A Motion Vehicle Detection And Tracking System Based On A Low-altitude Drone Platform

In recent years,low-altitude unmanned aerial vehicles developed rapidly.With the technology of hover,recharge mileage,image PTZ mature,a large number of UAV-based industry applications were born.At the same time with the development of computer vision technology and machine learning technology,the ability to detect and track the target is getting stronger.The combination of the two technology used in the city under the road vehicle detection and tracking have a good application scenarios.On the other hand,the study fits the current hotspots of intelligent traffic with the city,it is the trend of future.The difficulty of the study is both collect images in the UAV platform and handle specific motion target detection and tracking in mobile context.At the same time it need to overcome the weak computing power of the airborne computing equipment.There have high requirement to speed and efficiency.This is a very challenging subject.This article describes how to build a low-altitude UAV image capturing platform and airborne computing platform.Through the system to collect urban road video,then analyze the video,using the method of computer vision and machine learning to research the way of detect and track vehicle targets,the effectiveness of the proposed algorithm is verified by actual flight.It can achieve low-cost and efficient intelligent traffic monitoring system in the application level.Innovation:(1)This paper deployed the entire system that include image capturing,processing,calculation on the UAV platform.It is very suitable for the target that the moving vehicle in the city road scene.It cheaper,faster and more efficient than traditional camera.(2)In terms of detection algorithm,according to the characteristics of the scene to optimize the traditional algorithm,proposed a global background compensation method based on feature point matching and a machine learning method based on HOG and LBP Classifier.The algorithms are applied to the image video sequence of the moving vehicle that collected by low altitude unmanned aerial vehicle platform.Then compared and analyzed the advantages and disadvantages of several methods.(3)In terms of tracking algorithm,proposed a multi-feature fusion tracking strategy based on Kalman and Camshift algorithm and validated its validity by the platform.Details:(1)This paper described the platform design detailedly,analysis the hardware requirement of carrying,collecting,computing which low altitude unmanned aerial vehicle need,compared the hardware and platform performance differences and characteristics,introduces the hardware and software of the system and the parameters of the calculation platform.In terms of hardware solution,the flying platform uses the DJI Matrice100 developer kit,the image capturing equipment uses DJI X3 PTZ camera that support 320 P to 1080 P video capture,its three-axis mechanical PTZ with Matrice100 can effectively offset the high frequency mechanical vibration from platform flight,and provide convenience to capture image,compared performance between Intel Nuc computer and DJI Manifold airborne computer.In terms of software solution,video capturing and algorithm implementation uses CUDA that support GPU computing and OpenCV image processing library.(2)For the platform is movably and real-time changes in the scene,first using traditional detection algorithm like background extraction method,interface difference method,optical flow method to experiment,found that the results are not ideal,a vehicle target detection using motion-based global background compensation algorithm based on feature point matching is proposed.The principle is extract the feature points between the reference frame and the matching frame and find transformation relation between two frames by matching feature points,the transformation relation is used as the global background compensation parameter.Transform the mobile background target detection problem to static background target detection problem for reducing the impact of platform movement.(3)Through the analysis and experiment,compared the tracking effect of Meanshift,Kalman filter method and basic particle filter method,analyzed defects of this kind of application scene,proposed a multi-feature fusion strategy based on Kalman and Camshift tracking algorithm,tracking effect is better and able to deal with the static vehicle target,largely enhance the tracking speed.Experiments show that the system and algorithm can effectively detect the moving vehicle target in different scenes of urban road,and carry out effective and continuous tracking.