Open AccessAutonomous Parking Spot Detection System for Mobile Phones using Drones and Deep Learning
Author(s) -
Guilleum Budia Tirado,
Sudhanshu Kumar Semwal
Publication year - 2021
Publication title -
computer science research notes
Language(s) - English
Resource type - Conference proceedings
eISSN - 2464-4625
pISSN - 2464-4617
DOI - 10.24132/csrn.2021.3002.13
Subject(s) - computer science , drone , artificial intelligence , process (computing) , object detection , parking lot , parking guidance and information , path (computing) , object (grammar) , convolutional neural network , machine learning , real time computing , computer vision , pattern recognition (psychology) , computer network , engineering , operating system , transport engineering , genetics , civil engineering , biology
Many parking lot facilities suffer from capacity over-loads and many times there are no monitoring tools toprovide feedback. As a consequence, the people, want-ing to park, become frustrated as there is considerablyloss of time. In this paper, we present a novel proto-type of an automatic parking-lot analysis platform us-ing image-based machine learning to (a) guide a droneautonomously; and (b) to process useful information tobe handled into a smartphone application to communi-cate with the parking lot users.We have collected a reasonable amount of test imagesto build a classification model using Convolutional neu-ronal networks (CNNs) to classify parking lot images,and build different object detection models to identifyfree and occupied parking spots. Those models havebeen exported to the back-end module of our platformso it can control the drone and record the computed in-formation to its database. In addition, we have imple-mented an iOS application that requests and displaysthe parking lot status and its empty spots.We have been able to prove that this prototype is fea-sible, functional, and opens a path towards future im-provements and refinements. The flight control and thedata classification algorithms have been shown to workusing the machine learning models. In summary, wefound a clear and and concise way to display useful in-formation in real time to our users.