Research on real-time object tracking based on YOLO3开题报告

 2021-11-05 07:11

1. 研究目的与意义(文献综述包含参考文献)

Research background: The use of unmanned aerial vehicles (UAV) in the last few years is growing up rapidly in terms of use. There are still not commonly used in mass adoption, but UAVs have already broken through industry barriers which have been stayed impassable for years by similar technological innovations. Over the last decade, UAVs have become central to all manner of businesses, military appointments, governmental organizations and have managed to break through the field where certain industries were sluggish behind. From fast deliveries from point A to point B, scanning an unreachable place for human beings as well as real-time target tracking for vigilance purposes, UAVs are proving to be very useful and beneficial in places a man cannot reach or incapable to perform in a timely and efficient manner. Developing productivity and work efficiency, improving accuracy, decreasing workload and production costs, and solving law enforcement issues on an enormous scale is top uses UAVs in industries globally. Endorsements of the UAVs across several fields of industries ascends from the non-using stage to the mega-trend stage very rapidly as more people started to realize its potential in a scale of global coverage. Whether UAVs are controlled by remote control or accessed via a high technology computer, they manage the capability of reaching the most remote areas that require the least amount of time and energy. This is the reason why the UAVs are being used worldwide, especially by the military, personal and commercial sectors.With the availability of getting efficient, low-cost and commercially available unmanned aerial vehicles, target tracking using in UAVs has become important on daily use due to its many new applications, for example, traffic monitoring, on-duty surveillance, obstacle avoidance, rescue operations, etc. However, real-time target tracking presents many challenges, such as fast motion and image instability, changing the target and its background. To solve these problems, many algorithms springs up continuously. It is proposed to adaptively change the size of the receiving box scale and the relative position of the rectangular filter, as well as an algorithm for precise positioning. First of all, we propose an improved real-time correlation filter tracking algorithm, perception for existing correlation filter tracking algorithms in a complex context. Second, an improved real-time correlation filter tracking algorithm based on occlusion, detection is proposed to study the various levels of occlusion that may occur when tracking a target. Thirdly, improvements for embedded platforms to achieve stable real-time target tracking.Overview of real-time target trackingIntroduction: In the last few years, we have witnessed important development in the area of computer vision. A great deal of research has done in vision tasks, such as target tracking and recognition. As an important field of computer vision, visual object tracking plays an active role in many applications, in which target tracking using drones is a highlighted one. Since the UAV can track the target based on vision feedback and change its position and orientation to increase the tracking performance, target tracking is extensively applied to a diverse selection of objects, which can`t be physically tracked from the ground. Large numbers of new applications based on target tracking have been applied including person following, infrastructure inspection and obstacle avoidance. However, compared with other tracking systems, real-time target tracking requires analysis on a dynamic scene and handling new challenges defined on the UAV videos. The problem of target tracking in video is outlined as the task of nding the position of target in every frame of the video. The ability to track a target in a video depend on several factors, such as knowledge about the target, type of criterias being tracked and type of video showing the target. The vast field of application reects the importance of reliable and effective target tracking. There are several important steps towards effective target tracking, including the choice of model to represent the target, and target tracking method suitable for the task.Real-time target tracking with UAVs: Here are few ways to track an object in real-time by UAV:1. Mean Filter tracking:The proposed system is based on the Mean shift algorithm to perform object tracking. In order to improve the results, this standard tracking technique will be enhanced by previously filtering a back projected image, using the target predicted position given by a Kalman filter. Tracking system consist of these three modules: Image Registration, Motion Detection and Object Tracking.a. Image registrationThis first module is responsible to align the captured frames by incrementally making a mosaicing, where all the frames belong then to a unique world referential. This is a crucial operation to make possible to detect object motion with image subtraction. Image registration is accomplished by combining the following techniques: camera calibration, image homography, feature extraction, and optical flow estimation. A camera calibration method is used to correct geometric deformations present in the video frames. In this kind of projects, these deformations are usually strong, as the used lenses have a large field of view, which generate significant radial and tangential distortions. The align phase uses feature extraction to detect interest