Description
In summary, this project is trying to read license plate for videos that has taken by personal devices, which involves more complexity due to hardware limitation.
In this work, I first read the video into single frame images. Then I extract the license plate part from the image by detecting high-density vertical edge areas and filtered by color feature and boundary conditions. I run pre-processing to correct the rotation of the image, remove the additional noise and background, and segment the character portion of the license plate. Lastly, I use a template matching on small sets to find best sizing scaling factor and then with all 26 letter, 10 numbers and special characters to find the peaks and filter out the false signal by peak intensity and coordinates.
The software successfully reads videos and displays the license plate location in the original image with the recognition results.
The results processing time is roughly 5 second to read single frame of a video. The successful rate of extraction is low for low-quality (resolution, blur etc.) images, the auto-correction and recognition suffers accordingly. For good quality images with high-resolution, bright illumination condition, the successful rate of the license plate is estimated as 94.5%, will ~3% loss on both preprocessing and ~3% loss on recognition, while solid good on extraction. Multiple frames in a video can be used to improve the reliability and accuracy.
flow rate calculation in the microfluidic channels using Purday approximation
Reviews
There are no reviews yet.