My expertise include doing computer vision using OpenCV, tensorflow, mediapipe, Nvidia SDK.(in decreasing order of expertise)
I have worked on a couple of pose estimation(VideoPose3d, movenet, BlazePose, openpose) and object detection(efficientdet, YOLO, mobilenet) and tracking(lucas kanade, siamask, deep sort) methods as well.
Other miscellaneous skills also include perspective transformation, trajectory of moving objects and 3D geometry.
I have used state of the art models with fine tuning as well as created some basic models from scratch. I have some experience with techniques like quantisation with TensorRT for increasing the speed of models and also have implemented them on both servers(AWS) as well as edge(Jetson- nano, Xavier, TX2). Other optimization I am slightly familiar with are openvino for myriad/OAK, TFlite for raspberry Pi.
My other skills include using creating sophisticated and intelligent algorithms using all the above tools which are worth some value. Also for vision with deep learning I create Synthetic data using tools like Unity3D for times when data is not enough or properly balanced.
I am also very good at tinkering(hardware assemble and software both) Linux systems and set them up for research work environment with all NVIDIA GPUs Cuda capabilities and popular deep learning frameworks building from source/pip/conda.
Sardar Vallabhbhai National Institute of Technology, Surat — M.tech
Vishwakarma Institute of Information Technology, Pune — BE, Pune University. 72% First Class with distinction
Spearheaded development of an advanced cricket ball tracking system integrating precise camera calibration, real-time ball detection, and small object detection at 400 FPS in 4K resolution. Utilized GPU Direct across 10 cameras to reduce ball trajectory prediction error by 90% compared to state-of-the-art methods.
Engineered a novel on-ground camera calibration method using a total station combined with bundle adjustment, achieving millimeter-level triangulation accuracy from distances up to 200 m.
Developed a real-time scene classifier for QuickFlip—the world’s first vertical feed for sports—to enhance content relevance and viewer engagement.
Developed end-to-end Jetson-based surveillance & inventory detection for Ivory Soaps to detect inventory theft and monitor security using object detection and GenAI for pharmaceutical research.
Enhanced synthetic data model training for AuraML, improving YOLOv5 fine-tuning by 25% mAP, and implemented SLAM-based navigation with photogrammetry for 3D environments.
Created a novel algorithm to detect abandoned objects on highways using CUDA, MOG2, and object detection with 94% accuracy.
Developed highly optimized models and algorithms in C++ for Nvidia Jetson boards, leveraging DLAs to run multiple models concurrently on shared resources.
Developed 3D computer vision-based human body tracking and advanced human motion analysis for exercise routines, including detection of equipment and usage patterns during workouts.
Developed an ID card recognition system to extract information/photos from PAN, AADHAAR, and international passports (including MRZ codes).
Managed traffic density analysis by controlling traffic signal countdowns using TensorRT and YOLOv4 on Jetson devices; provided detailed traffic stats for toll plaza planning using DeepSort.
Implemented pothole and garbage detection on car cameras using EfficientDet and TensorRT, reporting GPS locations to government entities (BBMP and KRDCL).
Utilized object detection, pose estimation, and ConvLSTM for smart city surveillance to identify potential security threats and monitor human activity.
Developed exercise video analysis solutions including E-Gym tracking for Koofit and Mediapipe BlazePose–based exercise tracking for UPUGO.
Performed trajectory analysis of sports balls using object detection and tracking algorithms, visualizing data on a 2D pitch map with heatmaps.
Developed athlete form correction methods for cricket batting and bowling using advanced pose estimation techniques.
Automated B2C marketing campaigns and analytics across major social media platforms.
Developed object detection–based hashtag recommendation systems and performed sentiment analysis on large-scale social media data.
Accomplished Indian strength sport athlete with about a dozen awards on national level, state level(Gujarat and Maharashtra)and University Games.
Shortlisted to represent India for International powerlifting games July 2020
Qualified GATE 2019 with 2326 All India Rank