Object Tracking Algorithm Based On Feature Fusion Of Correlation Filtering

With the rapid development of artificial intelligence,deep learning and computer technology,object tracking is a hot research issue in the field of computer vision.Excellent object tracking algorithms are emerging one after another at home and abroad.However,due to the complexity of the object tracking background and the variability of the object itself,such as background clutter,target deformation,occlusion and other factors,designing a comprehensive robust object t tracking algorithm is still a huge challenge.Among them,the object tracking algorithm based on correlation filtering has attracted much attention from researchers,and its tracking speed and accuracy far exceed other object tracking algorithms.In this dissertation,the related filtering algorithms are deeply studied,and the traditional methods are improved.The main contents are as follows:Firstly,it summarizes the research background and significance of the subject,analyzes the research status of the object tracking technology at home and abroad,collects the challenging problems to be dealt with in the object tracking field,and explains the principle of the object tracking algorithm based on correlation filtering,lists some classic tracking algorithms.Secondly,an improved object tracking algorithm based on feature fusion is proposed for the kernel correlation filtering algorithm in the background clutter and target deformation problem.The algorithm introduces the color template and obtains the filtering template and color template to calculate the sampling score.And adopting the strategy of updating the target template every 20 frames,and detecting the initialized position according to the template with the larger threshold value score as the target.The comparative experiment proves that the improved algorithm is feasible.Thirdly,aiming at the target occlusion problem,a long-term tracking algorithm based on feature fusion anti-target occlusion is proposed.An online detection mechanism is set up.When the object disappears or is occluded,the tracker parameter update is stopped,and the online detection mechanism is opened to reposition the object location.The Hamming window extracts the RGB information of the target,introduces the color attribute,and increases the object feature description.It is proved by experiments that the anti-occlusion strategy is effective and feasible.Finally,for the problem that the object tracking process cannot be initialized and the object detection method is applied to the object tracking to prevent the object from being occluded,the real-time pedestrian monitoring using the VGG16 convolution neural network SSD object detection algorithm and the object tracking algorithm is proposed.The tracking and counting system proves that the system is effective and feasible through experiments.