为实现精确的红外-可见光视频序列的自动配准, 提出了一种新的基于目标轨迹线匹配的配准方法。首先, 利用运动目标检测技术提取目标前景, 并由基于相关滤波器(KCF)的多目标跟踪算法对每个前景顶点进行跟踪, 进而获取每个目标的运动轨迹。此后, 为每条轨迹线建立归一化运动方向描述子与归一化运动幅度描述子, 通过时序分析、方向描述子匹配及幅值描述子匹配建立分步约束的匹配机制, 完成轨迹线匹配工作。最后, 采用迭代更新的方式获取最佳全局配准矩阵, 实现对异源视频的配准。在LITIV数据库上的9组视频上进行测试验证, 实验的结果表明: 本文配准算法的重叠率误差一般小于0.2, 接近或已超过手动的Ground-Truth矩阵。通过充分利用目标的运动信息, 该算法实现了精确的红外-可见光图像序列配准。

In order to realize accurate and automatic infrared-visible video registration, a novel registration method was proposed based on matching the motion trajectories of targets. First, the top pixel of each foreground was tracked by using the multi-target tracking algorithm based on KCF. In this way, the trajectory of each target was obtained. Then, the normalized motion orientation descriptor and the normalized motion magnitude descriptor were established for each trajectory. The stepwise constraint matching framework was structured by using time analysis, orientation descriptor matching and magnitude descriptor matching. Finally, the best registration matrix was obtained with iterative updating. The method proposed was validated with the nine pairs of videos in the LITIV database. The results indicate that the overlap error of the proposed method is smaller than 0.2, which is close to or better than the manual Ground-Truth matrix. By adequately using the motion information of target, the algorithm can realize precise infrared-visible image sequence registration.