«I am currently working on the development of an innovative optical tracking system for robotic surgery. The array of cameras is capable of tracking objects’ 3D positions and communicating with robots fully autonomously. All the computations are handled on-board. Due to confidentiality reasons, I am not authorised to detail the system specifications.
From 2014 to 2018, I worked at Atracsys, a company recently acquired by Smith & Nephew. In the early stage of the project, as project manager and main software contributor, I actively worked on the first prototypes of cameras using development kits and contributed to its development into a medical grade product. All along the project, I contributed to the development of calibration tools in order to reach a sub-millimetric multi-view 3D position tracking. To improve the system's performance, I have co-authored a patented method that reduces cameras' readout delay without losing tracked scene features. Overall, I was involved with various stakeholders; physics researchers for the camera lens calibration, electronic engineers for platform based optimizations, mechanical designers concerning manufacturing constraints and medical regulatory delegates for clinical tests.
Since the end of 2018, I have continued to work on this project as consultant for Think Surgical. I have been working remotely from Zürich with Montreal and California based R&D engineers. I have contributed to the integration of the multi-view camera array within a behavior-tree based navigation software controlling the robotic system. I have also participated in the development of additional engineering debug tools that evaluate the guided-surgery robot performances.
This contribution to the development of an advanced robotic system in the medical field is, so far, the greatest professional experience I ever had. This managerial and coding experience on a challenging industrial project taught me to work with various collaborators and to design a time-critical embedded software with a deep understanding of the science of optical measurements.»
Experience
Profile not up-to-date
Coming soon... Feel free to write me!☎ Contact Form ✉ maxvell.arulraj@gmail.com
Medical grade 3D positioning tracking system for robotic surgery
Think Surgical Society (Montreal, Canada & Bay Area, United States) - Team size: 10 pers. Company size: ∽190 pers.Atracsys Society, acquired by Smith & Nephew (Lausanne, Switzerland) - Team size: 7 pers. Company size: ∽25 pers.
- Main software contributor of a multi-views camera system development
- Multi-threaded, multi-process, real-time 3D position tracking algorithm in C++
- Time critical algorithms interfaced with FPGA logic fabrics
- Multi-views camera calibration tools for tracking with sub-millimetric precision using non-linear numerical optimisation techniques
- Co-author of a patented method that reduces cameras' readout delay without losing tracked scene features
- Medical device software development classified IEC 62304 class C (highest standard)
- Intellectual Property analysis in medical optical tracking systems for robotics
Cross-platform Augmented Reality SDK
Vidinoti Society (Fribourg, Switzerland) - Company size: 5 pers.
- Real-time 2D object recognition & features tracking algorithms (used Oriented FAST and Rotated BRIEF features, SIFT keypoints detectors)
- Integration of the core C++ recognition engine on iOS and Android
- Plenoptic camera with AR proof-of-concept prototyping
- Intellectual Property analysis in Augmented Reality technologies
- Web application for content creation
«I contributed to the development of a cross-platform Augmented Reality SDK. The library allowed developers to embed AR functionality into their own iOS and Android applications. The SDK was able to recognize 2D targets in real-time and to overlay AR contents. Using a web-based interface, customers were able to create interactive AR content such as 3D animations, audio or videos in a few clicks.
Within a small team of three engineers, I worked on the development of the core recognition and tracking engines in C++. I created evaluation metrics and training tools for profiling and optimising the selection of invariant features. Based on research papers, I explored ideas to continuously reduce the recognition delay. I optimised the tracking of the 2D targets within critical illumination, wide angles and partial occultations. Each target image was first analysed offline to create a unique model of bag-of-words with invariant keypoints. The on-device recognition library was then run within an application for matching each camera frame with these pre-registered targets in real-time. When frame descriptors are matched with one target model, the computed homography is refined for estimating the orientation and pose of the target and the attached AR content was rendered on the top of the camera frame data.
This project was a challenging experience with a business-to-consumer product. I learned to work on an already existing and deployed product used by thousands of developers. It has been my first advanced coding experience on a sophisticated multi-platform software.»
Jan. 2013 - Oct. 2014
Within a small team of three engineers, I worked on the development of the core recognition and tracking engines in C++. I created evaluation metrics and training tools for profiling and optimising the selection of invariant features. Based on research papers, I explored ideas to continuously reduce the recognition delay. I optimised the tracking of the 2D targets within critical illumination, wide angles and partial occultations. Each target image was first analysed offline to create a unique model of bag-of-words with invariant keypoints. The on-device recognition library was then run within an application for matching each camera frame with these pre-registered targets in real-time. When frame descriptors are matched with one target model, the computed homography is refined for estimating the orientation and pose of the target and the attached AR content was rendered on the top of the camera frame data.
