Tumble detection method based on deep learning 3D attitude evaluation

Abstract

The invention discloses a tumble detection method based on deep learning 3D attitude evaluation. The tumble detection method comprises the following steps of collecting binocular camera left and right images of a target person in real time through a binocular camera; calling a pre-trained human skeleton key point detection model to detect 2D human skeleton key points in the left and right images of the binocular camera; and constructing a 3D human skeleton of the person based on the parameters of the binocular camera and the 2D human skeleton key points in the left and right images of the binocular camera, and judging whether the person tumbles or not according to a preset rule. According to the invention, binocular stereo vision and deep learning are combined to carry out human skeleton key point detection, the 3D human skeleton of a person in a scene is extracted in real time, efficient and accurate tumble detection is carried out based on the 3D human skeleton, and the method is convenient to use and low in use cost.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to a fall detection method based on deep learning 3D posture evaluation. Background technique [0002] my country is rapidly entering an aging society, with a large number of elderly people. The elderly may fall due to the decline of various physical functions, illness and medication. Due to social and economic development and changes in family structure, the number of elderly families living alone in empty nests is increasing sharply. Many falls often occur indoors, such as getting up to stand, going to the toilet, taking a bath, etc. Only a few falls occur in the living room. During dangerous activities, such as climbing ladders, carrying heavy objects, etc. If a fall occurs and cannot be rescued in time, it will cause serious consequences and even death. Detecting the fall behavior of indoor human body when moving, the current main methods are basically divided into two categ...

AI Technical Summary

Problems solved by technology

If a fall occurs and cannot be rescued in time, it will cause serious consequences and even death

Detecting the fall behavior of indoor human body when moving, the current main methods are basically divided into two categories: 1. Using monocular (single camera) deep learning human skeleton key point detection, which can generate 2D human skeleton (human skeleton), Then evaluate the posture of the human body or recognize actions, detect falls, etc., but the 2D human skeleton lacks depth information (distance information), action recognition will cause ambiguity, and the detection of falls is not accurate enough, especially for fake falls. 2 1. Use wearable sensing devices to judge the speed or acceleration of the human body in multiple directions in three-dimensional space through the sensors to make a fall judgment. Although this method is used indoors, although the setting and operation are simple, the comfort of carrying is poor and other reasons cause the elderly to fall. When sleeping or taking a bath at night when falls are prone to occur, you just don’t carry it or forget to carry it for other reasons. You need to wear relevant devices and equipment, and you can’t really and effectively call the police when a fall occurs.

Images