Infrared human body behavior recognition method based on MobileNetV3 network model

Abstract

The invention discloses an infrared human body behavior recognition method based on a MobileNetV3 network model. The infrared human body behavior recognition method comprises the following steps: S1,shooting an infrared video indoors; S2, selecting multiple persons to participate in shooting, and the implementing actions including sitting, standing, lying, walking, running and jumping; S3, 230 small videos are shot in total, the 230 small videos are cut into 62806 images, and the frame rate is 30 fps; s4, dividing the data sets into 88% of training sets, 10% of verification sets and 2% of test sets while dividing the data sets into six types of data sets; s5, generating a training and verification txt label file and generating seven read files to be conveniently sent to the network; S6, sending to a neural network for training; and S7, carrying out model prediction, and realizing action recognition and prediction. According to the method, the lightweight neural network model MobileNetV3 is used, the accuracy is improved while the parameters and the calculated amount are reduced, and the method can be used on various mobile terminals including mobile phones, embedded devices such as jetson nano and the like in the later period.

Description

technical field [0001] The invention relates to the field of deep learning technology, in particular to an infrared human behavior recognition method based on the MobileNetV3 network model. Background technique [0002] Human behavior recognition is an important research in the field of computer vision at present. It has a wide range of applications in many aspects such as video surveillance, smart home, and human-computer interaction. It not only has important scientific significance, but also has very critical practical value. [0003] Human behavior recognition is to obtain data sources from sensors, videos and other tools to form a reference database for subsequent network learning. Sensor data is mainly collected by accelerometers, gyroscopes, etc. to collect acceleration and angular velocity data of the human body during different actions to form one-dimensional data; video data is divided into various types, and the data sets available for download on the Internet inc...

AI Technical Summary

Problems solved by technology

[0005] Continuously increasing the depth of the network can improve the performance of the network, but with the continuous increase of the number of network layers, it will bring two main problems: one is over-fitting and insufficient training data; the other is the waste of computing resources, because in the network The output of the activation function of a large number of neurons will eventually become zero

Images