Computer Vision
German Traffic Sign Detection and Classification for E-Bike Safety
Computer-vision system for detecting and classifying German traffic signs relevant to e-bike scenarios. Implemented and custom-trained an SSD MobileNetV2 model and compared its performance with YOLOv11n.
Overview
This completed academic project focuses on detecting and classifying German traffic signs relevant to e-bike safety. The implementation uses a custom-trained SSD MobileNetV2 model and compares its predictions with YOLOv11n.
Project workflow
1
Dataset Preparation
2
Annotation and Preprocessing
3
SSD MobileNetV2 Training
4
SSD and YOLOv11n Comparison
What I built
- Annotated German traffic-sign scenes and prepared them for training.
- Prepared bounding boxes, resizing, normalization, augmentation, Pascal VOC XML, and TFRecord conversion.
- Implemented and custom-trained an SSD MobileNetV2 320×320 FPN-Lite model.
- Compared predictions with YOLOv11n on shared test images.
- Reviewed confidence scores, confusion matrices, and class-level error analysis.
Technical snapshot
Domain
Computer Vision
Use case
E-bike safety
Task
German traffic-sign detection and classification
Primary implementation
SSD MobileNetV2 320×320 FPN-Lite
Comparison model
YOLOv11n
Classes
20 German traffic-sign classes
Framework
TensorFlow