Driver Assistant Using Deep Learning

Abstract

In the development of intelligent vehicles, accurate traffic sign detection and recognition are critical. This project proposes an improved algorithm for traffic sign detection and recognition, aiming to address limitations of traditional methods such as environmental sensitivity and poor real-time performance of deep learning-based approaches. The algorithm employs HSV-based spatial threshold segmentation for effective traffic sign detection based on shape features. Furthermore, the algorithm enhances the classical convolutional neural network model by utilizing Gabor kernel for initial convolution, incorporating batch normalization after pooling, and employing the Adam optimizer algorithm. The proposed algorithm is evaluated using the German Traffic Sign Recognition Benchmark, achieving a favorable prediction and accurate recognition rate of 99.75%, with an average processing time of 5.4ms per frame. Compared to other algorithms, the proposed approach demonstrates superior accuracy, real-time performance, generalization ability, and training efficiency. The findings of this project are expected to contribute to reducing accident rates and enhancing road traffic safety through improved traffic sign recognition in intelligent vehicles.