Parkpal: A computer vision guided smart parking system in a parking space

ABSTRACT: STATEMENT OF THE PROBLEM: Finding parking is one of the most frustrating challenges for drivers after navigating through traffic. With expanding population and increasing car density, the problem of finding available parking spaces in cities has been exacerbated by ongoing economic growth. However, existing solutions to this issue are costly and ineffective. Therefore, researchers have proposed a computer vision-based parking system as a potential solution that requires further investigation to effectively address the problem. This system would leverage machine learning algorithms to analyze video footage and identify available parking spaces, helping drivers to locate parking more efficiently. In particular, the study intends to accomplish the following objectives: • To create a user-friendly website application for parking guests, enabling them to locate a parking spot before they arrive, while also providing a platform for parking monitoring and management for the administrator. • To assess the accuracy of the developed and trained model in identifying available parking spaces by utilizing Intersection Over Union (IOU) method. • To evaluate the overall performance of the vacancy detection model through the following metrics: - Recall - Precision - F1-score - Mean Average Precision (mAP) • To measure the level of acceptability of the developed system to the users in terms of usability RESEARCH METHODOLOGY: To gain a thorough understanding of the system’s requirements, enhance its functionality, and tailor it to the preferences and needs of its users, the researchers adopted a multi-faceted approach that combined qualitative and quantitative data collection and analysis. They employed exploratory research designs that allowed them to gather a broad understanding of the system’s requirements for further investigation. They utilized the Scrum software development methodology, which is an iterative and incremental approach that emphasizes and adaptability, to facilitate the development process. The primary aim of the post-questionnaire in this study is to assess the effectiveness of the prototype by evaluating its performance with regards to various performance metrics and themes of interest that were identified during the pre-interview stage. This paper collected data from voluntary participants which included parking guests at any establishment and of four-wheel vehicles. Overall, this research methodology provides a valuable framework for researchers looking to develop and improve software systems and tailor them to the needs and preferences of their users. SUMMARY OF FINDINGS: The goal of this study was to create a Raspberry Pi 4B-based system that could recognize and classify parking spots as filled or unoccupied. To achieve this goal, the Python programming language and Convolutional Neural Networks (CNN) with the YOLOv4 algorithm were used. Interviews and a survey using Google Forms were conducted to gather feedback on the constructed system from the volunteered respondents. Pre-interview data underwent content analysis to extract information for building a prototype that matched the demands of the target audience. The system prototype was built and then underwent thorough testing to evaluate its accuracy, recall, F1-score, intersection over union (IOU), and mean average precision (mAP). The tests resulted in a recall of 0.93, precision of 0.84, and F-1-score of 0.88, demonstrating the model’s accuracy in identifying positive cases. With a mAP of 89.35% and an average IOU of 73.35% at a threshold of 70%, object detection was performed with great precision and recall. The Post-Study System Usability Questionnaire (PSSUQ) was used to collect participant input after user testing of the system was completed. The evaluation yielded a score of 1.55, indicating the effectiveness of the method. Information quality earned a score of 1.53, and interface quality received a score of 1.51. The overall PSSUQ score was 1.53, indicating among usability of the developed system. CONCLUSION: To address the pervasive parking problems in the Philippines, this study aimed to design a parking system that could accurately detect and locate vacant spots. The system employed the YOLO method and training datasets to achieve precise detection capabilities. To ensure that consumers are well-informed about the availability of parking spaces and provide them with a convenient option to reserve their desired spots, a user-friendly online application was developed. This application serves as a platform through which users can access up-to-date information on parking availability, allowing them to make informed decisions and plan their parking accordingly. The implementation of real-time video feeds enabled effective monitoring and management of parking spaces. By utilizing this approach, authorities and administrators can closely monitor the occupancy of parking areas, identify any issues or inconsistencies, and take necessary actions to ensure smooth parking operations. Through the evaluation of the training model, it was observed that the system exhibited commendable accuracy in correctly identifying and classifying objects within the parking spaces. Although further refinements are required to accurately detect vacant spaces, the overall performance of the system indicates promising results in terms of effective parking space detection. The system successfully met user expectations by providing a user-friendly interface, reliable information quality, and seamless interaction. Users found the system intuitive and easy to navigate, enhancing their overall satisfaction and usability. This study presents prospective remedies that can potentially alleviate parking problems in metropolitan areas and advance the development of smart parking systems. RECOMMENDATION: Based on the data collected from both respondents and the evaluation of the model, the researcher has developed the following recommendations to enhance the study’s performance and provide avenues for future researchers to improve their research. These recommendations encompass various aspects, including increasing the sample size for user testing, incorporating license plate recognition, enabling multiple car registrations under a single account, implementing Random ID Generator APIs, establishing a QR code or barcode system to ensure authorized parking guests, and adjusting camera placement to a high position where car movements do not obstruct its line of sights.

Filipiniana