Optical Flow to Improve Image Segmentation
Combining YOLO with optical flow for more robust object indexing
A successful indexing of cars and pedestrians using YOLO and optical flow.
A team coursework project augmenting You Only Look Once (YOLO) with optical flow to improve object tracking consistency between frames. YOLO detects objects well but can occasionally swap identities between them across frames—optical flow provides a cheap geometric prior on where each object should have moved, making the indexing more robust to the occasional missed detection.
It was a modest project, but it was one of my first at Penn and gave me a solid grounding in image processing, camera integration, and applied machine learning that informed later work.