Research On Workpiece Visual Tracking Algorithm Based On Sparse Transforms Learning

The tracking research of video moving targets is the research field of artificial intelligence,and the tracking of workpiece motion video is a necessary means to implement monitoring in the automation operation of factory enterprises.At present,researchers have proposed many tracking algorithms for different tracking environments.These algorithms are proposed to promote the development of tracking technology.However,how to effectively track the illumination,occlusion,and pose changes in a poor environment still needs further research.In this paper,the workpiece is denoised and segmented before visual tracking,and some deficiencies in the process of denoising,segmentation and tracking are overcomed.The main research contents and innovations of this paper are as follows:(1)As the quality of light in visual measurement will directly affect the accuracy and efficiency of workpiece detection.To address the problem,an approach was proposed to denoise the workpiece image under different light sources with union-of-transforms learning.The proposed method first extracted the noise image patches from noise images,and studied the intra-cluster transform learning of the noise patches by alternated between sparse coding and transform update steps.Next,the transform sparse levels of clustering signal in each noise patch were calculated,and the minimum sparse level was selected as the sparse level of the denoising patches.Finally,denoised patches were clustered and the denoised image was estimated by the mean value of the final iterative denoised patches.The experimental results show that the proposed mothed is superior to the other algorithms in terms of both denoising effect and algorithm speed for workpiece images under different light sources,and gains better denoise performance.(2)As the poor segmentation accuracy and effect in workpiece image segmentation,a new approach was proposed to segment workpiece images in HSV color space with k-means clustering.Firstly,the RGB color space of the workpiece images was converted into the HSV color space,then the pixels of each channel were clustered by k-means in the HSV color space,and the pixels with similar colors were clustered into the same class to achieve the segmentation effect of the workpiece image.The experimental results show that the proposed method is superior to the traditional segmentation methods.(3)As the problem of pose change and occlusion in the process of workpiece tracking,an approach was proposed to tracked the object workpiece by jointing anti-sparse model and spatio-temporal context.Firstly,the candidate object was sampled by particle filter on the video frame of the workpiece,and then the new candidate target was selected and optimized by spatio-temporal context as the dictionary.Finally,the anti-sparse model is used to calculate the sparse coefficient,and the candidate target corresponding to the optimal sparse coefficient is the tracking result.The experimental results show that the proposed method has better tracking performance.There are 19 figures,7 tables and 67 references in this paper.