Pipeline Architecture for High Speed License Plate Character Recognition - Advances in Image and Graphics Technologies - page 74

Pipeline Architecture for High Speed License Plate
Character Recognition

Boyu Gu 1 , Qiang Zhang 2 , and Zhenhuan Zhao 2

1 Changchun University of Science and Technology
No.7089, Weixing Road, Changchun, 130022, China
[e-mail]
2 Continental Automotive Corporation (Lian Yun Gang) Co. Ltd. Changchun Branch
No.1981, Wuhan Road, Changchun, 130000, China

Abstract. An embedded hardware for license plate character recognition is
designed and implemented on an FPGA (field programmable gate array) with
pipeline architecture. The architecture is based on M2DPCA (modular
two-dimensional principal component analysis) algorithm. Three processing
elements are contained in the proposed pipeline architecture, projection element
is designed for matrix multiplication operations of feature extraction, the
distances between input character and each class in training database are
computed in distance element, and the nearest neighbor classification is carried
out in classification element, all functions are run in pipeline. Experimental
results show that very high speed is achieved, which provides approximately
28% speedup of equivalent software implementation, and also, the hardware
architecture performs extremely resource economical.

Keywords: License plate recognition, character recognition, FPGA, pipeline
processing, hardware architecture.

1 Introduction

Automatic license plate recognition technology has numerous important applications in
people's daily life[1,2]. In a license plate recognition system, very little time and
resource is allowed to be consumed by character recognition. One main difficulty of
license plate character recognition is that implementations on embedded application
should operate fast enough to ensure the whole system to fulfill the real-time
requirement[3], and the other obstacle is make the resource occupation of character
recognition functions as little as possible.
To efficaciously extract the feature in a high dimensional space is extremely crucial
for character recognition. Since high dimensional image data could projected into low
dimensional eigen space, the PCA based approach is very suitable for embedded
applications. As a statistical approach, PCA was proposed in [4], and has been widely
applied to pattern recognition[5,6]. To improve the accuracy to varying illumination
and angle, modular PCA was proposed in [7], by divide the images into sub-blocks, the
technique has been achieve a better recognition rate. 2DPCA was proposed in [8],