为了实现在复杂背景条件下对动态视频红外多目标进行稳健的检测与跟踪, 提出了一种新颖鲁棒的运动多目标的检测与跟踪算法。首先对相邻红外图像进行匹配校准, 利用图像累积差异图检测出运动的目标。为了达到实时高效的检测效果, 提出了网格采样策略, 大大降低了特征点的匹配复杂度并解决了特征点非均匀集中的问题。同时采用强度滤波和形态学操作等算法提升了目标的显著性特性, 滤除了虚假目标; 由于红外热像仪视场的变化, 目标的尺度将发生变化, 在检测到目标的基础上提出了尺度计算与区域检测算法; 最后采取了传统的卡尔曼滤波对检测到的目标进行跟踪。实验结果表明, 本文算法能够准确地检测动态场景下运动目标, 并在目标尺度变化时自适应的检测出目标的变化, 同时稳定地跟踪目标。

In order to achieve robust multi-object detection and tracking in dynamic infrared video under complex background conditions, a novel and efficient detection-and-tracking algorithm for moving targets is proposed in this paper. The algorithm firstly matches the adjacent infrared frames, and the image cumulative difference map is adopted to detect the moving target. In order to achieve real-time and efficient detection, a grid-sampling strategy is proposed, which greatly reduces the matching complexity of feature points and solves the problem of non-uniformity for feature points. At the same time, the intensity filtering and morphological operation are used to enhance the saliency of the object and also filter out the false object. Since the field of view of the infrared imager constantly changes, the object scale will also change. Therefore, a scale-and-region detection algorithm is proposed for solving the scale change. Finally, the traditional Kalman filter is adopted to track the detected objects. The experimental results show that the algorithm proposed in this paper can accurately detect multiple moving objects in a dynamic scene, and adaptively detect the change of target when the object scale changes, to realize robust object-tracking.