运动目标检测是实现目标跟踪、交通监控、行为分析等任务的基础。但在无人机获取的视频图像中, 无人机运动、旋翼震动或外界风力等客观因素使图像出现较为明显的背景、光照等变化, 会对运动目标的检测产生影响。因此, 如何降低干扰、提高检测精度, 让无人机在运动目标检测领域发挥作用在信息时代具有相当重要的意义。无人机视频图像的运动目标检测相比传统运动目标检测, 检测思路基本一致, 但干扰因素众多。本文以此为切入点, 分类综述了适用于无人机视频图像运动目标检测的算法及其改进, 主要包括运动估计算法、帧间差法、背景建模法、光流法等传统算法和近年出现的新型算法; 通过对无人机运动状态的划分探讨比较了上述方法的优缺点及适用场景。帧间差法更适合处理无人机悬停状态的数据, 背景建模法、光流法及新型算法对无人机悬停及巡航状态的数据均可处理; 上述算法均不能很好解决光照变化造成误检、漏检现象。所以处理无人机视频数据时, 要根据其运动信息及数据特点选择合适的算法, 才能获得好的检测结果。

Moving target detection is the basis for tasks such as target tracking, traffic monitoring and behavior analysis. However, in the video images captured by UAV, objective factors such as drone movement, propeller rotation or wind will affect the detection of moving targets, these uncertainties may cause failure during the detection. It is very important to reduce the interference, improve the detection accuracy, and make the UAV play an important role in the field of motion detection in the information age. Compared with the traditional moving target detection, the detection method of UAV video image is basically consistent with many interference factors. In this paper, the algorithm and its improvement for UAV video image moving target detection are summarized, including traditional algorithms such as motion estimation algorithm, frame difference method, background modeling method, optical flow method, and new algorithms appearing in recent years. The advantages, disadvantages and application scenarios of the above methods have been compared through the division of UAV movement status. The frame difference method is more suitable for the data of UAV hover state, the background modeling method, the optical flow method and the new algorithm can be used to deal with the UAV hover and cruise state data. None of them can solve the problem of false detection and missed inspection caused by illumination change. For processing UAV video data, it is necessary to select an appropriate algorithm according to its motion information and data characteristics to obtain good detection results.