https://www.linkedin.com/in/nikolicz/

https://zoran.art

PROFESSIONAL SUMMARY

I am an applied computer vision and deep learning systems expert with extensive industry experience turning perception technology into real products, including putting vision into production vehicles. I work at the intersection of software, hardware, mathematical modeling, optics, image processing, optimization, and machine learning to deliver embedded computer vision systems that run in real time under strict power, thermal, and safety constraints.

My key strengths are broad system‑level knowledge across traditional computer vision and modern deep learning, deep understanding of the automotive domain (functional safety, power/thermal challenges, reliability), and the ability to build strong customer relationships and translate their needs into SoC roadmaps and product architectures.

Over my career I have pioneered computer vision algorithms, defined and optimized advanced driver assistance systems (ADAS), specified embedded processor peripherals, and designed embedded computer vision product architectures and the semiconductors that implement them. This gives me a unique perspective and the ability to collaborate effectively with teams across computer vision, biometrics, digital image processing, biomedical engineering, embedded hardware and software, and semiconductor design.

I regularly collaborate with industry and academic partners and am available to advise on AI strategy, perception systems, sensor fusion, and safety‑critical intelligent systems for research institutes and universities.

EDUCATION

  • Ph.D., University of Miami, Coral Gables, FL, Dissertation: An Automatic Induction Of An Expert System For A Finite State Model Of Locomotion
  • M.S., University of Belgrade, Belgrade, Serbia (Major: Electronics)
  • B.S. University of Belgrade, Belgrade, Serbia (Major: Electronics)

TECHNICAL SKILLS & DOMAIN OF EXPERTISE

A very diverse background and a broad skill set across multiple disciplines and multiple technologies.  Deep understanding of a wide range of video and digital image processing techniques related to machine/computer vision, machine learning, knowledge-based systems, object recognition, object tracking, feature registration, structure recovery from stereo or camera motion, gesture recognition and biomedical engineering.

  • Signal processing and instrumentation
    • Expert‑level knowledge of imaging, night vision, radar, lidar, and multi‑sensor fusion systems
    • Experience across optics, CCD/CMOS image sensors, custom ISP pipelines (lens distortion correction, demosaicing, shading, blending), and real‑time computer vision and pattern recognition (face detection, facial key‑points, biometrics).
  • Mathematical methods
    • Strong background in linear algebra, numerical methods, and the impact of finite‑precision effects on algorithms
    • Practical experience with optimization methods (e.g., SLAM, bundle adjustment) and conversion of floating‑point algorithms to fixed‑point arithmetic for embedded deployment..
  • Deep Learning
    • Working with neural networks since the early 1990s; contributed to early applications of deep learning in biometrics and facial recognition in collaboration with Dr. David B. Hertz.
    • Architectures and performance: estimation and optimization of CNN, Swin‑T, and transformer‑based inference across heterogeneous hardware (FPGAs, GPUs, embedded SoCs).
    • Frameworks: PyTorch, TensorFlow, TorchVision, TensorBoard.
    • Radar DL: FFTRadNet and TFFTRadNet (Swin‑T backbone) for automotive radar perception, trained from scratch with mixed‑precision training and INT8 inference.
    • Model optimization: mixed‑precision training, INT8 quantization and calibration, and embedded inference performance tuning for radar and vision pipelines.
  • Programming and tools
    • Languages: C, C++, Python (NumPy/SciPy/Pandas), Matlab, R, PERL, VHDL, Verilog, SystemC, Visual Basic.
    • Libraries and CV stack: OpenCV, CERES, g2o,  plus custom ISP/vision pipelines on embedded platforms.
    • Strong proficiency with standard electrical and electronic CAD/EDA tools
  • Embedded and hardware platforms
    • Verilog/VHDL design on Xilinx FPGAs with multiple shipped projects.
    • .DSP programming and optimization on real‑time embedded platforms, including multi‑tasking RTOS environments.
    • ARM Cortex‑M4, Cortex‑A8, Cortex‑A15 and heterogeneous SoCs (CPU/DSP/FPGA/accelerators)
    • High‑speed microprocessor and DSP board design, EMI/EMC‑aware PCB layout, and EMI testing.
    • Stereo camera and surround‑view systems, including synchronization and transmission of multi‑camera video streams over Ethernet and LVDS
  • Automotive functional safety and regulations
    • Deep knowledge of automotive reliability and quality requirements (AEC‑Q100).
    • Functional safety and process standards: ISO 26262, IEC 61508, MISRA C, ISO/IEC 15504.
    • Automotive vision and ADAS standards and protocols: Euro NCAP, ISO 16505, and related EU and US automated‑driving regulations and legislation.
  • Additional domain exposure
    • Cross‑disciplinary exposure to biomechanics, physiology, rehabilitation engineering, and neurophysiology.
  • Principal areas of interest :
    • Advanced Driver Assistance Systems (ADAS) and Automated Driving
    • Real-Time Computer Vision algorithms,
    • Pattern Recognition and Applied Artificial / Computational Intelligence and Sensor fusion.
    • Medical instrumentation

