Development Of Dynamic Object Detection System Based On UAV Platform

Computer vision technology is now widely used in all areas of life,such as vision measurement,event monitoring,object detection,scene reconstruction,etc.The object detection technology,as one of the important research direction in the field of computer vision,it has important application prospect in military reconnaissance,wisdom cities,intelligent transportation,video monitoring and other fields.The commonly used object detection algorithms include optical flow method,frame difference method,background difference method and matching method.With the improving of the hardware performance in the recent years,and the new algorithm is proposed,the deep neural network method can autonomous learn the characteristics of objects.It shows strong robustness in the field of object detection,and even some object detection algorithm can be used on a server to detect static or dynamic object in real-time.This paper tries to develop a versatile hardware system on the embedded platform NVIDIA Jetson TX1,applied to UAV real-time vehicle detection,what's more,it can access to current UAV flight pose and location coordinates by inertial measure-ment sensor and GPS sensors.The main contents of the paper include:The first chapter elaborates the background and significance of this research,ana-lyzes the domestic and foreign research status of dynamic object detection,and gives the main research contents of this paper.The second chapter analyzes the demand of mobile terminal dynamic object de-tection system,gives the overall software and hardware scheme of the system,and com-pletes the selection of sensor,embedded platform and other hardware.Third chapter comparative analysis the optical flow method,the frame difference method,background difference method,the matching method,the deep convolution neural network used on object detection algorithm,and finally choose the deep convo-lution neural network algorithm to detect object in a variety of cases of dynamic scene.In the fourth chapter,we train the real-time object detection method based on the deep convolution neural network,improve the model and deploy it on the embedded system.First,make the vehicle data set VOCcar for training.Then improve the object detection model tiny-yolo.Finally,the deep learning model is deployed on NVIDIA Jetson TX1,and the dynamic vehicle object detection is realized on the mobile terminal.The detection frame rate is stable at about 20 frames.The fifth chapter realizes the integration of hardware system and object detection algorithm,and develops prototype system.Acquire UAV pose data information and geographic coordinate data information by using Arduino.And filter the raw data based on the kalman filter algorithm,in order to get more accurate pose information of UAV.The sixth chapter summarizes the main work and achievements of the paper,and looks forward to the future research and development of embedded artificial intelli-gence hardware system.