License plate detection in differently illuminated car images

• Main Author: Tan, Evelyn Gek Khim.
• Other Authors: Sung, Eric
• Format: Final Year Project
• Language: English
• Published: 2009
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Electronic systems::Signal processing
• Online Access: http://hdl.handle.net/10356/17956
• Institution: Nanyang Technological University

License Plate Detection (LPD) is the pivotal process in all Vehicle License Plate Recognition (LPR) system, in which the character recognition rate is greatly affected by License Plate (LP) detection rate of success and correctness. Although many research and development on License Plate Detection were conducted, however, most of the techniques have to be operated under many restricted conditions such as with fixed illumination, fixed background, stationary vehicle, designated range of distance between camera and vehicle [7]. In this final year project report, both linear and non-linear filtering methods are explored and implemented for detecting the presence of license plate regions of images taken in open roads. The license plate candidate selection is based on license plate features such as its shape characteristic, dimension and area. The proposed LPD technique consists of 3 key modules: convolution-based edge detection, detection for possible LP regions and LP candidate selection. Firstly, a convolution-based edge detector employs linear filtering method and is characterized by vertical and horizontal gradient derivatives to extract the edges from a cluttered image. Next, hysteresis method and dilation are employed to detect for connected components or possible license plate regions from the binary image. Lastly, license plate selections is based on pre-defined License Plate features like the background and characters color, LP range of dimension, intensity and area. To ascertain the detection performance accurately, these feature extraction techniques are implemented in Visual C++ and tested by using a dataset of 120 color car images (768 by 576 pixels) which include 4 common types of license plates found on Singapore roads. Two experiments have been conducted for the proposed technique, using 120 color car images taken from various angle, dynamic conditions and in different illumination of the day. The experimental results show that the proposed technique can locate license plate candidates quickly and correctly, whereby the average detection rate of success and average detection accuracy rate for experiment 1 are 88.4% and 87.8% and for experiment 2 are 88.9% and 92% respectively. The experimental results obtained by the proposed algorithm exhibit encouraging performance. A dialog-based User-Interface (UI) application was created for easy execution of the LPD program and reviewing of processed images in the experiments. Keywords: License Plate Detection (LPD), Vehicle License Plate Recognition (LPR) system, Vertical and Horizontal Gradient Detection, edge detection algorithm, dilation, hysteresis, license plate candidates, edge detector