3 cushion video object detection and auto cut to short clip

Automatically cut 3-cushion video into short clips. Use object detection to detect balls, table, cushion, ground. Divide ground into 512 square grid (e.g., A1–E5) and locate each ball’s position in the grid. Export results as output." Our video is a long recording with multiple non-fixed camera angles. I need to build a model that can detect objects regardless of the camera view. The automatic short clip cutting should start just before the balls begin to move and end when the shot is completed and all balls stop. We will split the long video into many short clips, each representing one shot. Then, we will name each short clip based on the detected coordinates of the three balls on a 512-square grid over the table surface — for example: a1-c3-d7.mp4 , i need expert on model ai like yolo mark r cnn , htc swin transformer and open cv and python