Street Parking Sign Detection, Recognition and Trust System

Abstract

Parking is one of the major problems in autonomous driving. Although cars can park in a parking spot automatically now, they can't find where they can park. In this thesis, we propose a novel street parking sign detection and recognition pipeline. This pipeline detects and recognizes street parking signs in images based on deep neural networks. After testing several models, we adopt RetinaNet\cite{lin2017focal} as our street parking sign detection model, and CTPN-EAST-CRNN as our street parking sign recognition pipeline. To train the neural network, we build a street parking sign dataset containing street parking sign images, bounding boxes, and text. To the best of our knowledge, this is the first work on street parking sign detection and recognition. In addition, we also build a street parking sign detection and recognition system, which includes a server and an app. Users can use the app for uploading and check any street parking signs around. For improving the data credibility, we purpose a trust system for evaluating the confidence scores and reputation level of parking sign uploading and users. Based on this trust system, our system will get much more reliable data and provide correct information for users.