| Summary: | The aim of this project is to develop motorcycle plate image detection and
recognition framework for traffic offender using Field Programmable Gate Array
(FPGA). The proposed system has processing time of 33.3 milliseconds in
various critical conditions, such as daylight, rainy daylight and night. Currently,
the available technology is lacking due to the system implementation is not
robust and less efficient. Benchmarking study for fast processing system
showed FPGA can carry out real-time processing at 128 × 128 resolution video
sequences at 30 frames per second (fps). Therefore FPGA was selected to
improve the plate number recognition for motorcycle. Comparison between
hardware (FPGA) and software (MATLAB) implementation of edge detection
was also performed. Currently, the time for processing motorcycle plate image
using software is 52 milliseconds. To meet the processing time constraints for
the developed framework, it is important to quantify the reduction of processing
time that can be achieved if the framework component are embedded into
hardware-based platform such as FPGA. MATLAB-Simulink was selected for
designing the detection system. This system was designed to detect static
images and moving objects in critical condition from 5 to 15 meters distances.
Then, the detection system was implemented on the FPGA for the detection
and recognition process. The output image was analyzed by comparing the
accuracy of bounding box and edges displayed in different conditions,
threshold levels, resolutions and distances.
From this proposed system, the daylight condition at 5 meter distance gives the
highest accuracy of 99% at threshold level ranging from 2 – 10. In addition, the
highest resolutions accuracy is 99% at 1024 ×768 pixels. This comparable with
the output from Matlab Simulink system that shows the best accuracy of 99%
at threshold level 2 and resolution pixel 1024 ×768 pixels. The second best
conditions is rainy daylight, in which the threshold accuracy is 99% at level 10
with 99% resolution accuracy at 1024 ×768 pixels. From the analysis, it can be
concluded that daylight is the best condition in detecting the motorcycle
images, followed by rainy daylight and night conditions. The speed to process
the image is 30 frames per second, which is 58 % faster than images
processed by Matlab Simulink. Thus, this proposed system on FPGA is more
flexible, efficient and robust for real time plate recognition for motorcycle. This
study can be implemented to efficiently identify road traffic offenders and
improve visual driver support system in the future.
|
|---|