posture estimation more stable, and can better handle the problem of fall detection in a multi-person scene.

Description

technical field [0001] The invention belongs to the technical field of computer vision, in particular to an embedded platform real-time fall detection method based on an improved posture estimation algorithm. Background technique [0002] With the improvement of medical conditions and the improvement of living standards, the average life expectancy of the population has increased significantly, which has accelerated the process of population aging. Population aging has brought many social problems, and reducing the losses caused by accidents of the elderly is an urgent problem to be solved. According to the WHO report, an estimated 646,000 people worldwide die each year due to falls without timely treatment, among which the elderly account for the largest proportion. Among the current fall detection methods for the elderly, the method based on wearable devices has relatively large limitations and high cost; although the computer vision fall detection algorithm based on pose...

AI Technical Summary

Problems solved by technology

[0004] In view of the shortcomings of the current attitude estimation algorithm, such as too large model, too many parameters, and no tracking of the human body during actual deployment, the present invention proposes a real-time fall detection method for embedded platforms based on the improved attitude estimation algorithm, and reduces the model by improving the network structure. Size, to complete the real-time estimation of human posture on the embedded platform and the purpose of detecting human falls

Images