Traffic Sign Detection and Recognition using CNN and KERAS

A traffic sign detection and recognition system based on CNN with high-accuracy and possibility of real-time implementation for autonomous driving applications has been presented. The Keras framework was used to develop the system, which overcomes limitations of traditional traffic sign recognition methods, commonly found to fail due to environmental variations such as changing light, object occlusions, or damaged traffic signs. It takes advantage of the comprehensive GTSRB dataset and performs data augmentations so as to generalize well to different driving environments. The optimized convolutional and pooling layers in the CNN architecture allow for efficient extraction of salient features, which is crucial as ASL often has similar sign classes that need to be differentiated across variations in scale, illumination, and orientation. The system ensures sign recognition instantaneously and with accurate classification performance, improving safety and efficiency of autonomous driving by providing a fast-accurate real-time recognition of a sign.