Monocular Depth Estimation deep learning model development

 Object Detection and Semantic Segmentation

Pruning, Quantization aware training and knowledge distillation for model compression

Developed pipeline for dense 3D reconstruction from drone aerial videos using COLMAP library.

Integrated drone’s GPS data to achieve accurate videogrammetry of the 3D reconstructed environment.

Implemented a pipeline to identify and rectify missed and false detections in YOLO object detection model for effective tracking using Deep SORT

Developed a cost-effective navigation stack (MinNav) for tiny aerial robots using only a monocular camera.

Used optical flow and its aleatoric uncertainty predicted from a single deep learning model for obstacle avoidance.

Deployed and evaluated MinNav on a tiny drone, achieving comparable results to depth-based systems while significantly reducing processing times.

Tested and developed transmission system for two-wheelers to comply with the BS VI emission standards.

◦Implemented an automated pneumatic system to reduce kick-start endurance testing cycle time by 90%.

 Received work commendation award for two years (2019 and 2020) consecutively.