Target detection method based on self-supervised generative adversarial learning background modeling

Abstract

In most foreground target detection scenes, the background has a relatively fixed mode. The change of brightness, the dynamic background, the movement of a camera and the like bring a lot of difficulties to a target detection algorithm. On the basis, the invention provides a target detection method based on self-supervised generative adversarial learning background modeling, and the method can effectively overcome the difficulty and can be applied to a real environment. The method comprises the following steps: firstly, forming training data by adopting a self-supervision method of automatic image completion, and carrying out self-supervision adversarial learning by using a generative adversarial neural network to construct a background reconstruction model; secondly, training a convolutional neural network, performing foreground target detection by using the difference between an original image and an image reconstructed by a background model. The method provided by the invention can be applied to contents shot by a static camera and a mobile camera, and has wide application scenes and practical application value.

Description

technical field [0001] The invention belongs to the technical field of digital image processing, in particular to a method for realizing foreground target detection in images or videos. Background technique [0002] The detection of foreground targets in images is the basis for target recognition and tracking, and is widely used in many fields such as video surveillance, home monitoring, and field environmental monitoring. There are generally three types of methods for foreground target detection, namely frame difference method, background subtraction and direct target detection. The basic idea of ​​the background subtraction method is to first establish a background model, and then subtract the background image from the current image to obtain the foreground target. [0003] After decades of development, especially with the development of deep learning since 2012, foreground object detection technology has made great progress. Compared with traditional unsupervised method...

AI Technical Summary

Problems solved by technology

But the current foreground object detection technology still faces great challenges

Changes in brightness, dynamic background, camera movement, etc. all bring great difficulties to the target detection algorithm.

For example, the current convolutional neural network-based method that can achieve the best results on videos captured by static cameras has great difficulties when applied to videos captured by moving cameras.

[0004] The present invention proposes a method for background modeling based on self-supervised generative confrontation learning to solve the difficulties of brightness changes, background dynamics and camera movement in foreground target detection.

Images