| CPC A63B 24/0062 (2013.01) [A63B 71/0622 (2013.01); A63B 2024/0034 (2013.01); A63B 2102/18 (2015.10); A63B 2102/24 (2015.10); A63B 2220/05 (2013.01); A63B 2220/806 (2013.01); A63B 2230/06 (2013.01); A63B 2243/0025 (2013.01); A63B 2243/0037 (2013.01); A63B 2243/0066 (2013.01); A63B 2243/0095 (2013.01)] | 5 Claims |
|
1. A device for efficiently extracting adaptively selected contactless multi-player heart rates applied to volleyball, basketball, soccer, rugby, baseball, and ice hockey on a court comprises:
cameras that are deployed in a plurality of angles to obtain videos from all angles of the field and cover the court;
a processor and a memory integrated with a microcontroller,
a bus to transmit data signals between the memory and the processor,
wherein the processor is configured to:
obtain videos to cover a court from a plurality of angles during players training by the cameras;
remove players with low contributions to training and games in the videos, wherein a movement trajectory of a ball in the videos is marked to obtain a ball position coordinate set (X, Y)Ball and then, a trajectory position of the players appearing in the videos is marked to obtain a player position coordinate set (X, Y)Obj; values of average change angles
 of the movement trajectory of the ball and the position coordinate set of the players are calculated respectively; if the average change angle is |θBall−θObj|<φ, wherein φ value is adjusted according to needs, and default φ value is 0.5° according to an actual measurement of volleyball training, change directions of the movement trajectory of the ball and the position coordinate set of the players are considered to be consistent approximately; and further, average values of position change
 of the movement trajectory of the ball and the trajectory position of the players are calculated, respectively; if the average position change rate is |kBall−kObj|<τ, wherein τ value is adjusted according to the needs, and default τ is 0.5° according to the actual measurement of volleyball training, the position change rates of the movement trajectory of the ball and the trajectory position of the observed players are considered to be consistent approximately, and when the movement trajectory of the ball is consistent with the movement direction and movement change rate of the players, the players are considered to be related to the movement trajectory of the ball, and the players are retained in the videos as valid players, otherwise the players who are not the valid players are removed from the videos;
detect whether the facial ROIs of the valid players contain perfect eye region features, and use the facial ROIs containing the eye region features as best facial ROIs for heart rate extraction; and
analyze the detected best facial ROIs by using blind source separation, and estimate a RGB signal by a JADE algorithm to obtain the heart rate values of the valid players, verifying whether each ROI contains the perfect eye region features, wherein the binocular eyeballs with closed contour are contained, if not, considering that the videos in this angle is invalid, and continuing to select the videos from other angles until all ROIs are retrieved to see whether they contain the perfect eye region features; and calculating the average value of the sum of roundness of a plurality of eyeballs in each ROI and selecting an angle video corresponding to a maximum average value as a best angle video for the heart rate extraction, when detecting a plurality of video angles that contain the perfect eye features;
wherein after obtaining the videos of the valid players in the best facial ROIs that contain the perfect eye region features and the maximum average value of roundness of the eyeballs, a blind source separation (BSS) method is used to extract the heart rates, and the JADE algorithm is used to estimate the heart rates from the RGB images of the best facial ROIs, thus realizing efficiently analyzing and calculating the videos and extracting heart rates with high accuracy;
wherein the processor is further configured to:
make a first judgement to judge whether a movement trajectory of a ball is consistent with a movement direction of an observed player; and
make a second judgement to judge whether the movement trajectory of the ball is consistent with a movement change rate of the observed player;
if the two judgments are consistent, the observed player is regarded as the valid players, otherwise the observed player is removed from the videos;
wherein the processor is further configured to:
have a first mark to mark the movement trajectory of the ball in a video clip to obtain a ball position coordinate set;
have a second mark to mark the trajectory position of the observed player appearing in the video clip to obtain the player position coordinate set; and
a first calculation to calculate values of the average change angles of the ball position coordinate set and the player position coordinate set respectively by using the obtained the ball position coordinate set and the player position coordinate set; if the average change angles of the ball position coordinate set and the player position coordinate set are less than a first threshold, the change directions of the two are considered to be consistent
wherein the processor is further configured to comprises:
have a second calculation to calculate average values of position change rates of the movement trajectory of the ball and the trajectory position of the observed player, if the average position change rate is less than a second threshold, the position change rates of the movement trajectory of the ball and the trajectory position of the observed player are considered to be consistent; and
wherein the processor is further configured to:
detect whether the facial ROIs of the valid players contain binocular eyeballs with closed contour;
calculate the average value of the sum of roundness of the eyeballs when the binocular eyeballs with closed contour are contained; and when detecting a plurality of video angles that contain the perfect eye features, calculate the average value of the sum of roundness of a plurality of eyeballs in each ROI and select the angle video corresponding to the maximum average value as the best angle video; and when the facial ROIs of the valid players do not contain binocular eyeballs with closed contour, the video from the current angle is considered invalid, and videos continue to be selected from other angles until all ROIs are retrieved to see whether they contain the perfect eye region features, which ensures that all ROIs are traversed and improves the detection accuracy; and
output all the ROIs that are retrieved to contain the perfect eye region features.
|