Using a public surveillance camera with a height of 3 to 4 meters and a resolution of 2MP / 4MP, real-time authentication of a vehicle license plate is problematic. To overcome the problem of existing systems, the proposed solution enhances the system's accuracy by combining the features of TensorFlow, OpenCV, and Keras-OCR.
- Capture video of the vehicle with the surveillance camera (to the given threshold).
- Identify the vehicle by comparing the trained datasets using TensorFlow. Capture the image with the detected vehicle.
- The image is captured and pre-processed to overcome noise, dim light, shadow, cross angle, etc. ... and crop the preprocessed image using OpenCV.
- Separate the letters from the number plate and separate the letters based on the number plate format using KerasOCR.
- Detected number plates can be managed by NoSQL MongoDB, computing data using the Hadoop service, and storing data in the cloud.