A New Vehicle Recognition Method by Using BiGaussian Filtering and Bag-of-Feature Algorithms

Abstract

In this work, we focused on object recognition algorithm for object matching and object recognition. We used Scale Invariant Feature Transform (SIFT) algorithm for object and image matching. In this work, we address the problem of vision based on-road vehicle recognition in real-world scenes. Our basic premise is that this problem is too difficult for any type of model or feature alone. Edge is one of the main characteristics of an object, which carries most of the information of an object in an image. Although, edge detection with different types of edge filtering on edge information present in image sequences, has been studied extensively for vehicle detection. However, we concluded that only edge based detection is not sufficient to render high recognition rate. We propose a new method that integrates edge based hypothesis generation and bag-of-feature based hypothesis verification. In hypothesis generation, we use Horizontal Edge Filtering on BiGaussian based edge filtering to filter long horizontal edges. In Hypothesis verification, we use Bag-of-Features with K nearest neighbor’s algorithm for verification of generated hypothesis. Qualitative and Quantitative results on a large data set confirm that our method is able to reliably detect vehicles, even for the vehicle far from camera. Our method is tested on different weather conditions in daytime and it shows recognition rate of 98.5% on average on roads inside a city and on highways.