Traffic Monitoring and Analysis At Toll Plaza

This paper presents a solution to Traffic monitoring and analysis at Indian toll plazas. In some of the areas, the work is done on Vehicle detection and localization, vehicle registration detection, and character recognition and vehicle make classifier (Currently concentrated on Indian Cars only). Accurate detecting and localizing of objects in computer vision has always been a core problem, to the rescue of which Tensor Flow Object detection API comes with implementations provided by RCNN family (Basic RCNN, Fast RCNN, and Faster RCNN) and Single-shot detection models. Experimenting over a range of models from Faster RCNN with inception v2, SSD MobileNet, and SSD inception v2 for vehicle and vehicle registration number detection problem we got higher accuracy with Faster RCNN with inception v2 for RMS Prop optimizer. For the detected vehicle registration number, we used an SVM-based multiclass classifier. A custom image dataset is used to train the model. The vehicle make classifier is implemented with VGG16 and the dataset was selected to contain only Indian cars.