WORK EXPERIENCE

NVIDIA Corporation.

SENIOR APP MANAGER (2023-Present)

Leading a team of engineers developing and optimizing computer vision, deep learning, radar, and lidar pipelines on the NVIDIA Tegra Programmable Vision Accelerator (PVA), enabling real‑time perception on embedded platforms.

  • Drive adoption of conventional and deep learning–based radar processing within the team, leveraging my system‑level background to define architectures, training workflows, and deployment patterns for radar sensors on heterogeneous SoCs.
  • Guided specification and definition of hardware accelerators for optical flow and stereo vision, including evaluation of classical and deep learning–based optical flow and stereo networks for Tegra integration.
  • Own performance and efficiency analysis of the PVA VLIW DSP architecture and tool-chain, designing benchmarks and micro‑kernels to uncover instruction‑set, memory, and tooling deficiencies and feeding findings back into silicon and compiler roadmaps.
  • Coordinate end‑to‑end optimization across CPU, GPU, DLA, and PVA accelerators, shaping requirements for future hardware and SDK features to support advanced CV and DL workloads

PRINCIPAL ARCHITECT TEGRA SoC (2015-2023)

Working across many NVIDIA teams as a focal point for EU NCAP and for imaging and radar sensors.  Collecting and understanding specific sensor requirements from different NVIDIA groups.  Consolidating the NVIDIA requirements and presenting and driving them externally.  As EU NCAP expert circulated the latest updates of EU NCAP protocols internally across different NVIDIA teams.

  • Person in charge for technical roadmap alignment between NVIDIA's Tegra SoC and imaging and radar sensors.  Working with key imaging and radar sensor manufacturers on the market to prevent technical issues on next generation of devices.  Identifying challenges and driving resolution of the gaps both internally and externally .
  • EU NCAP Expert inside NVIDIA.  Representing NVIDIA in CLEPA and actively participating in Autonomous Driving mirror group and EURO NCAP mirror groups relevant to active safety (LSS, AEB, SAS).  Monitoring automated driving regulations in EU and in the US.
  • Mapped customer use-cases to Xavier and Parker to identify key gaps in the Tegra SoC architecture.  Estimate of chip loading, power dissipation and memory bandwidth.
  • Guided specification and definition of hardware accelerators for optical flow and stereo vision.
  • Strategic analysis of technology trends and analysis of competitive landscape (FPGA, Multi-core CPUs)

TEXAS INSTRUMENTS INC., Houston, TX

CHIEF TECHNOLOGIST FOR  MACHINE VISION AND AUTOMOTIVE VISION (2005-2015)

Set technical direction for the automotive and machine vision inside Texas Instruments. Defined system requirements, chip roadmap and product software roadmap as well as development hardware and tools. Provided both engagement and technical leadership to the customers and to the internal team. This resulted in revenue growth from virtually 0 in 2005 to 15 million of TI’s advanced driver assistance systems (ADAS) System-on-Chip (SoC) devices on the road in 2014.

