设计并实现了一种适合于工程中动态目标识别应用的实时的复杂巡航场景运动检测技术。首先，对背景和当前帧两幅图像进行卷积处理，提取边缘图像。然后分块计算背景边缘图像的局部熵，根据局部熵大小合理选取配准模型组。采用模型组与当前帧边缘图像的相关性强度及运动偏移量结果，完成场景的运动估计。实验结果表明，本文方法用于红外和可见光图像均十分有效，在分辨率为320×256，帧频50 Hz的红外图像以及分辨率为768×576，帧频25 Hz的可见光图像中运行时间分别为13、24 ms，达到了工程中应用的实时性处理要求。同时，在后续的图像处理中场景运动检测有效剔除了复杂背景图像对目标识别的干扰，可准确检测出运动变化目标。

This paper designs and realizes a realtime motion detection technology for complex cruise scene. The technology is suitable for dynamic target recognition in engineering application. First of all, two images of the background and the current frame are processed by convolution operation and extract the edge images. Then we put parts of background edge image and compute their local entropy. According to the value of the local entropies, we can select reasonable registration model group and compute correlation strength and the value of movement offset. The value is the offset of the model group and the current edge image. We complete scene motion estimation by them. The experimental results show that the method can be used in the infrared and visible images. The infrared image processing time is 13 ms. Its resolution is 320×256 and frame rate is 50 Hz. The optical image processing time is 24 ms. Its resolution is 768×576 and frame rate is 25 Hz. They reach realtime processing requirement of the engineering application. Meanwhile, motion detection of scene effectively excludes the interference of target recognition which is produced by the complex background image. We can accurately detect the movement and changing target.