Aerial Vehicle Detection And Tracking Based On Convolutional Neural Network

With the advancement of technology,the continuous improvement of hardware,the emergence of large-scale data,deep learning has made great breakthroughs and developments in all aspects.This paper mainly studies the detection and tracking of aerial vehicles in the foggy background.Firstly,the development and current status of image dehazing,target detection and target tracking are introduced.At the same time,the convolutional neural network is summarized.The vehicles detected in the foggy background are mostly light and dark,and there is a problem that the restored image is dark and the image edge details are lost,which has an impact on subsequent target detection.In response to this problem,DehazeNet was used to obtain the transmittance maps of the four fog-related features;the atmospheric light value was obtained by local atmospheric light values instead of the single atmospheric light values;and the edge detection mean-guided filtering was proposed,and the initial transmittance map and initial The atmospheric light map is optimized to preserve edge detail information.The DehazeNet and edge detection mean-guided filtered image defogging module is designed to solve the problem of restoring image darkness and missing image edge detail information,and improving the subsequent detection of aerial vehicle target detection.Based on the selection of the SSD target detection algorithm based on convolutional neural network,according to the characteristics of aerial vehicles,only the vehicle needs to be detected.Therefore,an improved aerial vehicle detection module based on SSD is designed.Adjust the network structure,select a specific convolution layer as the feature layer,and modify the size of the anchor point box according to the a priori information of the aerial vehicle to improve the detection confidence and optimization model.When training the network,the data set obtained by data enhancement with the Stanford Cars data set and the data set collected by itself were used for network training.Finally,an experimental analysis is performed.According to the aerial camera,the video has little change and smoothness.Therefore,the GOTURN tracking algorithm based on convolutional neural network is selected to track the aerial vehicle,and the whole is built according to the image defogging module,the aerial vehicle detection module and the aerial vehicle tracking module.The system module implements an image defogging,detection and tracking system module for aerial vehicles.Finally,by analyzing the inadequacies of the experimental results,the direction of future improvement is proposed.