Computer Vision Research Group

Vehicle Make and Model Recognition under controlled conditions

Automatic vehicle type recognition (make and model) is very useful in secure access and traffic monitoring applications. It complements the number plate recognition systems by providing a higher level of security against fraudulent use of number plates in traffic crimes. We developed a simple but powerful probabilistic framework for vehicle type recognition that requires just a single representative car image in the database to recognize any incoming test image exhibiting strong appearance variations, as expected in outdoor image capture e.g. illumination, scale etc.

We used a new feature description, local energy based shape histogram 'LESH', in this problem that encodes the underlying shape and is invariant to illumination and other appearance variations such as scale, perspective distortions and color. Our method achieves high accuracy (above 94 %) as compared to the state of the art previous approaches on a standard benchmark car dataset. It provides a posterior over possible vehicle type matches which is especially attractive and very useful in practical traffic monitoring and/or surveillance video search (for a specific vehicle type) applications.

Our paper titled "Bayesian Prior Models for Car Make and Model Recognition" received best paper award at International Conference on Frontiers of Information Technology (FIT 2009).(see publication)

Vehicle Make and Model Recognition under Un-controlled conditions

The different approaches addressing vehicle MMR problem achieve high recognition rates on datasets of car images collected under constrained environments where the images are taken from similar view point, at same scale with minimum camera movement and contain a properly segmented car without cluttered background. Cars should be recognized even if they appear at different scales and viewpoints in highly cluttered backgrounds and so it remains quite challenging for an approach to achieve acceptable results based on minimum set of constraints (on background clutter, viewpoint and scale).

We are currently developing an approach that can recognize make and model of car in presence of scale variations, viewpoint changes and background clutter by automatically learning discriminative patches/regions for each vehicle class. To achieve this, we propose a novel patch selection criterion that allows automatic selection of discriminative patches for each vehicle class. Feature extraction is achieved through local description of interesting regions. Query patches of unknown query image are assigned class labels in a pure Bayesian setting.

In pursuit of a less constrained vehicle recognition approach, we have also introduced our own complex dataset of cars COMVis_cardataset_v1 prepared under uncontrolled environment with high in-class appearance variability and cluttered backgrounds. We are in process of improving the classification results of this proposed approach on COMVis_cardataset_v1.

In future, more variants of how to classify unknown query image once discriminative patches have been learnt for each vehicle class will be developed to improve recognition accuracy. Different interest point detectors and feature description methods will be tried to see variation in performance. Proposed approach will be tested on more datasets to check its robustness.

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