Moving Object Detection Method Based on Color Probability of Moving Area

Abstract

The invention discloses a moving target detection method based on the color probability of the moving area. The convolution feature of the input image pair is extracted based on the convolutional neural network; the significant motion area between the image pair is obtained by image feature difference; the color histogram statistics are performed on the input image and the significant motion area respectively, and according to the maximum a posteriori probability formula Solve the probability that each color histogram belongs to the target; use the probability to assign the input image pixel by pixel to obtain the target probability map; generate the initial moving target detection frame on the target probability map based on the image feature difference result, and use the gradient descent method to solve the optimal detection frame to get the moving target detection result. The invention can effectively identify the moving target in the scene, and represent the position of the target in the image in the form of a rectangular frame. The method has fast computing speed and high accuracy. It can be used as the front end of various algorithms, such as object tracking and pedestrian re-detection. It has important application significance in practice.

Description

technical field [0001] The invention belongs to the field of computer vision, relates to target detection, motion recognition, feature extraction and other technologies, in particular to a moving target detection method based on the color probability of a moving area. Background technique [0002] Moving object detection is to extract the moving object in the scene and detect its position in the scene. It has important application value in monitoring security, target recognition, human-computer interaction and other fields. Classical moving object detection methods include frame difference method, background elimination method and optical flow method. Among them, the background elimination method needs to model the scene, while the optical flow method needs to solve the motion of each pixel in the image. Therefore, these two methods are time-consuming and difficult to meet the current real-time detection needs. Although the frame difference method is fast, it is susceptibl...

AI Technical Summary

Problems solved by technology

Among them, saliency detection is to identify the most visually unique objects or regions in the image, and then segment them from the background. This type of method pays more attention to the semantic information of the image and ignores the motion information of the image, so the detection results sometimes do not match expectations. ; and object-like detection is to extract the area where a certain object is located in the scene. Many methods have achieved fast and accurate detection results, but this type of method needs to know the type of the detected object in advance, so in practical applications still subject to many restrictions

Images