A Real-Time Traffic Sign Detection and Recognition System on Hybrid Dataset using CNN

Abstract

We simplify almost everything in our lives today by automating tasks. When driving, we often miss signs on the side of a road because our attention is focused on the road. This is hazardous to us and to people around us. If a driver is alerted without having to shift his/her attention, then this problem can be avoided. Traffic sign detection and recognition (TSDR) plays a major role in this as it detects and recognizes traffic signs, thereby alerting a driver if any traffic signs are approaching. By doing so, besides the safety on road being ensured, the driver will feel more relaxed when navigating tricky or unfamiliar roads. Oftentimes, it is difficult to comprehend signs and other warnings. Using this approach, drivers will not have to struggle with translating signs. This study aims to propose a model for TSDR using deep learning wherein convolutional neural networks are used for recognizing traffic signs using a hybrid dataset that comprises German traffic sign recognition benchmark dataset from Kaggle and a self-created Indian traffic sign dataset. In addition, the proposed model is trained on both datasets individually so that the output from the proposed model can be compared with the output of existing models. Experimental results revealed that the proposed model attained an accuracy of 95.45% for hybrid datasets and for the Indian dataset alone 91.08%, whereas for the German dataset alone, accuracy yielded was 99.85%.

Topics & Concepts