Dangerous behavior detection method and system based on space-time double-flow convolutional neural network

Abstract

The invention relates to a dangerous behavior detection method based on a space-time double-flow convolutional neural network. The method comprises the following steps: shooting a real-time video image of a detection place; detecting whether a target object exists in an image frame of the video by using a target detection network; segmenting the real-time video into a plurality of time sequence segment images with equal time; carrying out the sparse sampling on the time sequence segment image; extracting an inter-frame optical flow from the image frame sequence, and fusing the time feature and the space feature of the video image by using a space-time double-flow convolutional neural network; inputting the fusion features into a classifier to obtain a classification result of target object behaviors, and outputting a dangerous behavior detection result. The invention further discloses a corresponding detection system. According to the method, the real-time video images of the detection place are screened before the target behavior is detected, so that the operand is reduced; according to the method, the feature information of time and space is extracted, the target behavior is detected after multi-scale fusion, the accuracy of behavior action recognition is improved, and the misjudgment rate is reduced.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a dangerous behavior detection method and system based on a spatio-temporal double-stream convolutional neural network. Background technique [0002] Many working environments have a certain degree of danger. Due to the lack of necessary supervision, the staff occasionally relax their vigilance during long-term work, do not dress in a standard manner, or work in accordance with strict operating procedures, which will cause some damage to life. Situations that pose a threat to security. [0003] In recent years, with the rapid development of deep learning and computer vision technology, more and more progress has been made in the image field. For example, real-time, fast, efficient and accurate detection has been achieved in the field of target recognition. Human action detection has also received more attention, and many detection methods for action recognition have em...

AI Technical Summary

Problems solved by technology

[0004] The technical problem of the present invention is that the existing action recognition method using a neural network has high complexity, a large amount of calculation, and a high false positive rate, and most of the methods cannot achieve real-time detection and cannot meet the real-time early warning of dangerous actions

Images