EDGE-CLOUD COLLABORATION BASED SYSTEM TO REDUCE LATENCY AND SAVE ENERGY IN PARKING OCCUPANCY DETECTION

• Main Author: Tritama Gamadita, Yudha
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86194
• Institution: Institut Teknologi Bandung

The increasing number of motor vehicles, particularly cars, has created a problem related to the limited availability of parking spaces, causing difficulties for drivers in finding empty parking spots. This situation not only worsens traffic congestion but also increases air pollution emissions, as drivers must continually move to find parking. One of the solution to address this issue is to provide real-time information about empty parking spaces to help drivers find parking quickly. Many studies have adopted IoT sensors as a technology to detect parking occupancy. However, the IoT sensor technology approach has constraints in terms of high installation and maintenance costs, especially if parking areas are expanded. Deep learning-based computer vision technology offers an alternative solution for more scalable and cost-efficient parking occupancy detection. The use of deep learning can accurately and flexibly detect parking occupancy according to the needs of the parking area. However, implementing deep learning-based computer vision requires high computational resources. The substantial energy consumption involved in using deep learning methods for parking occupancy detection is another drawback, as the system needs to operate continuously 24 hours a day. Some studies have adopted cloud computing as a solution to this problem, but this approach has the disadvantage of latency in data transmission. Therefore, this research aims to develop an edge-cloud collaboration-based system for parking occupancy detection. The edge-cloud collaboration-based system is developed by dividing the workload between edge computing and cloud computing. Edge computing is tasked with conducting the direct detection of parking occupancy from the data source, thereby reducing latency. Meanwhile, cloud computing is responsible for collecting samples, managing datasets, training deep learning models, and displaying parking occupancy detection results to users. A combination method of motion detection and object detection for parking occupancy detection is also designed in this study to save energy consumption on edge computing. Motion detection plays a role in detecting activity in parking areas and minimizing the use of object detection, which requires deep learning and consumes a lot of energy. The system developed in this study has successfully reduced overall latency in the process of parking occupancy detection. By utilizing the YOLOv5nu model and the MQTT protocol for faster data transmission, the system has achieved a speed improvement of 29.45% compared to using cloud computing alone. Moreover, by combining motion detection and object detection in the overall parking occupancy detection process, this method's design has successfully saved up to 37.93% of energy on edge computing compared to using exclusive deep learning object detection. Thus, the system developed in this study, by dividing the workload between edge computing and cloud computing, can reduce the latency of the parking occupancy detection process and simultaneously save energy during its implementation..