Vehicle Recognition using extensions of Pattern Descriptors

Abstract

Vehicle identification and classification for still images are incredibly useful and can be extended to a range of traffic surveillance operations. Reliable and accurate recognition of vehicles is however a challenging issue due to changes in vehicle appearance and illumination difference in real time scene. In this paper, we present a simple and effective way of vehicle recognition technique based on vehicle’s local texture features extraction and classification. The local features are extracted individually using the Local Binary Pattern (LBP), Median Binary Pattern (MBP), Gradient directional pattern (GDP), and Local Arc Pattern (LAP) descriptors and feed into Support Vector Machine (SVM) for classification. We also focus on vehicle classification using various color spaces like RGB, HSV, YCbCr for the texture descriptors extraction. The primary focus is to observe the effect of colour information on vehicle classification efficiency across different colour spaces. Initially, experiments are conducted for the classification of gray-level vehicle images of five different classes from the CompCars dataset. Then experiments are extended to different color spaces for the same dataset for color texture classification. The integration of different colour details increases the efficiency of vehicle classification, according to the experimental results.