Research On Human Object Tracking Technology In Complex Field Environment

This paper focuses on algorithms for tracking human targets in complex field environments.Thealgorithms specifically studied in this paper include two types,one of which is traditional object tracking algorithms and the emerging emerging deep learning-based object tracking algorithms.The complex field environment includes similar backgrounds and occlusion backgrounds.The paper first introduces the theoretical basic knowledge of the traditional object tracking algorithm,then selects the most classic MeanShift algorithm for research.The object tracking application of MeanShift algorithm in similar background environment and occlusion background environment is studied.The MeanShift algorithm is improved,and the multi-feature extraction method is integrated to make it accurately track the object human body in a similar background environment.Besides,the Kalman prediction mechanism is added to the MeanShift algorithm,so that the object can be accurately predicted when the tracking object is occluded,and the object can be continuously tracked.Research experiments were performed in these two complex background environments,and the traditional MeanShift algorithm and the two improved algorithms were compared and analyzed respectively.Object tracking based on deep learning is one of the most popular methods of object tracking.This paper first introduces the basic knowledge based on the deep learning network model and introduces a twin network model.Then we select the GOTURN model based on the twin structure to study the object tracking.Under the similar background environment,the GOTURN model is used for object tracking,and the experimental results are compared with the MeanShift algorithm based on multi-feature fusion.In the occlusion background environment,the GOTURN model adds a TLD detection module to solve the occlusion problem,and the experimental results are compared with the Kalan filter-based MeanShift algorithm.