May 2020 – Nov 2022 · 2 yrs 6 mos · Bengaluru, Karnataka, India

• Worked on Driving Test Automation.
• Trained an UNet for car segmentation with a dice coefficient
• of 92.60%.
• Created a driving test line and speed violation pipeline with
• the above model.
• Synchronized different angled camera streams in the same
• pipeline for reduction of perspective errors.
• Worked on Gait Recognition using GaitSet and UNet.
• Created a gait sequence dataset with more than 50-person
• data.
• Trained GaitSet model for gait feature generation with an
• accuracy of 96.00%.
• Created a pipeline for gait recognition using the above model
• with a pretrained Unet model for person segmentation and a
• KLT tracker for gait sequence generation.
• Achieved an accuracy of 75.93% for the pipeline on the
• above dataset.

May 2019 – Apr 2020 · 11 mos · Bengaluru, Karnataka, India

• Worked on Traffic Violation Detection.
• Trained a TinyYOLO deep learning model for detecting traffic
• violations like helmet violations and seat belt violations with
• mAP of 83.54% on client data.
• Optimized the deep learning model by using TensorRT to
• increase the fps of the model from 3 fps to 16 fps.
• Optimized the processing pipeline by using vectorization to
• increase the fps of the pipeline from 2 fps to 7 fps.
• Added object tracker in the pipeline to reduce the detection
• of the same traffic violation multiple times.
• Deployed the deep learning model and pipeline on the re-
• mote camera connected with Jetson Nano.