红外视频图像中的人体目标检测方法

Human detection method in infrared video images

人体目标检测是很多机器视觉应用的难点,如智能视频监控和车辆辅助驾驶,基于可见光图像的方法很难解决复杂背景和目标区域的分离问题,因此,越来越多的研究转向利用红外图像进行检测。提出了一种红外视频图像中的人体目标检测方法,该方法首先利用红外图像中像素值近似呈现单模态分布的特点,对高亮像素进行检测,然后采用灰度直方图和投影直方图相结合的2D直方图特征对目标进行检测。实验结果表明:该方法具有较高的检测率,但误报率也较高,其原因在于负样本的数量和代表性不足,因此,改进空间很大。

Human detection in images is a challenging problem in the applications of machine vision, such as intelligent video surveillance and vehicle assistance driving. Because visible-image-based methods have difficulty in extracting moving objects from complex backgrounds, more research interests on human detection have been addressed in infrared video images. In this paper, a new method for detecting human in infrared video images was proposed. Due to the pixel values in infrared images could be approximately modeled by the univariate Gaussian distribution firstly, the univariate Gaussian model was used to detect the pixels with large intensity in images, then, the 2D histogram feature vector was utilized to detect objects, which combined a gray histogram feature and a projection histogram feature. Experimental results show the detection rate （DR） of the presented method is high, but the fault detection rate （FDR） is also slightly high for lacking of quantity and representativeness of negative-sample.Therefore,more research should be done to decrease the FDR.