Anubhav Rohatgi
Expert Developer in Computer Vision & Robotics
EXPERIENCE
Intello Labs Pvt Ltd
– Present
- Set the technical direction for the Deep Learning and Backend Engineering teams. Lead, motivate and mentor a team of rockstar developers.
- Supported and managed talent acquisition and improved hiring timelines.
- Designed and developed high-speed large-scale object detection and instance segmentation on high resolution images 8000 x 10000, used for quality inspection of spices.
- Pioneered innovative high-speed 360-degree scanning using CMOS scanners for spices and grains for quality inspection and grading using Instance Segmentation, patent pending.
- Improvised low latency computer vision based apple detection and segmentation method using differential colour spaces used in the apple sorter machine with an accuracy of 97%
Nayan India Science & Technologies
Co-Founder, Vice President Engineering, – Mar 2022
- Designed & Orchestrated Kafka Backlog Based Kubernetes custom Horizontal Pod Autoscaler for AI model inference on a GPU resource-limited node.
- Develop 2D/3D Keypoint localization using CNN for automatic pose estimation of cars.
- Innovated Automatic Driver Assessment System for Smart Yard using a multi-camera system to detect and track car and assess driving behavior applying visual odometry. Patent Pending.
- Deep learning-based Lane Occlusion classification with 91% accuracy validated on real feeds.
- Work with cross-domain teams like DevOps/QA/Data Analytics/Annotation Specialist teams on technology decisions and technical designs
- To help develop backend micro-services for C3 crowdsourced data annotation engine with automatic data train-validate-test and model retraining capability with KPI dashboards
- Developed the logical architecture in support of the overall vision of the platform, identifying all of the required components and subsystems, including their individual responsibilities.
Vihaan Networks Limited
Sr. Lead Engineer – Feb 2020
- Designed and Developed Video Management System back-end.
- Integration of VLC library in VMS and Develop plugins for Real Time Video
Analytics, perimeter surveillance, face recognition, abandoned baggage
detection, Bipeds detection on FLIR video feeds. - Edge Analytics - Raspberry Pi + Intel Compute Integration with the camera for edge based real time video analysis for event triggers.
- System Integration – C/C++, VLC library, OpenCV, TensorFlow.
CoinTribe Technologies
Lead Researcher & Developer
- Consistently met deadlines and requirements for all production work
- Designing scalable system architecture and micro services.
- Face Recognition using Siamese Net.
- Highly Scalable, Modular OCR and Data Extraction RESTful services for images
- Leading a bunch of data scientist geeks and disruptors striving for a big breakthrough success in data extraction from different types of documents.
- Research and develop techniques and algorithms for object detection, segmentation, and recognition
- Machine Learning, Probability and Statistics, Linear algebra
- System Integration - C++ and Web (Node, JS, HTML & CSS)
CoinTribe Technologies
Sr. Researcher & Developer Jan 2016 – Jun 2018
- Document Image Analysis – Enhancement, Detection and OCR of complex unstructured text and graphical scanned documents. KYC Documents recognition e.g. ID cards, Insurance cards etc.
- Text Detection and Recognition with Image reconstruction and extraction of tables
- NLP – Dictionary lookup and correction of OCR results
- Transactional Narration Classification
- Optical Character Recognition through machine learning. Tesseract OCR-CNN-LSTM Training.
- Detection of Tabular data
- Optical Character Recognition through machine learning. Tesseract OCR Training
- Logo recognition and matching.
- RESTful API Programming for server and integration
- Image Binarization
- Face Detection and Matching using CNNs
VisioLucid TechnoSoft Pvt Ltd
Sr. Researcher & Developer Feb 2014 – Jan 2016
- Visio Ingenii's India Office- Merger between Lucidient and VisioIngenii
- Heading research team of 6 members .
- Crowd Detection and Management Video Analytics
- FHoG, Kalman Filters and Background Modelling.
- Visual Search and object classification in Videos
- Face Recognition using Deep Neural Networks and High dimensional LBP.
- Object Recognition and Image Retrieval.
- GPS Tracker (Hardware and Software (Android)).
- Energy Scavenging – Battery Charger – Body heat to Voltage using Linear Ultra Low Voltage Step –up converter and power manager.
- Other Projects: Online Video Stabilisation, Video Annotation, Bag of Visual Words, Augmented Reality, Real Time Video Annotation, Blur Detection, Camera Tampering. Omega Model for Humans.
Visio Ingenii
Researcher in Computer Vision Algorithms – Jan 2014
- Devising improved algorithms in Digital Image Processing using OpenCV.
- Particle Filters using Recursive Bayesian Estimation.
- Pattern Recognition
- Motion Detection
- Media Layer for Augmented Reality (OpenGL & Unity3D)
- Face Recognition
- Fire and Smoke Detection using wavelets
- Object Tracking - Moving Object and Static background as well as Moving Background and Moving Object.
Dr. Harry Witchel
Research Assistant – Oct 2012
- Analysis of various stimuli used in experiments to study body language.
- Mathematical representation of data using MS Excel and MATLAB/
- Signal Processing using MATLAB to perform transformations and regression analysis.
- Implementing various algorithms to optimize results and minimize errors
University of Sussex
Part Time Data Research Assistant – Jan 2012
- Analysed and tabulated the entry requirements for the 1994 group of institutions in UK.
- Involved data extraction from various sources.
- Timely delivered reports.
Telerex Etech
Graduate Engineer Trainee (GET) – Apr 2010
- As a Field-Engineer supervised the installation and serviced EPBAX systems.
