Research And Application Of Based On Object Patch And Global Video Tracking Method

In recent years,with the improvement of hardware technology,object tracking has made great progress.Object tracking refers to calculating the location and size of a target object on each frame by analyzing the video sequence,which will lay the foundation for later tasks such as object recognition and behavior analysis.However,location moving,rotating,zooming and other complicated conditions may come out when object moves in real scene.Meanwhile,the complex environment,the external illumination change and other factors contribute to the uncertain results of object tracking.Based on kernel correlation filtering algorithm,this paper proposes a tracking algorithm according to partial target based on Lazy Interaction and Irregular Patches;and combined with another tracking algorithm for holistic target based on convolutional sparse coding.These two methods are applied to cows tracking.The main contents are as follows:(1)Based on partial target tracking,this paper designs a tracking method based on Lazy Interaction and Irregular Patches.Firstly,dividing the targets into many irregular target patches by simple lazy interactive process.Then,each target patches is tracked based on kernel correlation filtering.At the same time,updating the patches model to satisfy the constant changes of the target and the environment.When a simple update fails to deal with target transformation,the lazy interactive process resampling for the related patches can build a more accurate target patch model.Finally,the target position is computed by the Houghvoting scheme according to the position of all the target patches on new frame.Evaluating different tracking methods by video on the Tracking Benchmark.Experimental results show that this method can obtain more accurate tracking results when dealing with illumination changes,rotations,and complex backgrounds.(2)Based on the overall object tracking,object tracking algorithm based on Convolutional Sparse Coding is proposed in this paper.Firstly,the method expands the target area by 2.5 times to form an expansion area,and the extended region is divided into a smooth component and 4 residual components with different fitting degrees based on the Convolution Sparse Coding.Then,the smooth component will be initialized modeled and tracked based on kernel correlation filtering algorithm.According to the initial value of the target area and the linear combination of 4 residual components with different fitting degrees,the detail part of the apparent model is initialized.Followed with the detailed tracking results between the sample and the model according to overlap rate and the Euclidean distance.Finally,tracking the results in the linear combination smoothing part and detail part,and determining the final position of the target in the new frame.Tracking Benchmark video sequence tracking results show that: Compared with the existing mainstream tracking algorithms,this algorithm can provide stable and accurate tracking results when light changes,beyond the visual range and occlusion happened.(3)Object tracking is applied in many areas,such as intelligent video surveillance,unmanned driving and so on.This paper applies object tracking to the field of agriculture,that is,achieving the cow tracking.Aiming at the challenge factors such as rotation,complicated background and postural changes of cows during the movement process,this paper uses lazily interactive mode to track discrete random block guided target tracking algorithm.For the motion blur,occlusion,out-of-view and other challenge factors,this paper applies the convolutional sparse coding based tracking algorithm to track it.This will lay the foundation for subsequent analysis of cow behaviors.