Deep Learning Based And Context-aware Object Tracking

Object detection and object tracking are hot research directions in the field of video surveillance.Pedestrian detection is one of the key research directions in object detection.Object detection and object tracking are widely used in many aspects of the society.Under the background of big data era,video surveillance has been gradually popularized.For massive video data,processing massive data manually is unrealistic.How to achieve intelligent object detection and object tracking is an extremely important research topic.This paper proposes a deep learning based pedestrian detection framework.Firstly,an optimized PVANet network is used to generate the corresponding feature maps for a given video sequence,then a region proposal network is used to generate the pedestrian candidates and the corresponding scores,and finally a decision tree classifier is used to complete the classification task to determine whether the generated pedestrian candidates are pedestrians.The proposed method is tested on the pedestrian detection datasets of Caltech and INRIA,and the experimental results prove the significance of the proposed pedestrian detection framework.This paper proposes a context-aware and correlation filter based object tracking method.Firstly,this method uses different discriminant correlation filters to estimate translation and scale respectively in the end-to-end representation learning for correlation filter based tracking,so as to improve the adaptability of the tracker to the scale change of the target.Then,the method combines global context information to locate the target position accurately in the tracker,so as to improve the tracking performance in complex background.The proposed method is tested on VOT2017 dataset and OTB2013.Compared with some advanced target tracking methods,the target tracking method proposed in this paper has better performance in the process of object tracking.