机场航站楼热成像视频下融入运动信息的显著人体检测方法

Salient Human Body Detection Method Incorporating Motion Information Under Thermal Imaging Video of Airport Terminal

弱光背景下的目标检测是航站楼夜间巡检机器人的主要任务之一。为充分提取与利用动态场景视频数据中的运动信息,提高检测方法在人体被遮挡、图像边缘的人体目标不完整等实际复杂情况下的检测效果,避免因单帧检测结果不完全引起的歧义,提出一种机场航站楼热成像视频下融入运动信息的显著人体检测方法。该方法首先通过背景模型初步分离图像数据的前景和背景,借由特征点轨迹聚类和运动估计仿射模型进一步分离因摄像机运动被误判的背景区域,最后将运动目标检测结果作为运动信息与单帧检测算法结果相融合。在4个数据集中的实验表明,该方法能够有效地从动态场景视频数据中提取运动信息,且能够有效提升单帧检测精度,并避免检测不完全情况的发生。

Target detection in low light background is one of the main tasks of night patrol robot for airport terminal. In order to sufficiently extract and make use of the motion information in the dynamic scene video data,improve the detection effect of the detection method in the face of the actual complex situations such as the human body is blocked and incomplete human target at the edge of the image,and avoid the ambiguity caused by the incomplete detection results of a single frame,a salient human body detection method incorporating motion information under thermal imaging video of the airport terminal is proposed. First,the foreground and background of image data are separated preliminarily by a background separation model. Then,the background area misjudged by camera motion is further separated by a feature point track clustering and motion estimation affine model. Finally,the motion target detection result is fused with the results of single frame detection algorithm as motion information.It can effectively extract motion information from dynamic scene video data. Experiments on four datasets show that this method can effectively improve the accuracy of single-frame detection and avoid the occurrence of incomplete detection.