At TATA Elxsi, I developed a computer vision proof of concept to detect foreign objects and railway signals on rail tracks using YOLOv8.
I collected and labelled a custom rail-track asset dataset, then applied GenAI-assisted augmentation, image preprocessing, and annotation to improve dataset diversity and model robustness.
I built training and validation pipelines, trained the YOLOv8 model, and evaluated it with mAP and precision-recall metrics. I also wrote Python scripts to streamline training workflows and dataset management.
On TATA Elxsi’s image pre-processing automation project, I used an SDK to process and optimise raw images, apply enhancement and retouching techniques, and analyse histograms to address dead and over-brightened pixels. Earlier, I conducted SIL testing for ADAS-LTA and LDA features, re-simulated vehicle data, analysed events, raised issues, and prepared reports.
