The study of deep learning for automotive logo recognition and classification

Abstract

Most vehicle manufacturer recognition (VMR) techniques are established in vehicle logo recognition because a vehicle's logo is the most obvious sign from the vehicle's manufacturer. However, due to the difficulty in accurately segmenting a vehicle logo on the picture with demand in resilience against many imaging scenarios, logo recognition can still be challenging. After subjective overview about this scope, a convolutional neural network (CNN) method for VMR is investigated in this research, which does away with the need for exact logo detection and segmentation. A powerful pertaining approach has also been developed to improve real-world applications to lower the high computational cost for kernel training on CNN-based systems. The contribution of this paper is to study the multiclass logo employing random forest ensemble learning and convolution mapping in nonlinear space. To boost accuracy by roughly 35%, 800 images from 15 types of car classes were investigated in the paper.