A Sports Video Behavior Recognition Using Local Spatiotemporal Patterns
Algorithm 1
Athlete behavior recognition algorithm based on deep spatiotemporal residual convolution neural network.
Input: Sports Video
Output: Behavior categories of all athletes
Initialization: Initialize target detection model D, target tracking model T, and behavior recognition model A
Initialize the athlete set image M<ID, S>, each athlete S has a cache C that stores the state image and a behavior v
For frame in video
boxes = D(frame)/ Get all athlete target images by athlete target detection /
IDs = T (boxes)/ Update athlete image assignment IDs by athlete target tracking /
For ID in IDs
If ID not in M:
M.add(ID, s)/ Add the nonexistent athlete S to the athlete set M/S.C.add(frame [ box ])/ Store the corresponding athlete target image to the corresponding athlete’s cache C/
If S. C. size = 16: S.v = A(S.C)/ Identify athlete behavior by behavior recognition model /
