Smoking behavior and calling behavior identification method based on video stream

Abstract

The invention provides a smoking behavior and calling behavior identification method based on a video stream. The method comprises the following steps: S1, collecting a data set of model training; S2,installing and compiling a caffe framework supporting 3D convolution; S3, building a classification network architecture based on Tiny_darknet-3D deep learning; S4, setting hyper-parameters of the training network model; S5, directly obtaining a monitoring video stream from monitoring equipment of a to-be-monitored scene; S6, sending the video data obtained in the step S5 into a pre-trained modelin the step S3 for 3D convolution processing, and outputting a classification result; and S7, setting a threshold value according to the confidence value of the target, removing the target with a relatively low confidence value, and storing the image. According to the smoking behavior and calling behavior recognition method based on the video stream, the problems that a traditional smoking and calling detection method is prone to causing target missing detection, the false detection rate is high, an original video cannot be directly recognized, and time dimension information is lost are solved.

Description

technical field [0001] The invention belongs to the technical field of video monitoring, and in particular relates to a method for identifying smoking behavior and calling behavior based on video streams. Background technique [0002] In recent years, many public places have posted no-smoking reminders, and the demand for no-smoking inspections in factories, offices, and public transportation is increasing. In some specific scenarios, such as examination classrooms, and some office areas, gas stations or confidential institutions, the detection demand for banned mobile phones is also increasing sharply. However, manual monitoring cannot play a real-time monitoring role, or after-the-fact surveillance video screening requires a lot of manpower and time costs, and it also has the effect of repairing a dead sheep, without initiative. Real-time detection of smoking in public places with high crowd density can detect fire hazards in the first place and prevent them from happenin...

AI Technical Summary

Problems solved by technology

[0004] In view of this, the present invention proposes a method for identifying smoking behavior and calling behavior based on video streams to solve the problem that traditional detection methods for smoking and calling will cause target leakage due to the influence of lighting and other conditions in a complex environment. The detection and false detection rates are high, and the original video cannot be directly recognized, resulting in the loss of time dimension information

Images