The invention discloses a multi-camera
system calibration method based on
pedestrian head recognition, and belongs to the technical field of
computer vision. The method comprises the following steps:
processing each frame of image, and extracting an
ellipse of a head portrait in the image by using a CNN method; calculating the three-dimensional position of the
human head in each frame under a camera coordinate
system according to the position and the size of the
ellipse; selecting any camera coordinate
system as a world coordinate system, and calculating external parameters of other cameras; and optimizing the obtained camera external parameters, and performing alignment conversion on the camera world coordinate system and the selected world coordinate system. According to the method, thehuman head is used as a feature point, the
point cloud formed by the motion trail of the
human head is used as a virtual calibration object, the method for calculating the three-dimensional coordinates of the head under the
monocular single-frame image is provided, and the multi-camera external parameter calibration problem is converted into the three-dimensional
point cloud alignment problem. Therefore, real-time online accurate external parameter calibration of the multi-camera system is completed by calculating the relative
pose between the three-dimensional point clouds.