Haider N.

I build computer vision systems that work beyond the notebook. From YOLO object detection and OpenCV video processing to tracking, GPU optimization, and application integration, I help turn AI ideas and existing code into working software. Specializing in Python, OpenCV, PyTorch, YOLO, deep learning, and video analytics. ◆ WHAT I'VE BUILT — AND WHAT I CAN BUILD FOR YOU 🎯 01 | OBJECT DETECTION, CLASSIFICATION, SEGMENTATION & TRACKING I develop and improve AI models that detect, classify, segment, count, measure, and track objects in images and videos. Delivered work includes: → Custom YOLOv8 signature detection for document images. → Horse-and-rider detection, tracking, and structured 2D trajectories. → Image recognition and feature extraction across approximately 25,000 images. → Sports video analysis, including push-up counting and basketball shot tracking. Tools: YOLO, PyTorch, TensorFlow, OpenCV, ByteTrack/DeepSORT, MediaPipe. 🎬 02 | VIDEO ANALYTICS & INTELLIGENT AUTOMATION I turn footage into useful data, actionable events, and automated workflows. Selected work: → Premiere Pro Wedding Engine: A long-running computer vision and video automation project covering footage analysis, unwanted-clip removal, important-moment detection, editing decisions, synchronization, GPU processing, and automated export. → Retail Behavior Analytics: Person detection and tracking, shelf-zone assignment, automated ROI mapping, and structured behavioral events. → AI Video Understanding: Action recognition, event classification, and video-segment analysis. I can help with video tracking, counting, event detection, analytics pipelines, and AI-assisted video editing. 📐 03 | 3D COMPUTER VISION, GEOMETRY & IMAGE MEASUREMENT My experience includes multi-view 3D reconstruction, triangulation, point-cloud processing, geometric calibration, and vision-based path planning. Relevant projects: → 3D reconstruction and point-cloud workflows at Burqstream Technologies. → Point-cloud segmentation and path planning for an Upwork robotics project. → Multispectral camera calibration and image-processing research, including OpenDroneMap-related work. → Image-based geometric measurement and inspection applications. 📄 04 | OCR, DOCUMENT VISION & MULTIMODAL DATA EXTRACTION I build pipelines that transform visual information into structured, usable data. Examples: → Image-to-knowledge-graph pipeline integrating PDF/image extraction, video frames, metadata, face processing, and Neo4j. → Reading numerical displays and gauge/needle positions from video into time-series data. → Document signature detection, image text extraction, and structured analysis workflows. ⚡ 05 | GPU ACCELERATION, MODEL DEPLOYMENT & SOFTWARE INTEGRATION Training a model is only part of the job. I also make computer vision pipelines faster, reusable, and easier to integrate. Examples: → Optimized a wedding-video processing workflow from approximately 4 hours to 90 minutes using GPU-oriented processing improvements. → Packaged YOLOv8 notebook inference into reusable Python software. → Trained and exported YOLO models to ONNX for application integration, including Unity-related workflows. → Built and debugged OpenCV/C++ applications, Python tools, video-processing scripts, and cross-platform GUI applications. Technologies: Python, C++, CUDA, OpenCV CUDA, FFmpeg, ONNX, FastAPI, and desktop application development. ◆ HOW I CAN HELP Whether you're starting with an idea, a research prototype, an existing codebase, or a working system that needs better accuracy or performance, I can help with design, development, model training, optimization, debugging, and integration. I focus on practical engineering, measurable improvements, transparent communication, and deliverables your team can actually use. → Have a computer vision problem or a pipeline that needs work? Send me a message with the problem and what success looks like. We can discuss the most practical approach.