points and optical flow to follow those features in the next frame. The set of corresponding points is then used to estimate a homography matrix, which is finally used to align consecutive frames into the same geometric referential.b. Motion DetectionMotion detection uses two different techniques: image subtraction and color information. In the first one, the aligned frames are returned from the previous module and subtracted to obtain a difference image that can be used as a starting point to detect the moving objects. A second approach is the use of a color histogram to model the desired target, and subsequently compute, for each new frame, the probability of each pixel color was produced by the moving target. This new image, containing the referred probabilities, is usually designed as back projected image.c. Object trackingThe system tracks one of the objects returned by the motion detection module: that one selected by the user. Tracking is then accomplished by combining Mean shift and Kalman filter algorithms. Meanshift is used to estimate the object location based on the local maximums of the back projected image. The algorithm performs an iterative local optimization, searching in each iteration for the maximum of the pdf in a small square search window, centered at the previous estimation. Kalman Filter is an algorithm that produces estimates of state variables based on a series of measurements. In this model, it is assumed that prediction and measurement values are affected by white Gaussian noise. The tracking algorithm operates as following: first, the targets location is predicted using the Kalman filter prediction equation, then Meanshift is used to refine the predicted position, which is finally used as the observation in the of the Kalman filter and give a define picture back to control system.References: [1] Wu Y., Lim J., Yang M.H. Online object tracking: A benchmark; Proceedings of the 2013 IEEE Conference on Computer Vision and Pattern Recognition; [2] Ibrahim, Aryo Winam, et al. Moving Objects Detection and Tracking Framework for UAV-based Surveillance. Fourth Pacific-Rim Symposium on Image and Video Technology. 2010.[3] Duc Phu Chau, Francois Bremond, and Monique Thonnat. Object tracking in videos: Approaches and issues. 2013. [4] P. Theodora kopoulos and S. Lacroix, "A strategy for tracking a ground target with a UAV," 2008 IEEE/RSJ International Conference on Intelligent Robots and Systems, Nice, 2008, pp. 1254-1259. [5] Pengfei Fang, JianJiang Lu, Yu Long Tian, Zhuang Miao. An Improved object tracking method in UAV videos. (2011) 634 638p [6] Fu C., Duan R., Kircali D. Onboard Robust Visual Tracking for UAVs Using a Reliable Global-Local Object Model. Sensors. 2016. [7] Alper Yilmaz, Omar Javed, and Mubarak Shah. Object tracking: A survey. Act computing surveys (CSUR), 38(4):13. 2006. [8] U. Zengin and A. Dogan, "Real-Time Target Tracking for Autonomous UAVs in Adversarial Environments: A Gradient Search Algorithm," in IEEE Transactions on Robotics, vol. 23, no. 2, pp. 294-307, April 2007. [9] Z. Li, N. Hovakimyan, V. Dobrokhodov and I. Kaminer, "Vision-based target tracking and motion estimation using a small UAV," 49th IEEE Conference on Decision and Control (CDC), Atlanta, GA [10] David Hurych. Linear Predictors for Real-time Object Tracking and Detection, CTUCMP201404. 2014. [11] Samuel Murray. Real-Time Multiple Object Tracking, 2017. [12] Comaniciu D, Ramesh V. Kernel-based object tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence; 2003, vol. 25, no. 5, pp.564-577. [13] Pengfei Fang *, Jianjiang Lu, Yulong Tian, Zhuang Miao. An improved object tracking method in UAV videos. (2011) 634 638pp. [14] Lin, S.; Garratt, M.A.; Lambert, A.J. Monocular Vision-Based Real-Time Target Recognition and Tracking for Autonomously Landing an UAV in a Cluttered Shipboard Environment. Auton. Robots . 2017. [15] Fu, C.; Duan, R.; Kircali, D.; Kayacan, E. Onboard Robust Visual Tracking for UAVs Using a Reliable Global-Local Object Model. Sensors. 2016. [16] Caifeng Shan, Tieniu Tan, Yucheng Wei. Real-time tracking using a mean shift embedded particle filter [17] Sanna Agren. Object tracking methods and their areas of application: A meta-analysis. [18] Ratnesh Kumar. Video segmentation and multiple object tracking. Other [cs.OH]. Universit Nice Sophia Antipolis, 2014. English.

2. 研究的基本内容、问题解决措施及方案

1. Research task of this projectThe goal is designing a software and programming the Tello UAVs internal program with camera in it using the program Python, which can track the objects in real-time and can distinguish the surrounding objects. Detailed contents as below: Design and create the function code for the Tello UAV that will track the targets in real-time. Design related software that can control the movement of the Tello UAV, collect target information and avoid the obstacles. 2. Research methods of this projectDuring this graduation project, we`ll use methods of theoretical and experimental modeling. Mainly by studying a large amount of literature, testing already existing target tracking algorithms, exploring the new one. In the middle of the semester, we`ll do experiments with the new algorithms in experimental class to get deep knowledge about object tracking with Tello UAV. Today real-time high accuracy target tracking algorithms on UAVs are not enough, we intend to create the new tracking algorithm, that will optimize output and speed of the video in real-time as well as bring us high accuracy target tracking.

剩余内容已隐藏,您需要先支付 10元 才能查看该篇文章全部内容!立即支付

课题毕业论文、开题报告、任务书、外文翻译、程序设计、图纸设计等资料可联系客服协助查找。