Aug 2020 – Jan 2024 · 3 yrs 5 mos · Bangalore Urban, Karnataka, India · Remote

Jan 2020 – Apr 2020 · 3 mos · Kharagpur, West Bengal, India

• Aim of the project was to develop a robust detection and tracking system of eyelids, eyeballs, lips, and head in order to efficiently monitor the activity of the driver while driving through a single camera.
• For detection and segmentation of facial regions, Mask Region Convolutional Neural network, or Mask R-CNN developed by Facebook AI Research Team (FAIR), an extension of faster R-CNN is used.
• To track the detected regions, Median Flow Tracker is used which is based on forward-backward error. The error is calculated by taking the difference of forward trajectory and backward trajectory.
• The mean absolute precision of the training set is 92.5 percent and the validation set is 96.2 percent. The proposed tracker works very well as long as the object is in the frame, once the object disappears from frame algorithm fails.

Aug 2019 – Nov 2019 · 3 mos · Shenzhen, Guangdong

• Responsible for the development of a counting system to count objects coming on the conveyor belt in order to increase counting accuracy, faster counting and can aid in reducing labor cost.
• Used background subtraction algorithm for the foreground detection which aims to detect the products coming on a belt through a single camera positioned at top of the belt.
• In order to track the centers of the detected objects, Kalman Filter along with Iterative - Hungarian algorithm is used. To increase the count, a line based method is used in such a way that if the contour's center is greater than line in nth frame and less than line in the n-1 frame, the count is increased.
• The maximum counting accuracy of the system with a single camera is 96.6 percent.
• PyQt5 and Qt Designer is used in order to create an user interface that can aid in increasing functionality of the system and can be used on wide range of products with varying dimensions.

May 2019 – Jul 2019 · 2 mos · Dongguan, Guangdong

• The project focuses on the development of the Automatic Visual Inspection System for light-emitting diode wafers containing thousands of LEDs in order to classify every LED to faulty or non-faulty.
• In order to extract LEDs or foreground from background Otsu Thresholding along with Morphological Operations are used. After this step, every LED from the wafer is separated from the background.
• To classify the extracted LEDs, a deep convolutional neural network is trained. The classification results along with the co-ordinates of LEDs are then sent to the sorting machine to remove faulty LEDs.
• Accuracy on the training set is 93.6 percent and validation set is 95.86 percent.

Jul 2018 – Jul 2019 · 1 yr

• •Coordinator of Gold winning Stage play and Silver winning Open-IIT street play among 20 participating teams

Jul 2018 – Jul 2019 · 1 yr

Jul 2017 – Apr 2018 · 9 mos · IIT Kharagpur

• Aggregated a total sponsorship of INR 30,000 from local sources to organize an introductory event for 1400 freshmen of the institute.
• Supervised a team of 28 Associate Members to organize 7 events which witnessed total participation of 700+ student.
• Enhanced participation by 28% YoY with designing 25 publicity posters, 2 event banners; each having a reach to 20,000 people.