Computer Vision Research Group


Vehicle make and model recognition system is now an integral part of public access control systems installed at parking lots, buildings and streets for implementing security measures. They also enhance the capability of automatic number plate recognition systems for monitoring traffic on highways. To recognize the make and model of car in real environments the proposed approach should be able to demonstrate its classification results on a dataset of car images with minimum controlled conditions and contain a large number of vehicle types.

Most of the publically available car datasets constitute of car images captured under controlled setups i.e. same scale, similar viewpoint, uniform illumination and contain properly segmented car. Existing literature relating to vehicle make and model recognition have used these datasets emphasizing very constrained imaging conditions to show their classification results.

There was an emerging need to collect the dataset of car images which should simulate less constrained imaging conditions and consist of more number of vehicle types. This COMVis_cardataset_v1 was collected over a period of three months and is prepared under uncontrolled environment presenting high in-class appearance variability. It consists of 38 different vehicle types, each having at least 15 frontal and 15 rear view car images where each type is placed in a specific folder. Car images in this dataset present scale, rotation and viewpoint variations, complex illumination and highly cluttered backgrounds. The main focus of this COMVis_cardataset_v1 is to testify the effectiveness of the approaches in realistic conditions developed for car make and model recognition.

Some sample images, each from different class type are given below


Download read_dirr.m

Recent News
© 2010 Department of Electrical Engineering, CIIT Lahore  |  Terms and Conditions