SSD Based Vehicle Number Plate Detection and Recognition

Vehicle Number Plate (VNP) technology is used in applications like parking management, traffic control and management. It involves tasks such as detecting and recognising vehicle licence plates. Most of the VNP systems in use today don't perform well enough in real-time image/video scenarios. In this work, a deep learning-based method for locating the licence plates of four-wheeler vehicles is proposed and put into practice, (OCR) Optical Character Recognition to instantly identify the characters and numbers on the extracted plates. The retrained Single Shot Detector (SSD) model of detection and recognition is stored in the (ONNX) Open Neural Network Exchange format. To trim and save the localised plate, NVIDIA Compute Unified Device Architecture (CUDA) APIs are employed, which resulted in speeding up the processing. The performance of three different optical character recognition (OCR) techniques─Easy OCR, Tesseract OCR, and ONNX OCR model─is compared to choose the best one. In terms of accuracy and time complexity, the Easy OCR model is proven to be the best. The author suggests using Single Shot Detector (SSD) Mobilenet V3-based architecture to locate licence plates and Easy OCR due to its superiority for character recognition. In terms of accuracy and time complexity, the Easy OCR model is proven to be the best. Finally, the Jetson Nano board is equipped with all of these blocks. These adjustments and implementations have produced frame rates of around 43 and 30 for detection and recognition respectively. Images and Videos can be processed for VNP with more than 97% accuracy using the suggested and implemented system on the Jetson Nano.