Human body positioning method based on multi-perspectives and lighting system

A human body positioning method based on multi-perspectives, comprising steps of:

• S1) simultaneously capturing pictures of a site from at least two angles by a plurality of camera devices;
• S2) extracting human body feature points of each human body in each picture;
• S3) matching the human body feature points of a same human body of a same position in different pictures to form matched point groups, and according to position, pose and internal parameters of each camera device, calculating 3D coordinates of corresponding 3D point of each matched point group in a 3D space;
• S4) calculating a spatial geometric error between the 3D point and each human body feature point of the corresponding matched point group in each picture, and weighting the spatial geometric error to obtain a reprojection error of the corresponding 3D point, and subsequent to projecting the corresponding 3D point onto a ground to obtain single-point 2D coordinates thereof on the ground, weighting the projected single-point 2D coordinates according to the reprojection error of the corresponding 3D point in the corresponding matched point group to obtain a projection distribution of the corresponding human body on the ground;
• and
• S5) projecting light beams, by light fixtures, according to positional information of the respective light fixtures and the projection distribution of the corresponding human body on the ground obtained in the step S4),
• wherein in the step S2), the human body in each picture is detected by a human body detection algorithm and framed by a body frame, and the human body feature points are extracted in the body frame,
• wherein in the step S3), there are a plurality of human bodies on the site, degree of match among region blocks defined by the body frames in different pictures is calculated, and the body frames with the best degree of match in different pictures are selected to refer to the same human body, and
• wherein image similarity among the region blocks is used to assess the degree of match among the body frames.

A human body positioning method based on multi-perspectives, comprising steps of:

• S1) simultaneously capturing pictures of a site from at least two angles by a plurality of camera devices;
• S2) extracting human body feature points of each human body in each picture;
• S3) matching the human body feature points of a same human body of a same position in different pictures to form matched point groups, and according to position, pose and internal parameters of each camera device, calculating 3D coordinates of corresponding 3D point of each matched point group in a 3D space;
• S4) calculating a spatial geometric error between the 3D point and each human body feature point of the corresponding matched point group in each picture, and weighting the spatial geometric error to obtain a reprojection error of the corresponding 3D point, and subsequent to projecting the corresponding 3D point onto a ground to obtain single-point 2D coordinates thereof on the ground, weighting the projected single-point 2D coordinates according to the reprojection error of the corresponding 3D point in the corresponding matched point group to obtain a projection distribution of the corresponding human body on the ground;
• and
• S5) projecting light beams, by light fixtures, according to positional information of the respective light fixtures and the projection distribution of the corresponding human body on the ground obtained in the step S4),
• wherein in the step S2), the human body in each picture is detected by a human body detection algorithm and framed by a body frame, and the human body feature points are extracted in the body frame,
• wherein in the step S3), there are a plurality of human bodies on the site, degree of match among region blocks defined by the body frames in different pictures is calculated, and the body frames with the best degree of match in different pictures are selected to refer to the same human body, and
• wherein subsequent to taking out one body frame from each picture, forced matching on the human body feature points in the body frame taken out from each picture is performed to form matched point groups, the 3D coordinates of the corresponding 3D point of each matched point group in a 3D space and the spatial geometric errors between the corresponding 3D point and the human body feature points in each picture are calculated, the spatial geometric errors of all human body feature points in corresponding matched point group are weighted to obtain the reprojection error of the corresponding 3D point of the corresponding matched point group, and the reprojection errors of the 3D points corresponding to all the human body feature points in the body frames taken out is weighted to obtain a weighted reprojection error, negative value of the weighted reprojection error taken as the degree of match among the body frames.