File pdf 1.1MB

Convolutional Neural Networks for Vehicle Tracking in Parking Lots

Recent improvements in convolutional neural networks’ (CNNs) object detection rate and accuracy are spurring new innovations to free humans from mundane, repetitive tasks. In our senior design capstone project, we implement a CNN application to monitor vehicle compliance with parking regulations (such as time and spaces) in public lots. Although a distributed sensor network could potentially achieve this monitoring, such a network exhibits a prohibitive high cost, and it may not achieve the same accuracy as CNNs. In our project, we will mount cameras in a parking lot to capture live video feed that will be analyzed through deep learning and object detection models. Our client application can then notify security and safety companies of potential violations of parking regulations. We also envision interfacing our application with drivers who can then track available parking spaces and their vehicle parking time. We will measure the success of our project by the accuracy of parking space detection, the accuracy of vehicle detection, and the accuracy of the timing mechanism for each vehicle in the lot. We are also involving our University Campus Public Safety to measure their satisfaction with the implementation to enforce local parking regulations.