I will build custom yolo object detection and tracking for images and video
Python Data Automation and Computer Vision Engineer
Level 1
Has met certain performance criteria and shows strong potential in the marketplace.
About this Gig
Your camera already captures the scene. The missing piece is a system that can tell you what appeared, where it moved and how many times it crossed a line.
I build custom YOLO object-detection and tracking systems for images, recorded video and live camera feeds. The solution is shaped around your real footage, target objects, lighting, camera angle and hardwarenot a generic demonstration trained on unrelated data.
I can deliver:
- Custom YOLO training and fine-tuning
- Multi-object tracking with stable IDs
- Counting, line crossing, zones and event rules
- Image, video, webcam or RTSP inference
- Validation metrics and annotated test results
- Python source code, trained weights and setup guidance
- FastAPI, Docker, ONNX or TensorRT deployment options
Send the target objects, sample footage, dataset status and deployment environment. I will assess what is realistic, define the right first version and build a system you can test on real data.
No inflated accuracy promises. No notebook-only delivery. Just a clear, tested vision pipeline built for the job.
APIs:
Amazon Rekognition
•
Other
Programming language:
Python
Tools:
Jupyter Notebook
•
OpenCV
•
TensorFlow
•
MLflow
Frameworks:
Scikit-learn
•
Keras
•
PyTorch
•
Other
My Portfolio
FAQ
Do I need to provide a dataset?
For custom objects, you normally need labeled images representing the real environment. I will review the dataset before training and explain whether its size, labels and variety are suitable.
Can you use a pretrained YOLO model?
Yes, when the required objects already exist in a supported pretrained dataset. For specialized objects, fine-tuning on your own labeled images usually produces a more relevant solution.
Can the system track and count objects?
Yes. I can assign persistent IDs, count objects, detect line crossings and monitor agreed regions inside recorded video or live feeds.
Will it work in real time?
That depends on model size, video resolution, camera count and target hardware. I will assess the expected environment before promising a real-time speed.
Will I receive the source code and trained weights?
Yes. Every package includes the agreed Python source code, trained model weights and instructions for running the delivered system.
Can you deploy it to a server or edge device?
Yes. Deployment to an agreed cloud server, NVIDIA Jetson, supported local machine or another suitable environment is available in Premium or as an extra.
Can you annotate my dataset?
Small corrections can be discussed, but full annotation is handled through the separate image and video annotation Gig. This keeps the scope and quality expectations clear.
Can you guarantee a specific accuracy?
No serious engineer can guarantee accuracy before reviewing the data and operating conditions. I provide validation results, explain limitations and show how the system performs on held-out examples.
Can you identify specific people?
This Gig focuses on object classes, movement and counting—not identifying individuals or building unauthorized biometric surveillance systems.

