Embedded Files
Research Projects
Research Projects
Multi-camera Markerless Motion Capture (MMC) - May 2022
Multi-camera Markerless Motion Capture (MMC) - May 2022
Mentors: Dr. Samarth Brahmbhatt, Dr. Rakshit Kothari and Dr. Bhupendra Fataniya
Students: Toshita Sharma, Maitry Dumaswala
Motivation: Machine learning (ML) has revolutionised computer vision based applications and contributes positively in almost every sector of our daily lives. ML models based on Neural Networks typically consist of millions of tunable parameters which are adapted to the data and task a model is trained on. To ensure that a ML model can generalise to everyday applications, we require large amounts of data captured across a wide distribution of variables. Variables include race, gender, age, viewing conditions, camera quality etc. The end goal of our MMC system is to develop a flexible, easy to understand and use dataset acquisition system for a wide variety of applications. Applications include gaze and head pose estimation, body pose estimation, hand pose estimation, group speech to text, emotion recognition, social interaction etc.
Expected output
A fully functional and well documented MMC system with data quality assessment using Machine Learning and Computer Vision
Open source code with the ability to provide active support for future students who will depend on this project
Project should be developed in a modular fashion to allow more cameras / sensors to be attached
A well documented report and detailed project slide deck
Collision avoidance using Jetbot - January 2023
Collision avoidance using Jetbot - January 2023
Mentors: Dr. Samarth Brahmbhatt, Dr. Rakshit Kothari and Dr. Bhupendra Fataniya
Students: Arya Tripathi (20BEC130) , Yash Purohit (20BEC137 )
About Project:
The aim is that the jetbot avoids any obstacle that comes in its way. To train the jetbot, a custom dataset was created using the openCV library where the obstacles were added to the blocked class and Jetbot's free path was added to the free class. Using PyTorch, the model was trained for 125 epochs with a batch size of 8 images for a dataset of about 560 total images and was later trained using the AlexNet model.
Page updated
Report abuse