半导体光电, 2019, 40 (1): 93, 网络出版: 2019-03-25
基于改进FAsT-Match算法的特定建筑区锁定跟踪
Lock Tracking for Specific Building Areas Based on Improved FAst-Match Algorithm
机载遥感视频 FAsT-Match算法 直方图规定划 仿射变换参数 锁定跟踪 airborne remote sensing video FAsT-Match algorithm histogram definition affine transformation parameter lock tracking
摘要
机载遥感视频背景复杂, 且指定建筑目标面积小、分布离散, 传统区域提取算法难以准确锁定并跟踪这类目标。提出了一种基于改进FAsT-Match算法的特定建筑区锁定跟踪方法。该方法首先以模板图为基准对目标图像进行直方图规定划, 以适应不同的光照变化; 然后构建仿射变换参数网络, 并根据上一帧得到的最佳仿射变换参数限制当前帧图像的仿射变换参数范围, 以提升匹配效率; 最后将与仿射变换匹配的平行四边形图像数据经过逆仿射变换成矩形图像作为下一帧模板, 从而解决旋转、尺度、形变等变化对目标跟踪准确性的影响。由实验分析可知, 该算法AUC指标可达0.820, 较NCC算法准确率提升40.5%, 且跟踪效果好、效率高、对各种场景的适应性好, 可在特定建筑区域准确、实时、高效地锁定跟踪。
Abstract
The airborne remote sensing video has a complex background, and the specified building has a small target area and discrete distribution. It is difficult for the traditional region extraction algorithm to accurately lock and track these targets. In view of this, a lock-tracking method for specific building area based on improved FAsT-Match algorithm was proposed. First, the template map was used as a reference to perform histogram definition for the target image to adapt to different illumination changes. Then, the affine transformation parameter network was constructed, and the affine transformation parameters of the current frame were limited according to the best affine transformation parameters obtained in the previous frame to improve the matching efficiency. Ultimately, the parallelogram image data matched by affine transformation is inverse affine transformed into a rectangular image as the next frame template, so as to solve the impact of rotation, scale, deformation and other changes on the target tracking accuracy. Experimental results show that the AUC index of this algorithm can reach 0.820, and the accuracy is 40.5% higher than that of the NCC algorithm. It can realize good tracking effect, high efficiency, good robustness to various scenes and environments, and can achieve accurate, real-time, specific, and efficient lock tracking for specific building areas.
尚飞, 秦艳艳, 杨志华. 基于改进FAsT-Match算法的特定建筑区锁定跟踪[J]. 半导体光电, 2019, 40(1): 93. SHANG Fei, QIN Yanyan, YANG Zhihua. Lock Tracking for Specific Building Areas Based on Improved FAst-Match Algorithm[J]. Semiconductor Optoelectronics, 2019, 40(1): 93.