3rd International Engineering Conference on Developments in Civil & Computer Engineering Applications (IEC2017)
Title: Multi-class vehicle type’s recognition system using spatial visual words with minimized feature set
Authors: Abbas M. Ali, Moayad Y. Potrus, Amin S. Mohammad
DOI: 10.23918/iec2017.12
Abstract: Visual vehicle detection and recognition are very important issues in road development study. It gives insightful statistics for the purpose of building-up road infrastructure, road maintenance and future development of road capacity. It is very important to keep stability of road traffic and safety with comfortable travel. In the literature, many approaches have been introduced to improve the multi class vehicle recognition. Through computer vision techniques, the paper introduces a method that uses the correlation among patches to measure the similarity between images. According to this approach, a set of HOG and SURF local features is extracted from a given automobile image. Then, the distance between these local features is computed against the visual codebook, which was previously constructed by K-means. The analysis applied to the Naively combined features result in getting better computational performance in recognition. KNN, and SVM, were used for data analysis and the results indicate that the proposed method gives promising results on the UIUC car detection dataset. Moreover, a combinatorial gray wolf optimization method was employed to determine the best recognition with less local feature set.
Keywords: Vehicles recognition, Local feature extraction, SURF, RBF, SVM.