
YOLO Currency Detector
Real-Time Malaysian Ringgit Denomination Recognition
“A model is only as good as the data you give it, and building that data is half the project.”
Overview
The YOLO Currency Detector is a real-time object detection system trained to recognise and classify Malaysian Ringgit banknotes by denomination. The system processes a live camera feed and overlays bounding boxes with confidence scores on each detected note, identifying RM1, RM5, RM10, RM20, RM50, and RM100 notes in any combination.
Dataset Construction
No public dataset existed for Malaysian Ringgit denomination detection, so the entire dataset was built from scratch. Notes were photographed under different lighting conditions (natural light, artificial light, overexposed, underexposed), at various angles, with partial occlusions, and against diverse backgrounds. Each image was manually labelled using annotation software, producing bounding boxes and class labels for every visible note. Data augmentation (rotation, scaling, brightness jitter, noise) was applied to increase effective dataset size.
Model Training
The detector was built on YOLO, fine-tuned on the custom dataset using transfer learning from a pre-trained checkpoint. Training involved iterative hyperparameter refinement: adjusting anchor sizes, learning rate schedules, and augmentation intensities to maximise detection recall while minimising false positives. The final model runs at real-time frame rates on a standard laptop CPU, sufficient for live deployment.
Project Gallery
10 items




Explore More Projects
Swipe or click arrows to see other work
















