Target tracking method for aerial image

Abstract

The invention discloses a target tracking method for an aerial image. The target tracking method comprises the steps: S100, executing an incremental search strategy around a target position of a previous frame; S200, enabling the sampled sample to pass through a twin network and a region recommendation network to obtain a target of a prediction position; S300, judging whether prediction succeeds or not by calculating the similarity between the target at the prediction position and the target template; if the prediction succeeds, storing the specific target position for the next prediction; when prediction fails, if n frames of prediction of the system in continuous N frames fail, judging that the target is lost, and entering the step S400, otherwise, judging that the target is detected mistakenly; and S400, initializing the detection frame, placing the detection frame at the central point of the current frame image, executing an incremental search strategy, and performing detection andtracking again until the target is repositioned. According to the target tracking method, the re-detection speed after the target is shielded is obviously improved, and the error sampling calculationof the detection frame is reduced, and rapid re-positioning can be realized.

Description

technical field [0001] The invention belongs to the technical field of image processing and computer vision, and in particular relates to a target tracking method for aerial images. Background technique [0002] Target tracking is an important research direction in computer vision. As an important research content of video information analysis, its main purpose is to estimate the motion trajectory of the target of interest in the video sequence, and automatically obtain the target object in the entire video sequence. It provides an important research basis for more advanced computer vision tasks such as scene recognition, action recognition, and unmanned driving. [0003] Modern object tracking can be roughly divided into two branches. The first branch is based on the correlation filter, which uses the recurrent property to train the regressor and performs the correlation operation in the frequency domain. It enables online tracking while efficiently updating filter weight...

AI Technical Summary

Problems solved by technology

However, since deep learning uses a large number of model parameters, it is difficult to apply it to the actual tracking speed.

Images