Abstract: In view of small moving target tracking problem on aerial video with existing of 3D scene occlusion, a small moving target detection and tracking method based on multi-view aerial video registration is proposed. Firstly the image sequence is sampled interval, and Harris detector is utilized to extract the feature points, the initial matching is achieved using Delaunay triangulation registration, then the differential images are calculated by integrated transformation model, and then it detects the target by calculating the accumulated energy. At last, the Kalman filtering is used to eliminate jitter of moving target tracking. Experimental results show that the algorithm can detect the minimum 30 pixels of slow moving targets on aerial video of urban and suburban scene.

Key words: aerial video, small moving target detection and tracking, multi-view images, transformation model

摘要： 针对存在3D场景遮挡的航拍视频运动小目标跟踪问题，提出一种基于多视角航拍配准的运动小目标检测和跟踪算法。该算法首先对图像序列间隔采样，利用Harris检测器提取全局特征点，通过Delaunay三角网对待配准图像实现初始匹配，然后利用整合变换模型计算差分图像，并利用累积能量检测出目标，最后通过卡尔曼运动滤波消除运动目标跟踪的抖动。实验结果表明，该算法对城市和郊区场景的航拍视频可以检测出最小30个像素的缓慢运动目标。

关键词: 航拍视频, 小目标检测与跟踪, 多视角图像, 变换模型