Nitish Kumar

Built an extra tab in the GUI using Tkinter to send missing data to the Tracksure portal along with its backend functionality using multi-threading. Created SKU classification model using colour masking and a deep learning (SqueezeNet) model, which gave an accuracy of 99.95%. Used SqueezeNet to design two process classification models, one for SKU and another for packet size, which gave accuracy of 100% on 15000 testing images.

Worked on Global Navigation Satellite System Jamming and spoofing of Unnamed Aerial Vehicles within a range of 10 Kms. Used GNU radio software, MATLAB and GPS-SDR-SIM simulator to generate and transmit the signal using HackRF-One and proper antenna. Achieved our objectives with 100 % accuracy for GPS frequencies within 20 metres and jammed all GNSS frequency bands.

Performed various Advance Driving Assistance System (ADAS) tasks like car parking assistance, object identification, parking sign detections. Developed an algorithm to detect and inspect parking spaces and to see if they are vacant, the algorithm worked for both labeled and unlabelled data. Effectuated models MobileNetV2 for Parking Context Recognition (PCR) and YOLOV3 for Parking Slot Detection (PSD) further tested on realistic data. Achieved F1-score of 98.29% on Parking-slot dataset 2.0 and 84.81% on 4518 realistic images in which 4026 have the parking slot.