Non-specific Target Tracking Algorithm Based On Deep Learning

With the development of society,machine vision has been applied in the fields of social production and life.Target tracking is one of the components.To calculate the position information of the target,mark the target image in the video.The non-specific target tracking studied in this paper refers to the use of algorithms to track the target categories without restricting them.In recent years,the target tracking algorithm has emerged in an endless stream,but the traditional algorithm cannot accurately track the target when the target has morphological changes,illumination changes and occlusion.,the scene of the target tracking task is complex and variable,moreover solving target tracking in complex environments has become a mainstream task.This paper proposes a neural network with combined convolution structure for target feature extraction.The main research results of this paper is as follows:(1)Design a combined neural network structure for convolution.Optimize the convolutional layer structure,reduce the parameters that the neural network needs to train,optimize the network transmission efficiency,and improve the recognition accuracy.(2)Using migration learning for network training,to solve the problem that the neural network cannot be trained because of the lack of target tracking data sets.(3)Using sample image generation,difficult sample mining and model updating,the target tracking algorithm is optimized to improve the recognition accuracy of the target in morphological changes,illumination changes and occlusion.Experimental verification of the combined convolutional neural network algorithm using the OTB-100 target tracking test data set.The experimental test accuracy is 0.833,and the success rate is 0.608.Contrast experiments using six other target tracking algorithms.Qualitative analysis and quantitative analysis of experimental results.The results show the tracking effect of the algorithm in this paper have high accuracy,stable effect,and the target position information can be accurately calculated under common interference.