Study On The Optimization Method Of Target Tracking Under UAV Platform

Target tracking has always been one of the important research directions in computational vision.With the continuous research on target tracking algorithms in fixed viewpoint and the continuous popularization of UAV iteration,people start to focus on the optimization of target tracking algorithms in UAV viewpoint.When target tracking is performed under UAV platform,the corresponding algorithm model needs to be as simple as possible on the basis of guaranteed performance due to its insufficient arithmetic power,high real-time requirement and complex tracking situation.In order to meet the real-time tracking requirements of UAV platforms,this paper uses traditional features to optimize the relevant filtered target tracking algorithm,and the main research includes the following aspects:Ⅰ.To address the problems of complex background,fast motion,scale shift,camera motion and algorithm tracking loss and drift in low frame rate scenes,we propose a correlation filtering target tracking algorithm based on likelihood detection and multifeature fusion,which uses likelihood detection for full-frame target detection to solve the problem of tracking failure due to targets beyond the search range in fast motion and low frame rate scenes.In order to improve the tracking performance in complex scenes,different weights are assigned according to the tracking effect of different features,and multi-feature dynamic weighting fusion is performed to improve the accuracy of the tracking algorithm in complex scenes.Ⅱ.To address the problem of object occlusion and field of view disappearance of the target that often occurs during tracking from the UAV viewpoint,we design a correlation filtering algorithm based on the chunking confidence mechanism and re-detection search to establish a multi-parameter combined occlusion detection mechanism.When an occlusion is detected,a temporal consistency term is introduced in the template update to increase the similarity between the current model and the past model to improve the robustness of the algorithm;and the object is quadratically blocked after an occlusion is detected,and the degree of occlusion is discriminated by calculating the confidence parameters of the sub-blocks to classify the degree of occlusion into partial and complete occlusion;for partial occlusion,the global filter and the sub-correlation filter are For partial occlusion,the global filter and sub-correlation filter are weighted and fused to get the predicted position of the target;for complete occlusion,all tracking will be stopped and the full-frame search will be performed with the unoccluded target template until the target reappears to improve the re-tracking rate after the object is completely occluded or disappears.According to the problems of complex background,fast motion,scale transformation,camera motion and low frame rate in target tracking under UAV scenes,video sequences with these comprehensive scenes are selected on the UAV123 standard UAV dataset to verify the performance of the optimized algorithm in this paper;further algorithm optimization is carried out for the occlusion scenes,and the optimized target tracking algorithm is compared with the high-performance tracking algorithm on the public dataset for experimental analysis to objectively evaluate the target tracking effect of the proposed algorithm in the UAV perspective.