Research On Object Tracking Algorithm Based On Deep Regression Network

In recent years,object tracking has become one of the hot topics in the computer vision field,which can be widely used in many practical systems,such as unmanned aerial vehicle(UAV),video surveillance,human-computer interaction and so on.The task of the object tracking algorithm is to use a given initial frame to predict the object position and size in the subsequent frames.Although many excellent object tracking algorithms have been studied in the past decade,object tracking is still an unsolved problem due to limited training samples and video challenges.In this paper,based on the deep learning algorithm,aiming at the problems of current tracking algorithms,two new algorithms for object tracking are proposed.The main contents of this paper are as follows:In view of the fact that the existing correlation filtering algorithms can not benefit from end-to-end training,and that it is challenging for deep learning algorithms to train deep networks online with one or a few samples,a deep ensemble object tracking algorithm based on temporal and spatial networks is proposed.The algorithm mainly includes four aspects: feature extraction,a deep regression network,a branch network and adaptive ensemble learning.The convolutional neural network is used for feature extraction to obtain image information.The deep regression network integrates feature extraction and a correlation filtering algorithm into a convolutional neural network for end-to-end training.The branch network is composed of a temporal network and a spatial network.The temporal and spatial networks capture the object temporal and spatial information and further refine the object position.Adaptive ensemble learning compensates for the object information deficiency and improves tracking accuracy.Experimental results on tracking benchmark datasets demonstrate that the proposed algorithm performs favorably compared with state-of-the-art tracking algorithms.Aiming at the complex model structure and a large number of trainable parameters for the above algorithm,and the previous regression model tracking algorithms can not find a balance between accuracy and speed,a deep regression object tracking algorithm based on multi-kernel convolutional aggregation is proposed.The algorithm is composed of only three convolutional layers,in which a multi-kernel convolutional aggregation layer is introduced for regression output.In this layer,multiple parallel convolutional kernels are used to capture multi-region object information to improve the overall performance of the algorithm.On this basis,the paper also proposes a scale estimation method to determine the object size by linear interpolation,which makes the predicted size in the current frame smoother by integrating the predicted size in the previous frame.Experimental results on large-scale benchmark datasets demonstrate that the proposed algorithm performs well comparable with state-of-the-art trackers and achieves fast tracking.