Method for segmenting video using background model learned with pixelwise adaptive learning rate

摘要

#$%^&*AU2017265110A120190606.pdf#####ABSTRACT A method (400) of segmenting a current image (426) of a video (425) capturing an action on a pitch in a scene, the method comprising the steps of receiving (410, 420) a background model (405) of the scene and a pitch mask (415) of the pitch; determining (440) a foreground mask (445) for the current image (426) based on the background model (405); identifying (470) 5 connected components of pixels within the foreground mask that are connected to the pitch mask, said identified connected components being referred to as on-pitch connected components; updating (480) the background model (405) of the scene using a variable learning rate for each pixel of the scene, the variable learning rate based on a degree of proximity between the on-pitch connected components and said each pixel, to thereby 10 construct an updated background model; and segmenting a subsequent image of the video based on the updated background model. P273917 / 13929085_2-1/10(Wide-Area) Communications Network 120 Printr 11 Microphone 124 180 121 117 (Local-Area) Video Communications Disp Network 122 10 A - Modem Loca NA Pt 116 100 U5 101 Audio-Video I/O Interfaces Locl13311Devirg toag Interface 107 108 I/face 111 HD 3 Devices 104 118 119 Processor I/O Interface Memory Optical Disk 105 113 106 Drive 112 Keyboard102 Scanner 126 Disk Storage 103 Medium 125 Camera 127 P273917 /13929094_1 Fig. 1A