应用于立体车库的人体检测算法

Research on Human Detection for Stereo Garage

摘要:

针对立体车库的安全监控需求,提出了一种基于机器视觉的人体检测算法,用于在立体车库内检测非法侵入的人员。首先,采用自适应双线性滤波方法滤除视频帧中的噪声。接着,差分视频帧与背景图像,并采用多尺度小波分析法二值化差分图像。然后,将二值图像进行形态学滤波以检测出人体。最后,利用高斯混合模型(Gaussian Mixed Model, GMM)更新背景图像。实验结果表明,提出算法可以在立体车库内实时准确地检测出人体,为立体车库的安全监控提供了可靠的保障。

Abstract:

For the safety monitoring requirements of stereo garage, a machine vision based human detection algorithm is proposed to detect unlawfully intruded persons in the stereo garage. First, an adaptive bilinear filtering method is used to remove the noise of the video frame. Next, a difference image of the video frame and the background image is calculated, and the multi-scale wavelet analysis algorithm is employed to obtain binary image of the difference image. Then, human is detected from the binary image by morphology processing. Final, the Gaussian Mixed Model (GMM) algorithm is adopted to update the background image. Experimental results show that the proposed algorithm can detect human in real time and accurately in the stereo garage. It provides a reliable guarantee for the improved safety monitoring of stereo garage.