Multi-peak value judgment-based dim small target tracking loss redetection method

Abstract

The invention discloses a multi-peak value judgment-based dim small target tracking loss redetection method. The method comprises the steps of selecting a search domain of target redetection in a to-be-detected image after a tracked target is lost; calculating a grayscale Gaussian distribution feature F of a frame image before the tracked target is lost, performing frequency domain response valueanalysis on the grayscale Gaussian distribution feature F in the search domain to obtain a frequency domain response graph, taking the size of the tracked target as a boundary, extracting 4-7 regionscomprising local peak values in sequence from the frequency domain response graph and recording peak value information of the regions; replacing tracked target image information in the frame image before loss to the central position of the search domain, and calculating the grayscale Gaussian distribution feature F of the tracked target and a response value delta of the central position of the search domain after replacement; comparing difference values between the 4-7 peak value information with the response value delta, and searching for a peak value position with the minimum difference value and a difference value ratio less than 20% to serve as position information of tracked target redetection. The method can quickly and effectively determine the position information of the lost tracked target.

Description

technical field [0001] The invention belongs to the technical field of target detection and tracking, and in particular relates to a rechecking method for weak and small target tracking loss based on multi-peak determination. Background technique [0002] Target tracking is the process of determining the target position information in continuous frame images, and it is an important content in computer vision research. According to the relationship between the camera and the moving target, target detection can be divided into target detection under static background and dynamic background. For the detection of moving objects when the background is not moving, there are mainly background subtraction method, frame difference method and optical flow method. For the target detection of background motion, the global motion parameters between adjacent frames of the video image are often obtained through the global motion compensation method, and the current image is matched with t...

AI Technical Summary

Problems solved by technology

[0003] During the target tracking process, background clutter, light and dark changes, partial or complete occlusion, target attitude changes, and rapid target movement in the video image will all lead to the failure of continuous target tracking.

Images