Real-time behavior recognition method and system

Abstract

The embodiment of the present invention provides a real-time behavior recognition method and system, by sampling video frames in the video to be recognized, and inputting the sampling result to the learned preset convolutional neural network, the learned preset convolution neural network The neural network outputs the real-time optical flow generated by the video to be recognized, which can greatly reduce the time to obtain the real-time optical flow, and then based on the real-time optical flow, determine the category of the behavior in the video to be recognized, which can improve the speed of behavior recognition. At the same time, since the learned preset convolutional neural network is used in the embodiment of the present invention, the problem of inaccurate recognition results generated by a single calculation can be avoided. Moreover, in the embodiment of the present invention, after obtaining the sampling results, it is not necessary to save the sampling results, but to directly input the sampling results into the learned preset convolutional neural network, and there is no need to limit the storage space. Further saving the cost of behavior recognition.

Description

Technical field [0001] Embodiments of the present invention relate to computer visual technology, and more particularly to real-time behavioral identification methods and systems. Background technique [0002] At present, human behavior identification technology is an important branch and frontier technology in the field of machine visual. It can be widely used in smart video surveillance, robotic vision, human-machine interaction, game control, etc., the application market prospects are wide. [0003] The human behavior identification method in the prior art is mainly based on the following three convolutional neural network models: 1) Time and space double streamline neural network; 2) three-dimensional homix neural network; 3) In the top of the time and space dual-flow, the neur network top stacking model, such as Long short-term memory time recursive neural network. When these three convolutional neural network models are realized, they need to determine the light stream in t...

AI Technical Summary

Problems solved by technology

The local optical flow refers to the optical flow obtained by sampling video frames within a certain period of time and calculating the optical flow based on the sampled video frames. This process requires a large amount of calculation and will greatly reduce the speed of behavior recognition.

The scheme of using motion vector instead of optical flow to represent motion information will also greatly reduce the speed of behavior recognition due to the large amount of calculation.

[0004] Therefore, there is an urgent need to provide a real-time behavior recognition method to solve the slow problem of behavior recognition in the prior art

Images