  • Built the customer base from the ground up securing design-wins at over 10 automotive OEM and tier-one companies such as BMW, Daimler, GM, Continental, VW, Autoliv, Valeo, Hella, TRW to name a few, with applications such as: Pedestrian Detection, Lane Departure Warning, Forward Collision Warning, Automatic Cruise Control, Head Beam Control, Driver monitor, Surround View, Night Vision, Rear View, Surround View and Blind Spot Detect.
  • Lead algorithm development, system partitioning, code optimization and benchmarking of the most performance demanding algorithms including dense stereo vision, traffic sign recognition, pedestrian detection, lane departure warning and optical flow.
  • Leveraged extremely deep computer vision expertise to solve customer problems and enhance TI solutions. Initiated and collaborated with R&D lab on development of Traffic Sign Recognition, Back Over Prevention, Optical Flow and Stereo Vision algorithms and demos
  • Guided specification and definition of hardware accelerators for optical flow and stereo vision.
  • Guided design of complex automotive systems based on Radar, Lidar, Infra-red, Ultrasound and imaging sensors using FPGA, DSP and Micro-controllers.
  • Lead strategic analysis of technology trends and analyzed competitive landscape (FPGA, Multi-core CPUs and GPUs).
  • Facilitated customer code migration from PC/workstation to embedded platforms.
  • Specified and drove development of software libraries optimized to run on TI C6000 processors (Imaging Library, Vision Library, IQMath Lib).

LEAD ENGINEER FOR PROFESSIONAL AUDIO (2002-2005)

  • Invented cutting edge DMA engine geared for professional audio applications by recognizing customer need to overcome inefficiencies of conventional DMA engines for data movement required for audio effects (http://focus.ti.com/lit/ug/spru795d/spru795d.pdf). Lead the team and participated in the implementation and testing of the module.
  • Defined system requirements for software and hardware reference designs for TMS320C672x floating point DSPs and communicated those requirements to integration partners.

SENIOR SYSTEM ARCHITECT (1997-2002)

  • Influenced Texas Instruments Inc. investment into video space by bringing into production a video FPGA daughter card for C6711 DSP (complex FPGA design with DRAM controller interface, and internal cache implementation). Co-invented a Video Port Interface used on various TI DSPs.
  • Designed firmware for multiple FPGA+DSP systems (firmware included VHDL and DSP code). Supported all facets of the customer design flow from presales to production test via presentations, travel, conference calls, architectural design partitioning and schematic and code reviews.

IDENTIFICATION TECHNOLOGIES INTERNATIONAL INC. Coral Gables, FL

CHIEF SCIENTIST (1995 - 1997)

  • Developed key algorithm improvements to the company core product for face recognition. An algorithm was based on extracting the most representative spatial features, and produced as a result a 72-byte long signature for each face.
  • Collaborated with Dr.Tzay Y.Young at University of Miami college of Engineering to incorporate stereo vision into the company product.
  • Responsible for leading software project that took the advanced algorithms from Research and developed it into professional-quality software application.
  • Coordinated the company presentation at 1997 CeBIT show in Hanover, Germany (at the show the sales contract was signed with German Company Leicher).

THE MIAMI PROJECT TO CURE PARALYSIS, School of Medicine, Miami, FL

Research Assistant  (1991 - 1996)

  • An Automatic Induction of an Expert System for a Finite State Model of Locomotion.  Control using a skill-based expert system can be applicable to gait restoration.  Rule-based systems have several advantages for this application: they generate a fast response (they are not computationally intensive) and they are easy to comprehend and implement.  A major problem with using such systems is the inability of users to determine its rules.  I developed an automatic method using artificial neural network for obtaining the production rules from a set of examples.  The production rule is able to estimate muscle activity pattern using sensory information.  The method is based on minimization of entropy.  The rule base was automatically induced using external or afferent sensory signals (input) and EMG patterns (output).
  • Biomechanical Models for Functional Electrical Stimulation of Paralyzed Muscles (project funded by MRC, NCE and the Miami Project to Cure Paralysis). 
  • An electronic circuit for analog processing of neural (Electroneurogram or ENG) and muscular (Electromyogram or EMG) signals in functional electrical stimulation (FES) systems (project funded by National Institute for Health – NIH grant NO1-NS-3-2380).  Signal levels well below 1uV are extracted from extremely noisy environment. This device was tested in chronic recordings using a tri-phasic cuff electrode for nerves and epimysial electrodes for muscles in the hind limbs of cats.
  • An EMG-controlled grasping system for tetraplegics.

© Copyright zoran