Research On Video Object Tracking Based On Neural Network Learning Model

Object tracking as a very important direction in the field of computer vision,it has been widely used in many fields such as intelligent monitoring,intelligent transportation,national defense industry and human-computer interaction,and has always been one of the hot spots of research.In recent years,object tracking has been developed by leaps and bounds,but limited by the influence of some factors,such as the shape change of the object,the change of the scale,the complexity and diversity of the background,there is still a lot of mining space for the research of the object tracking.In order to explore the application of deep learning in the field of object tracking,this paper has carried on the following research work:First of all,this paper makes a comprehensive review of the research status of the object tracking technology,and clarifies the challenge of object tracking.Through the detailed description of the object tracking method based on discriminant model,it is found that robust feature expression has a good effect on the object tracking.Secondly,this paper proposes an object tracking method based on convolution neural network learning model.In this paper,the pooling operation in convolution neural network is analyzed in detail,and the advantages and disadvantages of pooling operation are obtained.Pooling operation can improve the computational efficiency of network model to a large extent,and make the model have good robustness to local deformation,but it has lost the spatial structure of the object to a large extent,these spatial structure features have a good effect on the precise positioning of the object.In this paper,the pooling operation in the network model has been improved,as little as possible the use of pooling operation,so as to achieve the retention of more objectives of the low-level structure features.At the same time,in order to deal with the phenomenon of object tracking loss caused by occlusion and other factors,this paper adds an object detection module to the tracking model to re-detect the object when the object tracking is lost and improve the tracking performance of the model.In this paper,the proposed method is experimented on the OTB100 dataset,the experimental results show that the proposed method has a good effect in improving the tracking accuracy.Finally,this paper constructs a network model which combines the features of multi-layer convolution layer,and proposes an object tracking method which combines multi-layer features.Then,the features of each layer of convolution neural network is analyzed,the lower layer of the convolution neural network extracts some spatial structures,textures and shapes information of the object,while the higher layer extracts the more abstract semantic information of the object.In this paper,the information of the spatial structure extracted from the lower layer and the semantic information extracted by the higher layer of the convolution neural network are merged,and the fusion feature is used to classify and locate the object.The spatial structure of the low-level layer has a good effect on the precise positioning of the object.In this paper,the proposed method is analyzed on the OTB100 dataset,the experimental results show that the network framework with the low-level convolution layer can achieve more accurate positioning of the object.