Road sign recognitionon IoT rover
Curated and augmented 4,000+ image dataset to train a TensorFlow classifier. Optimized inference pipeline for 70% latency reduction, enabling real-time processing on a live camera feed mounted to a Raspberry Pi rover.
Dataset size
4,000+ images
Latency reduction
70%
Real-time processing
enabled on Pi
Team size
4 members led
Data-centric ML
Rigorous data curation and augmentation created a high-quality training set. Inference optimization focused on edge-device constraints, trading off a small amount of accuracy for 70% speed improvement.