AI based Smart Parking System

A new venture has been endorsed to thrive a user-friendly smart parking system with minimal human interference. The proposed system scans for the vehicle type and vehicle number when a new vehicle enters with the help of a transfer-learning based model and a CNN model. A suitable parking spot is searched based on the vehicle type and assigned to the vehicle. A customer would be monitored with the help of webcams and number plate recognition model after entering the parking lot and would be alerted through message if he/she has parked his/her vehicle in a spot that is not assigned to him. The customer would then be displayed with a bill at the exit point with automated calculation of parking charges by the system. YOLOv3 and Tiny-Yolov3 models has been proposed for vehicle detection. The models have been compared by detection time per frame, detection accuracy and its F1 Score. On exploration, an EfficientNet based TL model, mobileNet Extended model and light-weight CNN model has been proposed for vehicle type classification. Testing is done to compare the classification accuracy and time. The EfficientNet based TL model performed extremely well in test data as well as in real-time data. For number plate detection, a pre-trained cascade classifier has been proposed. On the other hand, a CNN model, a ResNet50 extended model and a ResNet152 extended model has been proposed for the purpose of character recognition. The models have been compared based on their classification accuracy and average recognition time per character. The output system has been delivered as a website to the end-user with 2 separate portals for customer and parking lot owner. Each portal has its own features that satisfy its target user. The proposed system has been efficient in satisfying needs of the customer and the vehicle type classification model has given a good accuracy score.