Tumble detection method based on deep learning and network compression

Abstract

The invention provides a tumble detection method based on deep learning and network compression. A detection part of the method is composed of a posture estimation network and a circulation network, ahuman body posture model is used for obtaining position information of the head of a human body, the centers of two hips and the midpoint of the connecting line of the centers of the two hips from animage sequence, displacement of the body centers of adjacent images is calculated to form a displacement sequence. The group of displacement sequences is sent into a cyclic network to carry out fall-down identification. In order to expand to multi-angle recognition, recognition probabilities output by cameras at multiple positions are sent to an SVM classifier for voting judgment. In order to improve the recognition speed, according to the redundancy of the features output by the convolution kernel, the human body posture estimation network occupying the longest time is cut; the method is a pure visual detection method, not only overcomes the limitation of sensor detection, but also improves the speed and precision of visual detection. The tumble detection method is flexible and practical.

Description

technical field [0001] The invention relates to fall behavior detection, health monitoring and machine vision recognition technology, and provides a fall detection method based on deep learning and network compression. Background technique [0002] As one of the behaviors that can have a direct impact on the human body, falling will not only have a direct negative impact on human health, but also bring potential health risks. The most critical point is that falling itself is an important sign reflecting the deterioration of human health. For unattended old people and patients, whether their falling behavior can be found in time is directly related to their life safety. Because it needs a lot of manpower and material resources to monitor the falling behavior directly through the video. Therefore, intelligent fall detection came into being. [0003] The current intelligent fall detection mainly focuses on sensor detection and visual detection. The method based on sensor de...

AI Technical Summary

Problems solved by technology

The main problem is that the threshold needs to be set manually, and the cost is high

Images