光电跟踪系统中的运动目标跟踪方法的研究

发布时间：2019-04-25 08:33

【摘要】：光电跟踪系统以可见光相机或热像仪为传感器,探测环境中的运动目标并进行跟踪监视。由于其以光波被动的方式对目标进行探测,不发射电磁波,所以抗干扰能力强,在军事和国防安全领域已经得到了广泛的应用。随着社会需求的多样化和光电技术的发展,基于智能图像跟踪算法的光电跟踪系统如今已经应用在民用领域、天文观测、航空摄影、靶场测试和激光通讯等领域。光电跟踪系统目标跟踪策略决定着系统跟踪性能的好坏,尤其是图像跟踪算法需要具备实时性好、抗遮挡性能优越的特点。而传统的光电目标跟踪策略难以满足抗遮挡性要求。本文提出了一种满足系统抗遮挡性要求的目标跟踪策略。首先针对系统现有结构,分析了光电跟踪原理,通过建立二轴户外指向器跟踪坐标系,及分析电视测量原理,并结合镜头视场角、焦距,将像素平面脱靶量转化为空间角度脱靶量。其次,为了满足系统的抗遮挡性及实时性需求,提出了改进检测器和模型更新策略的TLD跟踪算法,并进行了实验验证。论文主要内容如下:1.分析了光电跟踪原理,并实现了图像闭环PID算法、陀螺稳定PID算法。2.调试了光电跟踪伺服驱动系统,匹配了模拟驱动器与直流力矩电机参数。3.建立了二轴户外指向器的跟踪测量坐标系,通过分析电视测量原理,结合镜头视场角、焦距将图像像素脱靶量转化为空间角度脱靶量。4.为满足光电跟踪系统对跟踪算法的实时性和抗遮挡性需求,结合应用背景,通过实验分析TLD跟踪算法的时间复杂度及最优参数,提出了基于Otsu阈值的动态差分图像阈值分割算法,并对TLD检测器和目标模型更新策略进行了改进。5.对改进TLD跟踪算法进行了实验验证。以民航飞机飞行视频序列作为数据集,对TLD算法与改进TLD算法进行了实时性数值对比实验,并使用Recall等评价指标对TLD算法、改进TLD算法以及经典Mean-Shift算法进行了算法性能评价及分析。
[Abstract]:The photoelectric tracking system uses visible light camera or thermal imager as the sensor to detect the moving objects in the environment and track and monitor them. It has been widely used in the field of military and national defense security due to its passive detection of targets and non-emission of electromagnetic waves, so it has strong anti-jamming ability and has been widely used in the field of military and national defense security. With the diversification of social demand and the development of photoelectric technology, the photoelectric tracking system based on intelligent image tracking algorithm has been applied in civil field, astronomical observation, aerial photography, range test and laser communication and so on. The target tracking strategy of the photoelectric tracking system determines the performance of the tracking system, especially the image tracking algorithm needs to have the characteristics of good real-time performance and good anti-occlusion performance. However, the traditional photoelectric target tracking strategy is difficult to meet the requirements of anti-masking. In this paper, a target tracking strategy is proposed to meet the requirements of system anti-masking. Firstly, according to the existing structure of the system, the principle of photoelectric tracking is analyzed. The tracking coordinate system of the biaxial outdoor director is established, and the principle of television measurement is analyzed, and the lens field angle and focal length are combined with each other. The pixel plane miss distance is transformed into the spatial angle miss distance. Secondly, in order to meet the requirements of anti-masking and real-time performance of the system, an improved TLD tracking algorithm with improved detector and model update strategy is proposed and verified by experiments. The main contents of this paper are as follows: 1. The principle of photoelectric tracking is analyzed, and the image closed loop PID algorithm and gyro stabilized PID algorithm are realized. 2. The photoelectric tracking servo drive system has been debugged to match the parameters of analog driver and DC torque motor. 3. The tracking and measuring coordinate system of biaxial outdoor director is established. By analyzing the principle of television measurement, combining the lens field of view angle and focal length, the image pixel miss distance is transformed into the spatial angle miss distance. 4. In order to meet the real-time and anti-blocking requirements of the photoelectric tracking system, combined with the application background, the time complexity and the optimal parameters of the TLD tracking algorithm are analyzed by experiments. A threshold segmentation algorithm for dynamic differential image based on Otsu threshold is proposed, and the TLD detector and target model updating strategy are improved. 5. The improved TLD tracking algorithm is verified by experiments. Taking the flight video sequence of civil aviation aircraft as data set, the real-time numerical comparison experiment between TLD algorithm and improved TLD algorithm is carried out, and the TLD algorithm is evaluated by using Recall and other evaluation indexes. The performance of the improved TLD algorithm and the classical Mean-Shift algorithm is evaluated and analyzed.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】