Experience

IntelliSee

• United States
• Public Safety
• 1 - 100 Employee

• Computer Vision Engineer

  • Jul 2021 - Present

  Developing Deep Learning and ML models for the real-time video analytics Computer Vision risk detection and risk mitigation product with a mission to make the world safer. Working on rescaling the frames parsed on video decoding to increase the spatial resolution on the frames. Optimizing the detection capability of the model Model evaluating and testing. Profiling the real-time GPU access and the real-time memory allocation inside the pipeline framework with NVIDIA Nsight. Developing Deep Learning and ML models for the real-time video analytics Computer Vision risk detection and risk mitigation product with a mission to make the world safer. Working on rescaling the frames parsed on video decoding to increase the spatial resolution on the frames. Optimizing the detection capability of the model Model evaluating and testing. Profiling the real-time GPU access and the real-time memory allocation inside the pipeline framework with NVIDIA Nsight.



University of North Carolina at Charlotte

• United States
• Higher Education
• 700 & Above Employee

• MS Thesis | Graduate Research Assistant at TeCSAR Lab

  • Sep 2020 - Feb 2021

  Got funded by VDOT to work collaboratively with Leidos team. • Reduced the risk of the VDOT laborers involved in the manual inspection of the highways asset items•Automated the inspection process by building an efficient and scalable AI framework for highway asset detection• Curated, augmented, and preprocessed the custom dataset for customized highway assets detection• Developed an algorithm for distributed training in Pytorch framework and scaled single highway asset dataset for localization into customized multi-object scene dataset for multi-object detection training on the NVIDIA TitanV GPUs• Developed and end to end vision pipeline with converting the trained customized highway assets detection Pytorch FP32model into an ONNX graph followed by generating optimized FP16 TensorRT engine for the deployment on the mobile edge NVIDIA Jetson AGX Xavier module achieving an inference benchmarking of 50 FPS with an aim to achieving real-time onboard processing Show less

• MS Thesis, Graduate Research Assistant at TeCSAR lab

  • Jan 2020 - Aug 2020

  With an aim to devise the detector and anomaly prediction framework on edge-cutting mobile devices, I had researched and learnt the conversion of TensorFlow model configuration conversion to fully quantised 8 bit INT Tensorflow lite model based on tensorflow to deploy on the Google Coral TPU accelerator. Achieved TF-TRT i.e, Tensorflow to Tensorrt conversion of the Efficientdet-D0 model with inference results of 25 FPS on the AGX Xavier.Learnt the MobileNet SSD model architecture to replace the backbone with a lighter version efficientnet backbone. Show less

• Graduate Teaching Assistant

  • Jan 2020 - May 2020

  Conducting the Labs for Undergraduate Students on Signals and Systems. Teaching is a good practice to keep learning. It also helps in strengthening the confidence and boosting the communication skills.

• Graduate Research Assistant

  • Nov 2018 - Jun 2019

  • Studied and learnt various algorithms such as Stereovision, Multi-view Stereo technique, Structure from motion, etc. • Deployed drone to capture aerial imagery and 2D thermal images with a FLIR thermal camera. • Implemented Feature matching, Feature descriptors algorithm such as SIFT, SURF, etc on all the 2D Images in OpenCV. • Generated dense 3D point clouds using Point Cloud Library. Built a 3D reconstructed model out of point clouds using meshing technique in the MeshLab. • Studied and learnt various algorithms such as Stereovision, Multi-view Stereo technique, Structure from motion, etc. • Deployed drone to capture aerial imagery and 2D thermal images with a FLIR thermal camera. • Implemented Feature matching, Feature descriptors algorithm such as SIFT, SURF, etc on all the 2D Images in OpenCV. • Generated dense 3D point clouds using Point Cloud Library. Built a 3D reconstructed model out of point clouds using meshing technique in the MeshLab.



Siemens Electric, Pvt Ltd

• Project Engineering Intern

  • Jun 2018 - Jul 2018

  •Worked on detecting the incorrect and faulty electrical ratings on the electrical rating plate •Built an algorithm in OpenCV PyTesseract to extract the letters and digits from the rating plates and comparedit to the ideal rating plate using the classical computer vision image difference technique •Proposed to implement the project in real time for the efficient, cost effective and avoiding faulty manufacturing of theelectrical rating plates to maximize the profit of the company •Worked on detecting the incorrect and faulty electrical ratings on the electrical rating plate •Built an algorithm in OpenCV PyTesseract to extract the letters and digits from the rating plates and comparedit to the ideal rating plate using the classical computer vision image difference technique •Proposed to implement the project in real time for the efficient, cost effective and avoiding faulty manufacturing of theelectrical rating plates to maximize the profit of the company