Real-time vision target detection method robust to ambient light change

Abstract

The invention provides a real-time vision target detection method robust to an ambient light change. The method comprises a step of acquiring a historical image sequence in a video stream, a step of initializing a pixel-level mixed multivariate Gaussian background model, a step of designing a background modeling method of a spherical K-means expectation maximization algorithm and updating parameters of the background model in an online way to adapt to a change of an environment once a new image frame is obtained, a foreground target detection step of using a statistical framework determined bya color space area and calculating a latest image frame and the background model in the statistical framework to obtain a pixel area where a foreground target is located, and a step of weakening detection noise through an iterative Bayesian decision step, wherein a target contour can be enhanced in the process. According to the method, the foreground target position and contour in a visual fieldcan be accurately detected in real time in the visual field, a correct detection rate is high, a false alarm rate is low, the disturbance of a detection result caused by the ambient light change can be resisted, and the method is particularly suitable for an intelligent video surveillance system.

Description

technical field [0001] The invention relates to the technical field of video monitoring and image processing, in particular to a background model-based real-time visual target detection method that is robust to ambient light changes. Background technique [0002] Visual object detection, also often called foreground detection, is to use algorithms to detect objects such as people or vehicles that we are interested in in the scene, and mark them on the video or image. Visual object detection is the most fundamental and most important problem in the intelligent video surveillance system, and it is the basis of subsequent processing. In a practical environment, although the camera can remain still, the scene is still full of background dynamics and various noises, especially the challenges brought by unstable lighting conditions. For example, outdoors with sunlight as the light source, due to the movement of clouds and changes in the direct sunlight angle, changes in global il...

AI Technical Summary

Problems solved by technology

3D is the future development direction of image processing and computer vision, but as far as the current research status is concerned, special camera sensors are generally required to achieve 3D target detection, and 3D calibration can only be applied to a specific environment

[0010] The above-mentioned existing visual target detection algorithms generally have the following defects: the performance is not reliable enough, the accuracy of algorithm detection needs to be improved, and the anti-interference for dynamic background and illumination changes is weak

Images