Automatic vehicle Bangla license plate detection and recognition

Automated license plate recognition is important in many contexts like- security and law enforcement, monitoring vehicles, automated parking control, etc. To enable these automated services in Bangladesh, we are reviewing and combining several established methods in this paper. The task of recognizing vehicle number plates consists of three-stage: plate detection, plate extraction, and character recognition. Each stage has many sub-steps. For every substep, we have reviewed many methods, and chosen the one that proved to be the best solution after thorough testing and observation. The main objective of this research is to gain high accuracy using as little CPU Time as possible, keeping into consideration the facts like- lighting conditions, vehicle motion, noisy plates, segmented words in the input image. The primary target of this thesis is to extract a clean image of license plates of private or community vehicles. Although we target our system to be able to detect standard license plates, we also tested our methods on non-standard plates. We have gained an overall accuracy of 98.41% in plate detection, 82.25% in plate extraction, and 94.11% in character segmentation, resulting in 76.19% overall accuracy using the available dataset. In terms of efficiency, the runtime of our plate detection module is 0.841 seconds, plate extraction is 0.132 seconds, character segmentation required 0.009 seconds, and the average total runtime of the entire system is 0.982 seconds; in our 64-bit machine having 4.00GB main memory and Intel(R) Core (TM) i3-4005U CPU @ 2 × 1.70. In this paper, our aim is to process the input image up to Character Segmentation so that it is easy to recognize the character by sending it to any existing Optical Character Recognition (OCR) system