摘要
目前针对背景复杂的海上舰船作为跟踪对象的跟踪算法研究较少。另外,多目标跟踪算法在变化频繁的背景条件下的精确性和实时性不足。为了提高多目标舰船跟踪的视觉跟踪算法的实时性和鲁棒性,本文提出了一种改进的基于TLD(Tracking-Learning-Detection)多目标舰船自动跟踪算法。该算法实现了海上多个目标舰船准确实时跟踪。首先,算法应用海天线检测与提取方法,利用最大类间方差阈值分割和Hough变换提取海天线;其次,算法利用Kalman滤波原理对海天线上的目标舰船进行定位与检测,分离出目标舰船;最后通过提取出来的目标用其最小外接矩形生成目标初始跟踪框,跟踪器利用初始跟踪框的位置坐标信息对目标进行实时自动跟踪。算法运行过程中都在海天线附近进行扫描检测,缩小了图像遍历范围,提高了算法的实时性。实验表明,通过对比典型的Mean-shift算法以及原始TLD目标跟踪算法,本文算法跟踪结果的精度较高,实时性较好。
At thepresent stage,ships in the complex sea conditionsare taken by scholar on as the tracking object in few research. In addition, the accuracy and real - time of the multi - target tracking algorithm un- der the background of frequent changes are insufficient. To improve the real-time and robustness of the multi-target visual tracking algorithm in tracking ship. An improved automatic tracking algorithm for multi-target ships based on TLD（Traeking-Learning-Detection） is proposed in this paper. The goal of auto- matically real-time tracking ships at sea have been achieved in the algorithm. Firstly, the detection and ex- traction method is proposed to split sea-sky-line. By using Otsu threshold segmentation and Hough trans- form, the sea-sky-line is easily extracted; Secondly, through using the principle of Kalman filter, the target ships near the sea-sky-line has been positioned and detected in the algorithm, and segregate the target ships. Finally, the initial target tracking frame has been generated with its minimum bounding rectangle by the extracted target, the tracker uses the position coordinate information of the initial tracking frame to automatically track the target in real time. During the running of the algorithm, scanning detection is car- ried out near the sea-sky-line, which reduces the traversing range and improves the real-time performance of the algorithm. Experimental analysis shows, bycomparing with the traditional Mean-shift algorithm andthe previous target tracking algorithm called TLD, the proposed algorithm is more accurate and more efficient in ship tracking.
出处
《中国海洋大学学报（自然科学版）》
CAS
CSCD
北大核心
2017年第2期128-136,共9页
Periodical of Ocean University of China
基金
海洋公益性行业科研专项经费(201505002)
中央高校基本科研业务费重点项目(2015ZZ028)
自主系统与网络控制教育部重点实验室资助~~
