Abu Dhabi Emirate, United Arab Emirates
Computer vision reasearcher/engineer with experience in visual odometry, projective geometry, hardware (FPGA), deep learning and machine learning. I am specialized in embedded vision, where the challenge is to adapt/conceive computational and memory efficient 3D CV algorithms for autonomous navigation. Worked in both Academia (4y) and Industry (7y). As part of industrial experience, most challenging was the self-employed activity (2y) where I was responsible for seven persons/colleagues. I had the opportunity to develop a sound work ethic in both environments and very importantly I learned how to conduct and stimulate efficient human communication on demanding projects which require a significant team effort. My roles on these past projects were ranging from pure development, R&D, to team lead, project management and recruitment. My interests lie in deploying computer vision on low-powered (edge) devices. One way of doing so is software-hardware co-design (a.k.a. algorithmic and hardware approaches) - but besides deploying existing solutions I also innovate on state-of-the-art.
Responsible for developing of the visual perception stack for aerial drones in GPS-denied environments: - custom relative localization, use of state-of-the-art visual-inertial solutions - custom global localization (US Patent Application No. 18/766,823), use of state-of-the-art visual- and multi-spectral deep learning solutions Built and led a diverse visual localization team consisting of software engineers and research engineers. Directed R&D efforts with external collaborators from NYU, USA. Interacted with project stakeholders.
Deep learning (CNN) for visual perception and human action recognition in retail: - Human pose estimation and keypoints based tracking (OpenVINO, Hungarian algorithm) - Video Transformer Network architecture fine-tuning and transfer learning with Pytorch Responsibilities: custom dataset and annotation process conception and management, coding of preprocessing and postprocessing logic in Python and interfacing with Amazon AWS
Established and led administration of a Venture Capital start-up. Managed and led deep learning projects for external clients and in-house. Recruited and mentored software engineers for computer vision tasks: - SSD object detection framework for retail semantics, applying transfer learning on Tensorflow models (MobileNetV2, Neural Architecture Search network), data mining and data analysis - machine perception with ORB keypoints detection - internal hardware development of human fall detector using Raspbian and Raspberry Pi sensors coupled with Intel NCS
Improving online localization accuracy of a road user within a dynamic map in the scope of Road Database Client project. Multiple View Reconstruction based numerical optimization on relative position of the in-vehicle camera. Matlab based POCs and evaluation vs an RTK-GPS ground truth. Development in Visual Studio 2010.
Spartan6 (slx25) FPGA based integration of Xilinx's video IPs for image scaling operations. A microblaze-centered design. On-the-fly reconfiguration of the FPGA. Input video format: Aptina 960p25. Output videos (both in parallel on slx25 device): 720p50 and 576i50. Development kit used: Xilinx ISE 14.7. Xilinx video IPs: video I/O to AXI stream, video Scaler, video DMA and video Timing Controller. Reconfiguration: ICAP primitive w/ multiboot Flash. Debug: ChipScope and Oscilloscope.