Muhammad Asif Manzoor
Published: 2018
Total Pages: 0
Get eBook
Automatic License Plate Recognition (LPR) is a basic computer vision problem. LPR has been thoroughly explored and is widely considered to be a well-understood problem. LPR systems are widely deployed all across the globe for a variety of situations. LPR systems perform poorly in the situations where the license plate is ambiguous, forged, or damaged. Similarly, LPR systems cannot work if only partial license plate information is available or if only vehicle's description is available; hit and run, hot pursuit, and amber alert are a few of the situations where a vehicle's description is often available. A Vehicle Make and Model Recognition (VMMR) system can provide great value in terms of vehicle monitoring and identification based on vehicle appearance instead of the vehicles' attached license plate. A VMMR system and an LPR system can be used to complement each other. A real-time VMMR system is an important component of many Intelligent Transportation System (ITS) applications, such as automatic vehicle surveillance, traffic management, driver assistance systems, traffic behavior analysis, and traffic monitoring, etc. The VMMR system can reduce the cost for such applications. VMMR systems have a unique set of challenges and issues. A few of the challenges are related to computer vision includes image acquisition, variations in lighting and illuminations, variations in weather, occlusion, shadows, and reflections, etc. A few of the challenges are due to the nature of the problem, such as the large variety of vehicles, the inter-class and intra-class similarities, addition/deletion of vehicles models over time, etc. The VMMR system is a multi-class classification/recognition problem; thus the selection of machine learning algorithm for robust and reliable VMMR system is another challenging task. In this thesis, we present a unique and robust real-time VMMR system which can handle the challenges described above and recognize vehicles with high accuracy. We extract image features from vehicle images and create many different feature vectors to represent the dataset. We use two existing classification algorithms, Random Forest and Support Vector Machine, in our work. We also proposed a two-level Support Vector Machine classification algorithm in our work. We use a realistic dataset to test and evaluate the proposed VMMR system. The vehicles' images in the dataset reflect realworld situations such as different weather conditions, different lighting exposure, occluded images (e.g. pedestrians), and different viewing angles, etc. The proposed VMMR system recognizes vehicles on the basis of make, model, and generation (group of consecutive manufacturing years) while the existing VMMR systems can only identify the make and model. Image feature descriptors are used to represent the vehicle dataset. Image features can be broadly divided into two categories: the first type of features is extracted using the prominent and distinctive points/patches in the image and uses the selected points/patches to represent the image while the second type of features uses the entire image for feature extraction and representation. We create optimized visual dictionaries to encode images for the first type of features. We also compare the proposed system with the existing VMMR research. The underlying goal of the proposed VMMR system is centered on discovering the ability of supervised learning to resolve the computer vision problem that results from the stringent limitations of the problem environment. The proposed real-time VMMR system can produce valuable information for law enforcement agencies.