A Heterogeneous Approach for the Traffic Sign Segmentation and Classification
Syed Tahir Hussain Rizvi, Denis Patti, Francesco Savarese, Mohsin Ali Khan · 2021
Deep learning-based image segmentation provides state-of-the-art accuracy and performance for real-time scenarios, but the segmentation of a street is a challenging and complex task. This complexity derives from the huge number of classes that are required to be taken into consideration. In some cases, a vast number of classes can also decrease the classification accuracy of a network. Moreover, the process of image segmentation cannot distinguish between two objects of the same class. In this work, a heterogeneous approach for the traffic sign segmentation and classification is presented. The proposed heterogeneous approach is comprised of an image segmentation block that identifies the different objects on a street especially traffic signs using a deep architecture named ENet (efficient neural network) and the second block distinguishes between segmented traffic signs using Support Vector Machine (SVM) model. The results obtained by the proposed approach are promising.