Research On Object Tracking Algorithm For Power Vision Technology

With the rapid development and wide application of artificial intelligence,the close integration of the power industry with artificial intelligence technology and computer vision technology will greatly promote technological innovation in the field of power systems.Power vision technology is a power artificial intelligence technology that combines digital image processing,machine learning,pattern recognition,big data analysis and other technologies with professional knowledge in the power field to solve the problems related to vision application in the power field.It will help to promote the safe operation and reliable development of power system.In recent years,power application scenarios pose a severe challenge to the reliability and efficiency of object tracking in power vision technology.In practical application,the target tracking algorithm mainly solves two key problems:one is to meet the real-time requirements of engineering applications;Second,it can effectively restrain the influence of external factors and the change of target itself on tracking.Therefore,the design of power vision target tracking algorithm with good real-time performance and strong robustness is still a problem to be solved,which has important theoretical research value and engineering application value.Correlation filter tracking algorithms have the characteristics of good real-time performance and easy implementation,and are widely used in fields such as intelligent monitoring,human-computer interaction,and power inspection.The existing correlation filtering algorithms are prone to tracking-drift or tracking-failure in the case of the real environment,and the increasingly complex tracking model leads to large parameters,over fitting and poor real-time performance.Therefore,this thesis,which originates from two perspectives of sample optimization and evaluation learning,effectively improves the robustness and accuracy of existing algorithms while ensuring real-time performance.The main work of this thesis is as follows:(1)Aiming at problems such as model drift and target loss caused by the contamination of training samples in the correlation filter tracking algorithm,this thesis proposes a correlation filter tracking algorithm based on the PSR sample optimization based on the SRDCFDecon algorithm.The internal relationship among the sample contamination characteristics,filter response graph and tracking results are established,and the sample classification mechanism and parameter dynamic updating strategy are studied.We divide the training samples into sample sets reflecting different pollution levels through sample classification thresholds,and train filters corresponding to various sample sets;dynamically update the sample weights and filter parameters of a specific sample set according to the PSR;Finally,weighted fusion is performed on the correlation filters of various sample sets to achieve object tracking.Compared with the original algorithm,the algorithm in this thesis not only reduces the capacity of the sample set,but also improves the tracking accuracy and real-time performance of the original algorithm.(2)In order to solve the problem that many interference factors in the real tracking environment easily lead to tracking failure,tracking evaluation is taken as a component of object tracking,a correlation filter tracking algorithm named ESASRCF based on confidence evaluation learning is proposed.Firstly,a confidence evaluation model is established,which combines objective similarity,smoothness and coincidence;Then,the principle of filter template drift failure judgment and the method of filter parameter updating are studied;Finally,GMM is introduced to construct and manage the sample set,so that the filter template can be recovered accurately after tracking failure.Compared with the original algorithm,the proposed algorithm significantly improves the tracking accuracy and robustness on the premise of meeting the real-time requirements.(3)In order to test the performance of the object tracking algorithm proposed in this thesis in actual power application scenarios,a target tracking data set for power application scenarios is developed,and the effectiveness of ESASRCF algorithm is verified on the data set.Finally,from the perspective of engineering applications,a tracking test software based on related filter tracking algorithms is developed on the QT platform,which is convenient for data acquisition and target tracking in actual engineering applications.