At Quidich Innovation Labs, I calibrate six-camera systems for 3D point mapping and build ground-truth datasets to evaluate geometric accuracy. I also design and train deep learning pipelines for real-time video matting and foreground segmentation in broadcast footage.
I’ve developed real-time computer vision pipelines for multi-camera sports video, including object detection and multi-object tracking. I’ve trained and evaluated models on footage with occlusion, rapid motion, and changing camera viewpoints, and optimized inference for low-latency analysis across camera feeds.
As a Computer Vision Engineer Intern, I fine-tuned YOLO models to detect small, fast-moving objects in cricket video and integrated DeepSORT with YOLO detectors for temporal tracking. My research has also included sports video analysis and using Twitter data for landslide prediction.
