The invention discloses a human body detection method and device and solves a problem of low accuracy of human body detection existing in the prior art when two or more persons parallelly stand back and forth in an overlapped mode. The method comprises steps that a to-be-detected first point cloud data cluster is divided into multiple first sub point cloud data clusters according to preset rules; each local highest point of the first point cloud data cluster is determined, if a projection distance of two local highest points on a horizontal plane is smaller than a set threshold, the first sub point cloud data clusters in areas corresponding to the first sub point cloud data clusters of the two local highest points are merged to be a second point cloud data cluster, if not, a second point cloud data cluster containing the local highest points is determined; according to an HOG value of an image corresponding to the second point cloud data cluster, whether the image of the second point cloud data cluster is a human body image is determined. The method is advantaged in that two-person or more-person back and forth parallel overlapped standing conditions can be better distinguished in the human body determination mode, and human body detection accuracy is improved.