Advanced Driver-Assistance System with Traffic Sign Recognition for Safe and Efficient Driving

Abstract

Advanced Driver-Assistance Systems (ADAS) coupled with traffic sign recognition could lead to safer driving environments. This study presents a sophisticated, yet robust and accurate traffic sign detection system using computer vision and ML, for ADAS. Unavailability of large local traffic sign datasets and the unbalances of traffic sign distribution are the key bottlenecks of this research. Hence, we choose to work with support vector machines (SVM) with a custom-built unbalance dataset, to build a lightweight model with excellent classification accuracy. The SVM model delivered optimum performance with the radial basis kernel, C=10, and gamma=0.0001. In the proposed method, same priority was given to processing time (testing time) and accuracy, as traffic sign identification is time critical. The final accuracy obtained was 87% (with confidence interval 84%-90%) with a processing time of 0.64s (with confidence interval of 0.57s-0.67s) for correct detection at testing, which emphasizes the effectiveness of the proposed method.