Browse categories
Explore
Fiverr Pro
English
$
USD
I will do object detection, tracking in yolov8
Simulation and computer vision engineer
I build simulators and computer vision pipelines.
I have delivered a licensed simulator to a paying client, shipped as a packaged installer rather than a demo. My own work includes a lattice-Boltzman...
About this Gig
Object detection, tracking and counting in images, video and live streams.
WHAT I BUILD
- Detection tuned to your objects, including open-vocabulary detection with no training set
- Multi-object tracking that holds IDs across frames, with camera-motion compensation
- Counting, direction and dwell-time analysis
- Frame triage for long video: pulling the few frames that matter out of hours of footage
- Preprocessing, annotation and dataset preparation
WHERE THAT COMES FROM
I built a multi-signal keyframe selection framework for live video that combines ORB structural change detection with YOLOv8 and BoT-SORT tracking under an online adaptive threshold. It is validated on a 148-clip UCF-Crime benchmark with ablation and sensitivity studies, runs on a dual-GPU pipeline, and is written up in IEEE format.
STACK
YOLOv8, YOLO-World, ByteTrack, BoT-SORT, ORB, OpenCV, PyTorch, Python. Vision-language models such as Qwen2.5-VL where detection alone is not enough.
Send a sample clip or a handful of images and tell me what you need detected. I will tell you what accuracy is realistic before you order.
My Portfolio
FAQ
What do you need from me to start?
A representative sample: a short clip or 20-30 images from the same camera and conditions as the real data, plus a plain description of what counts as a detection. Sample data matters more than volume. I will tell you what is achievable before you order.
Do I need a labelled dataset?
Not always. Open-vocabulary detection can find objects from a text description with no training data, which is often enough for common objects. For unusual objects or high accuracy targets, a labelled set helps, and I can handle annotation as part of the work.
What accuracy can you promise?
None, before seeing your data, and be careful with anyone who does. Accuracy depends on object size, occlusion, lighting and camera motion. Send a sample and I will give you an honest estimate and tell you if the footage is the real problem.
Can you track objects, not just detect them?
Yes. Tracking holds an identity across frames so you can count, measure dwell time or follow direction. I use BoT-SORT and ByteTrack, including camera-motion compensation for footage from a moving or shaking camera.
