A pedestrian abnormal behavior identification method based on 3D convolution

Abstract

The invention discloses a pedestrian abnormal behavior identification method based on 3D convolution. The pedestrian abnormal behavior identification method comprises the steps of S1, creating a dataset containing abnormal behaviors such as fighting, dog walking and falling; S2, in combination with the latest video behavior identification scheme, constructing a 3D convolutional neural network considering both precision and speed; S3, preprocessing the images in the data set, and sending the preprocessed images into a 3D convolutional neural network to obtain a video abnormal behavior recognition model; And S4, inputting a tested pedestrian monitoring video, and outputting an abnormal behavior type. According to the identification method provided by the invention, the lightweight 2D convolutional network MobileNet idea is migrated to the 3D network, so that the calculation cost can be reduced on the basis of maintaining the identification performance; Meanwhile, a self-adaptive poolinglayer and a sparse time sampling strategy are adopted, so that a large amount of redundant information and fuzzy noise contained in continuous frames can be reduced.

Description

technical field [0001] The invention belongs to the technical field of video processing, and mainly relates to identification of abnormal behavior of pedestrians, specifically, a method for identifying abnormal behavior of pedestrians based on 3D convolution. Background technique [0002] Behavior recognition has a wide range of applications in real life and has aroused the interest of a large number of research teams. With the rapid development of deep learning technology in the image field, researchers began to believe that deep learning methods can also be used for tasks such as video analysis and understanding. Compared with traditional artificial feature-based methods, models using deep learning methods can automatically obtain meaningful hierarchical feature representations. However, video clips obtained from the Internet or movies are more complex than the video samples in previous standard databases, and these video clips contain a large number of motion components....

AI Technical Summary

Problems solved by technology

State-of-the-art 3D CNN models such as Res 3D and I3D build CNN models in this straightforward manner and use multi-layer 3D convolutions to learn powerful video features, achieving the highest accuracy on multiple datasets, but computational very expensive

Images