This project was a challenging experience with a business-to-consumer product. I learned to work on an already existing and deployed product used by thousands of developers. It has been my first advanced coding experience on a sophisticated multi-platform software.»
Android application with cloud-based face detection
Personal project - Team size: 2 pers.
- Face tracking Android application for demonstration purpose
«Based on the knowledge I learned from my previous experience, a friend and I decided to create an application to detect faces on smartphones. The core engine was developed in C++, inspired from the OpenCV approach. Each camera frame was processed asynchronously to extract faces using preliminary windowing filtering and Viola-Jones classifiers. The application was working with the remote assistance of a RESTful web service.
This short experience taught me how hard funds raising for independent projects can be. We finally decided to stop the development and moved forward to other tech challenges. Even though I failed to successfully finalised that demonstration product, my motivation to work in this field remained unchanged. Since my wife was working in Switzerland, I decided to look for job opportunities there.»
Jun. 2012 - Dec. 2012
This short experience taught me how hard funds raising for independent projects can be. We finally decided to stop the development and moved forward to other tech challenges. Even though I failed to successfully finalised that demonstration product, my motivation to work in this field remained unchanged. Since my wife was working in Switzerland, I decided to look for job opportunities there.»
Real-time audience tracking on embedded camera
SelfAdvert Society (Nantes, France) - Company size: 3 pers.
- Real-time face detection in C++ using Viola-Jones algorithm
- AdaBoost Machine Learning boosted classifiers training
- Background / foreground subtraction in real-time
- Optimisation on embedded Blackfin DSP
- Research-based fast-prototyping
«The project was to develop an embedded camera system for measuring the amount of time shoppers spent in front of advertisements. The main challenge was the integration of advanced mathematical algorithms on a low powered embedded DSP processor. With another engineer, we developed and oversaw the whole creation of this embedded camera system.
My main contributions in the project were the following: First, I integrated the Viola-Jones face detection algorithm based on research papers and OpenCV library implementation. Second, I used state of the art Computer Vision algorithms, such as SIFT, to analyse the scene. The real-time keypoints extraction has been optimised under tight computational costs for extracting foreground objects in real-time. The computation of Image Integrals and face detection classifiers has been significantly sped-up by embedding pre-computed information. Finally, I created specific training tools using AdaBoost machine learning algorithm and iteratively improved the detection rate by reducing the camera's sensitivity to noisy data and outliers. The product has been marketed and used in stores to study shoppers’ behavior near advertisements.
This professional experience taught me how to manage the development of an industrial product and to build platform-optimised algorithms starting from a mathematical concept.»
Jan. 2011 - May 2012
My main contributions in the project were the following: First, I integrated the Viola-Jones face detection algorithm based on research papers and OpenCV library implementation. Second, I used state of the art Computer Vision algorithms, such as SIFT, to analyse the scene. The real-time keypoints extraction has been optimised under tight computational costs for extracting foreground objects in real-time. The computation of Image Integrals and face detection classifiers has been significantly sped-up by embedding pre-computed information. Finally, I created specific training tools using AdaBoost machine learning algorithm and iteratively improved the detection rate by reducing the camera's sensitivity to noisy data and outliers. The product has been marketed and used in stores to study shoppers’ behavior near advertisements.
This professional experience taught me how to manage the development of an industrial product and to build platform-optimised algorithms starting from a mathematical concept.»
Strathclyde University Labs
Contribution to a 2D object recognition algorithmResearch assistant (Glasgow, Scotland)
- Automated selective matching algorithm with SIFT keypoints detectors
- Debugging tools to evaluate and improve keypoints selection for 2D object recognition
University of Strathclyde
Computer EngineeringGlasgow, Scotland
- Specialization in Image Processing, Computer Vision
- Double Master program (final year)
- Master of Engineering in Computer Science - Strathclyde University
- Master of Engineering in Electronics & Computer Science - INSA de Rennes
- Graduated with distinction
LiSSi University Labs
Contribution to a segmentation algorithm for medical imagingResearch assistant (Créteil, France)
- Automated multi-criteria segmentation algorithm with non-linear numerical optimisation techniques (based on Particle Swarm Optimisation)
National Institute of Applied Sciences
Electronics and Computer EngineeringRennes, France
- First year of Master Degree
- Bachelor year
- Started Double Master program with Strathclyde University (Scotland)
- Awarded merit scholarship
Lafayette Preparatory Class CPGE
Specialization in Mathematics & PhysicsChampagne-sur-Seine, France
- Head of the class
- Awarded merit scholarship
Follow
