Research On Target Detection And Tracking Algorithm Based On Yolov3

In this paper,we focus on the study of target detection and recognition and visual tracking algorithms based on Yolov3.This paper mainly carries out the following work:1.This paper studies target detection.This paper proposes a single output layer Yolov3 improved target detection algorithm that introduces channel pruning.In order to reduce the amount of algorithm parameters and calculations,the algorithm uses channel pruning operations to streamline the network level without affecting accuracy.Under the premise of,a network with a smaller overall width can be obtained.At the same time,in order to prevent forced layered prediction,the algorithm changes the original multi-output structure of the Yolov3 algorithm to a single-output layer structure.2.This paper studies target tracking.In Chapter 3,this paper proposes a dynamic update network target tracking algorithm that introduces deep features.The core network layer of the algorithm uses the CIRes Net network as the core feature extraction network of the algorithm.3.In order to take into account the real-time and accuracy requirements of the algorithm,this paper designs a candidate network target tracking algorithm that introduces the color perception module in Chapter 4,integrates the color features into the twin network structure,and uses the ROI Align operation to process the two parts The obtained information is fused as output.Among them,the color perception module is mainly used to count the color features in the video frame,and a linear interpolation method is added to the color perception module to update the target model online.4.This paper designs a visualization system that turns the target detection and target tracking process into a user-oriented application operating system.