- Programmed EPBAX CPU using the OMC software according to the client’s specifications.
- Designed the layout for carefully laying out the cables and conducted the final tests.
- Inspected the splicing and integration of cables and connectors.
- Identified and analyzed the programming faults and repaired PCB cards in the OXO cabinet.
EDUCATION
University of Sussex
Master of Science Embedded System Design
Completed with distinction 74.330%
Activities: General Secretary of Indian Student Society,
University of Pune
Bachelor of Engineering Electronics & Telecommunication
Attained 1st Class
Modern School
School Science
Very good mathematical programming and analytical skills,
Activities: Actively participated in Science quizzes and debates, Athletics, Mathematics Olympiad, Intel Science Talent Search Participant.Projects
Photos & videos from my recent projects.
Link to my Repo : Press Here
Hexapod Robot
A study of gaits
Individual project. Interfacing Basic Atom Pro processor and SSC-32 controller in Master/Slave configuration for an existing Hexapod chassis. Programming in MBASIC language to realize complex walking algorithms. Implemented basic locomotion using Inverse Kinematics. Extensive use and in-depth understanding of PWM generation algorithms.
Camera Tampering Detection
State-of-the-art Tampering Detection
Algorithm deals with camera tampering imposed by the actions like : blocked view, change in viewing angle, camera shake, camera defocus, change in camera scene and change in (>70%)illumination. System is based upon proprietry state-of-the-art algorithm. This incomplete GUI was my first hands on with QT.
Crowd Analytics
Crowd Management System
Crowd Analytic based on video. Crowd Density Heat-map Group Tracking and Direction Determination and Prediction. Used C++11, TBB and OpenCV
License Plate Detection
ANPR on Static Images
Number plate detection performed on images. The system works in real time even on videos. The algorithm behind this is not starightforward wherein we could have performed some edge detection, morphological operations and then searching for rectangle. The OCR part is proprietry and hence not a part of this demo video. Designed in pure C++ & OpenCV and not yet multithreaded yet it runs at 25fps on 640x480 videos. Works at different angles.
Video Stabilization
My first attempt at stabilization
In this I am not using OpenCVs stabiliz-ation module. The pipeline is a composite approach from optical flow detection, direction of motion blur, perspective tran-sformormation, wiener deconvolution and some pre processing.
UAV Video Stabilization
My second attempt at stabilization
Second attempt at stabilizing the UAV aerial footage. Almost near. Next target :- online full frame stabilization. Attempt "Warping" as used by Google and Adobe. I dont know yet if that can be done in real time. "Hang on" while I dive into that subject soon.
Fire Detection in Video
Real Time Video based Fire Detection
Real Time Video based fire detection for both outdoor and indoor scenarios. State-of-the-art system works on aerial videos as well. The algorithm is not only based on Color Segmentation but is accompanied by various dynamic analytics like wavel-ets, than color based static analytics.
Simple OCR
Number recognition on images
A simple ocr example using Tesseract. An overview of steps to be performed for OCR on an Image.
ARInk
Simple Multi-Threaded Responsive UI for Filter Effects
Simple multi-threaded QT UI for applying different filtering effects on the image.
OpenCV + QT + C++
Multi Target Tracking
Fast Multi Target Tracker
MTT using modified KCF filter. Trained over a set of 3 key features namely DPM(Deformable Parts Model) , LBP and grayscale.
SMD Based Circuit Design
First hands on smd soldering and circuit design experience with SMD chips LTC3109
Energy Harvesting using peltier modules. Circuit design using SMD LTC3109 and SMD coils from coilcraft. Circuit designed using PCB express, then printed on glossy paper, then ironed on the board and compenets soldered using the hot air soldering gun. Flawless mounting. but bad design since the circuit is energy harvesting I overlooked the power dissipation and voltage over the track length.
Face Recognition
Face Recognition & Tracking
Face Rec using OpenCV + QT (GUI + Qt-concurrency) + Siamese Net. The tracking is done on each frame but detection on every 15th . Used SIMD optimisations. Landmark Model borrowed from Dlib and Simaese Net trained using Tensorflow. Tracking using Kernelized Correlation Filter working on HoG features and Colour Space.
Automatic Foot Posture Detection
Detect foot posture one at a time for estimating misalignment of feet
Identify the alignment of foot one at a time to recommend right insoles to rectify the misalignment pronation or supination using machine learning.
Github Repo Link
A link to my repository
A link to my repo on GitHub.
Patents Link
A link to my patents
Link to my patents.
REFERENCES
Jayesh Chhatpar
SSD Hardware Design Engineer at Micron Technology
Anubhav and I have graduated from University of Pune, and we have taken several classes together and worked on many projects. During that time, I have gotten to know Anubhav quite well and can thoroughly vouch for his character and abilities. Anubhav is a pleasure to work with because his pleasant and positive attitude can make nearly any work seem fun and interesting. He always brought new ideas to the table for every project. His ability to explain ideas clearly, express his views and address concerns showed leadership qualities in him. He always paid attention to details and took the initiative to make sure the project was done exceptionally well.
Manas Dalal
Senior Research Engineer at Visio Ingenii
Anubhav is extremely technical! which makes conversation with him even more interesting! He has a tremendous grasping capability which makes him suitable for any project. He is very intellectual and knows what he does. He has excelled in his academics with perseverance and smart work! A wonderful guitarist and a true friend.
Pavan Hanagal
Senior Software Architect at VisioLucid TechnoSoft
Anubhav is very dedicated Employee in our Company. His work is more precise and up to the mark. It's been pleasure working him.
Good Luck!!!LET'S CHAT!
