Sergey T.

I build, optimize, and troubleshoot computer vision and real-time video systems—from camera streams and video processing to YOLO object detection and production deployment. My expertise combines 𝗢𝗽𝗲𝗻𝗖𝗩, 𝗬𝗢𝗟𝗢, 𝗙𝗙𝗺𝗽𝗲𝗴, 𝗚𝗦𝘁𝗿𝗲𝗮𝗺𝗲𝗿, 𝗣𝘆𝘁𝗵𝗼𝗻, 𝗮𝗻𝗱 𝗟𝗶𝗻𝘂𝘅. I work across both model development and the surrounding infrastructure, helping clients turn computer vision prototypes into reliable applications. 𝗪𝗵𝗮𝘁 𝗜 𝗰𝗮𝗻 𝗵𝗲𝗹𝗽 𝘆𝗼𝘂 𝘄𝗶𝘁𝗵: • 𝗖𝗼𝗺𝗽𝘂𝘁𝗲𝗿 𝘃𝗶𝘀𝗶𝗼𝗻: YOLOv5/v8 training and fine-tuning, object detection, dataset preparation, model evaluation, and OpenCV processing. • 𝗥𝗲𝗮𝗹-𝘁𝗶𝗺𝗲 𝘃𝗶𝗱𝗲𝗼: FFmpeg, GStreamer, RTSP, low-latency streaming, video pipeline troubleshooting, and performance optimization. • 𝗜𝗻𝗳𝗲𝗿𝗲𝗻𝗰𝗲 𝗮𝗻𝗱 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁: ONNX Runtime, Python APIs, Docker, Raspberry Pi, embedded Linux, and cloud services. • 𝗣𝗿𝗼𝗱𝘂𝗰𝘁𝗶𝗼𝗻 𝘁𝗿𝗼𝘂𝗯𝗹𝗲𝘀𝗵𝗼𝗼𝘁𝗶𝗻𝗴: Diagnosing bottlenecks, dropped frames, resource consumption, networking issues, and unreliable application behavior. 𝗥𝗲𝗹𝗲𝘃𝗮𝗻𝘁 𝗲𝘅𝗽𝗲𝗿𝗶𝗲𝗻𝗰𝗲 My completed projects include automatic license plate recognition (ALPR) model training and fine-tuning, serverless object detection using YOLOv8 and Azure Functions, Raspberry Pi camera systems, and low-latency video streaming. I also bring more than 20 years of Linux experience, including Python development, deployment automation, networking, and production infrastructure. 𝗛𝗼𝘄 𝗜 𝘄𝗼𝗿𝗸 I start by understanding the requirements and establishing a measurable baseline. I then identify the bottlenecks or failure points, agree on a focused scope, and verify the results. Depending on the project, that means measuring detection accuracy, precision and recall, FPS, inference latency, resource usage, or system stability. If you have a computer vision model that works but is difficult to deploy, a video processing pipeline that is unreliable, or a camera-based application to develop, I can help you identify a practical solution.