Towards improving the performance of traffic sign recognition using support vector machine based deep learning model

Nowadays autonomous vehicles are evolving due to the advancements in cutting edge technologies. In order to recognize the traffic signatures with high efficacy, traffic recognition system is required. Sign detection and classification are the two parts of the recognition system. The sign detection algorithm detects the size and coordinates of the sign board in an image and in sign classification, the representation of traffic signal is identified and classified into one of their traffic sign sub-classes. In order to achieve these goals, an extremely fast detection module using support vector machine is proposed to detect the traffic sign into one of the traffic classes such as, prohibitory, danger, mandatory, and non-sign. Further classification is carried out using deep convolutional neural networks to determine the sub-classes of each super-class, such as, prohibitory, danger, and mandatory. Based on publicly available benchmark traffic sign image datasets, we have demonstrated that the proposed approach has significantly improved traffic sign recognition accuracy compared with state-of-the-art systems.

Automated summary

Nowadays, autonomous vehicles are evolving with the advancements in cutting edge technologies. Traffic recognition system is required to efficiently recognize traffic signals. It consists of sign detection and classification. The proposed approach utilizes a support vector machine based fast detection module to detect traffic signs into different traffic classes, and deep convolutional neural networks for further classification into subclasses. Experimental results on benchmark traffic sign image datasets demonstrate that the proposed approach significantly improves traffic sign recognition accuracy compared to state-of-the-art systems.