Fall detection method based on residual bidirectional SRU network

Abstract

The invention discloses a fall detection method based on a residual bidirectional SRU network, and the method comprises the steps: collecting image data of human behaviors and actions, and forming a sample data set of the behaviors and actions; constructing a residual bidirectional SRU network for behavior and action detection; adopting a balance simple and easy-to-separate sample control focus loss function, and optimizing the residual bidirectional SRU network; training and testing the residual bidirectional SRU network by using the sample data set to enable the residual bidirectional SRU network to reach detection precision; and acquiring a real-time image of a person, inputting the real-time image into the trained residual bidirectional SRU network, and detecting whether a falling behavior exists or not. The invention provides a new neural network model, namely a residual bidirectional SRU network, which is used for detecting the falling behavior of a person, and compared with an existing neural network model, the model has the advantages of high convergence speed, higher stability and higher precision and accuracy; the residual bidirectional SRU network solves the problem of gradient disappearance and is easy to train.

Description

technical field [0001] The invention belongs to the field of behavior recognition, and in particular relates to a fall detection method based on a residual bidirectional SRU network. Background technique [0002] With the rapid development of science and technology, artificial intelligence has been known to more and more people, and artificial intelligence is more and more widely used in people's lives, and the deep learning technology is particularly hot. Behavior recognition technology, as a sub-problem in deep learning, is very important in people's daily life. It is helpful to identify dangerous behaviors such as falls of the elderly, and plays a very important role in elderly care. [0003] In the field of intelligent recognition, people use different methods to classify and recognize human body movements according to different behaviors of the human body, such as methods based on wearable devices, posture estimation, and video positioning, and design different network ...

AI Technical Summary

Problems solved by technology

Therefore, the final trained model will not fully reflect the correctness of action classification

Images