Moving target detection method based on differential convolutional neural network

Abstract

The invention provides a moving target detection method based on a differential convolutional neural network, which detects a moving target by constructing the convolutional neural network, and has the advantages that the current frame of a video and a corresponding background frame are differentiated to obtain a differential image, the differential image is used as the input of a model, the input of the model does not contain scene information for instance the background. The convolutional neural network only learns the change of the scene without specific information about the scene, so that the convolutional neural network has relatively strong generalization ability, videos of the same type can be detected only by training once, the application range of the convolutional neural network model is greatly expanded, and the convolutional neural network has good application prospects. Through testing, the F-measure value is 0.2-0.3 when an existing moving target detection model based on deep learning is used for testing videos irrelevant to training videos, and the F-measure value is 0.73 when the moving target detection model based on deep learning is used for testing videos irrelevant to training videos.

Description

technical field [0001] The invention belongs to the technical field of video monitoring, and in particular relates to a moving target detection method based on a differential convolutional neural network. Background technique [0002] In recent years, video has been widely used. However, current video surveillance is just a simple record. In recent years, intelligent video surveillance has gradually become one of the core technologies in the security field. Moving object detection is an important foundation of intelligent video surveillance, and it has an important impact on object modeling, tracking and recognition. [0003] Moving object detection in dynamic background is still a challenging task. With the continuous efforts of scholars, many methods have been proposed. These methods can be classified into the following categories: deep learning methods, pixel-based methods, and subspace learning methods. These methods have achieved satisfactory practical results, esp...

AI Technical Summary

Problems solved by technology

However, these deep learning methods have a big flaw, that is, poor generalization ability

They can only work with data related to the training data