Reseach And Implementation Of Object Detection And Tracking System Based On Recurrent Neural Network

Object tracking is of high value in both research and practical application.Therefore,object tracking has become one of the important research fields of computer vision and has been widely used in both civilian and military applications.The intelligent tracking system can effectively reduce the working intensity and improve the efficiency of related staff.With the rapid development of artificial intelligence technology,object tracking technology is becoming more and more mature.However,object tracking in real scenes is still faced with various non-constraint conditions,for example,the appearance variation of the tracked target object,in-plane or out-of-plane rotation,the scale variation,the illumination variation of the background space,and the occlusion of the background object and the target object,etc.The change of these non-constraint conditions will affect the performance of the object tracking algorithm and the application of object tracking in real life cannot be realized.The detection and tracking task of video sequence is a typical sequential task.The temporal information of the front and rear frames can help improve the performance of the tracking algorithm.Based on this,this paper studies the problem of object detection and tracking in complex scenarios,introduces the inter-frame information of video frame,and applies the recurrent neural network to the object tracking task.We design and implement a object detection and tracking system based on recurrent neural network.In this paper,an object detection and tracking algorithm based on recurrent neural network is proposed and improved to realize the performance improvement under partial occlusion.At the same time,an interactive object detection and tracking system based on the algorithm proposed in this paper is designed and implemented.Below we will briefly introduce the main research content and innovation points of this paper.1.In order to reduce the search scope of the candidate area and the dependence of the object tracking on the detection,a direction prediction model based on Long Term Memory network(LSTM)is proposed to determine the motion state of the object and to help determine the region of interest for object detection.Through experimental verification,compared with other tracking algorithms,the tracking algorithm in this paper makes full use of the inter-frame tilming information,so that the algorithm proposed in this paper has certain advantages in both accuracy and speed.At the same time,the object detection and tracking algorithm based on motion direction prediction proposed in this paper is more robust to the scene where the target scale changes dramatically.2.In order to further reduce the dependence on object detection performance and speed up the tracking algorithm,an adaptive detection mechanism on the basis of the object detection and tracking algorithm based on the recurrent neural network is proposed.And by combining correlation filtering with deep learning,non-frame-by-frame detection is realized.The experimental results show that the correlation filtering method is helpful to improve the tracking performance,reduce the dependence on object detection and improve the tracking speed.At the same time,the object tracking algorithm based on adaptive detection proposed in this paper is more robust to the scene where the target is partially occluded.3.An interactive object detection and tracking system is designed and implemented based on the object detection and tracking algorithm based on the recurrent neural network proposed in this paper.It is verified that the object detection and tracking algorithm proposed in this paper can be applied to the actual complex scenes.The interactive system designed and implemented in this paper has certain practical value,so the research content in this paper has extensive research significance and application value.In conclusion,the object detection and tracking algorithm based on motion direction prediction and the object tracking algorithm based on adaptive detection proposed in this paper are tested with OTB and VOT datasets,and good evaluation results are obtained in both datasets.In addition,the interactive object detection and tracking system designed in this paper is tested,which proves that the interactive system designed and implemented in this paper has certain application value.