Horse Video Auto-Cutter

I need a program that can automatically identify the walking direction of a horse in yearling parade videos, detect the frame where the horse is most perpendicular to the camera, and extract a standardised clip around that moment.

Requirements:
- Process MP4 yearling parade videos
- Use YOLO to detect the horse and SAM2 (or YOLO-seg) to generate a segmentation mask for each frame
- Calculate the mask centroid per frame and determine walking direction based on consistent left-to-right X-movement
- Select only the frames where the horse is clearly walking left-to-right
- From those frames, detect the moment where the horse is most perpendicular to the camera
- Extract a standardised segment around this point (e.g., ±75 frames or ±2 seconds depending on FPS)
- Exclude videos that do not contain enough valid frames after the perpendicular moment
- Output the clipped video using the original file naming structure

Ideal Skills and Experience:
- Strong experience with computer vision and video processing (Python, OpenCV, ffmpeg)
- Proficiency with object detection and segmentation frameworks (YOLO, SAM2, etc.)
- Ability to compute and analyse per-frame motion metrics (centroid tracking, direction inference)
- Understanding of pose/angle detection or geometric heuristics for “perpendicular” classification
- Experience building efficient pipelines capable of batch-processing large video datasets
- Clear attention to detail and ability to implement robust quality-control rules

Looking for a freelancer who can deliver a fast, scalable, and reliable solution suitable for processing thousands of videos.