Senior Computer Vision & Perception Engineer

Email · LinkedIn · ICPR 2024 Paper

I build real-time perception systems that ship on edge hardware — detection, multi-object tracking, SLAM, and multi-sensor fusion. My work sits at the intersection of applied research and production engineering: taking state-of-the-art algorithms, adapting them to constraints they were never tested against, and making them run fast enough to be useful.

• Detection & tracking. YOLOv8 / v11, RT-DETR, SAM, DeepSORT, ByteTrack. I modified DeepSORT's track-management logic for my ICPR 2024 paper and reduced identity switches by 70% in the target application.
• 3D vision & state estimation. Visual SLAM (2D/3D), structure-from-motion, camera calibration, Kalman / Extended Kalman filtering, Hungarian assignment.
• Multi-sensor perception. Camera + radar fusion, sensor-driven target prioritisation, closed-loop perception-control pipelines.
• Edge deployment. NVIDIA Jetson (Orin / Xavier / Nano), TensorRT, ONNX Runtime, FP16 / INT8 quantisation. 71% latency reduction on production workloads, sub-100ms perception loops.
• Generative vision. LoRA, IP-Adapter, character-consistent image generation for synthetic dataset generation in data-scarce scenarios.

Akpu, C. H., Wei, H., & Hong, X. (2024). Deep learning in automated worm identification and tracking for C. elegans mating behaviour analysis. International Conference on Pattern Recognition (ICPR 2024), Springer Nature Switzerland.
→ First-author · Novel YOLOv8 + SAM + custom-modified DeepSORT pipeline · Open-source implementation ↗

MSc Data Science, University of Reading (Distinction) · 6+ years shipping CV systems across healthcare, quest platforms, autonomous systems, and behavioural research.

Languages Python · C / C++ · SQL
CV / ML PyTorch · TensorFlow · OpenCV · NumPy · SciPy
Edge NVIDIA Jetson · TensorRT · ONNX Runtime · CUDA · RKNN
Infra Docker · FastAPI · Apache Airflow · PySpark · GCP · PostgreSQL · Linux

_{Reach me at holaryc@gmail.com · LinkedIn}