摘要： 为克服单个可见光摄像头检测准确率低的问题，提出一种融合双通道视频的人员检测系统。由可见光摄像头和红外热像仪分别获取同一场景的可见光和红外线视频数据，使用自适应学习速率的神经网络背景模型在2个通道中分别检测运动区域。通过图像配准对2个通道的结果进行“或”融合，并采用高斯滤波以消除噪声，利用积分图像快速检测近似长方形响应的人体区域。实验结果表明，该系统对行人和骑自行车人员的检测准确率达到98%，比单一通道具有更高的可靠性。

关键词: 视频监控, 运动检测, 背景模型, 双通道视频融合, 神经网络, 积分图像

Abstract: To overcome the low detection precision of single visible camera, a human detection system by fusion of bi-channel video is proposed. Visible and infrared video are obtained by a visible camera and a thermal infrared video of the same scene. Motion regions are detected separately in two videos by neural network background model with adaptive learning rate. Detected results of two channels are fused by image registration and logical “or” operation and noise are removed by Gauss filter. Human like rectangular regions are detected efficiently by using integral image. Experimental results show that, bi-channel video fusion can detect pedestrian and bicycle with presion of 98%, which is more reliable than a single channel video.

Key words: video surveillance, motion detection, background model, bi-channel video fusion, neural network, integral image

中图分类号: 

• TP391.41