Research On Target Tracking Based On UAV

Target tracking is a popular research direction in the field of computer vision.Target tracking is the process of detecting the pixel coordinates of the target center point in the image,and then transmitting the coordinates as a control amount to the smart device,so that the smart device can obtain the process of the target location in real time.As a representative of smart devices with portability and high maneuverability,UAVs have been widely researched and applied in the field of target tracking.Because most UAVs have limited processor performance,they must encode and compress the captured images and return them to the ground,and then process them at the ground station.On the one hand,the processing affects the efficiency of tracking;on the other hand,when the UAV and the ground station are transmitting data,they are susceptible to electromagnetic interference and become unstable.When using the UAV platform for target tracking,the algorithm needs to meet real-time,accuracy and robustness.The traditional computer vision-based tracking algorithm has low tracking accuracy due to its small number of training samples;the target tracking algorithm based on deep learning has been trained in advance with massive data,so the accuracy of the algorithm is very high,but Because the model it trains is usually relatively large,it cannot meet the requirements in the occasions where the real-time and computing resource requirements are strict.In order to improve the accuracy of the tracking algorithm on the premise of ensuring real-time performance,this article conducts research in the following aspects:?Research the target tracking algorithm based on Kernel Correlation Filter(KCF).Compared with the traditional computer vision tracking algorithm,the kernel correlation filter algorithm has the characteristics of strong real-time performance and high accuracy.To study the principle of the algorithm,according to the characteristics of target occlusion,target confusion and low frame rate when tracking target s in the UAV scene,select a data set that meets the characteristics of the scene on the OTB standard data set,and verify its Performance under different conditions;combined with simulation results,discuss the advantages and disadvantages of the kernel-based filter tracking algorithm;for its shortcomings that it is easy to lose the target in the case of target occlusion and low frame rate,the target template and filter update method are carried out Optimization;comparing the optimization algorithm with the original algorithm on the standard data set,it is concluded that the optimization algorithm has a slightly higher time complexity than the original algorithm,but the accuracy has been improved in the case of target occlusion and low frame rate in conclusion.? Research on target detection and tracking algorithms based on deep learning.Research the principles and advantages and disadvantages of commonly used target detection and tracking algorithms based on deep learning,and use the PASCAL VOC standard data set to simulate the accuracy of its detection;for two of the best detection algorithms,verify on the OTB standard data set Its accuracy and time complexity in short target tracking provide theoretical support for the UAV platform to achieve intelligent tracking.?Propose a target tracking solution based on UAV and build a target tracking platform based on UAV.This solution realizes the real-time detection and tracking of ground target s by the drone with a co-processor and a tracking algorithm based on deep learning.Build a hardware and software platform based on the solution,use the platform to run a tracking algorithm based on deep learning,and use the Tensor RT framework to accelerate the algorithm.According to the platform test results,it is concluded that the algorithm can detect and track target s on the platform in real time.