The recent increase in personal mobility (PM)sharing services is associated with an increase in the number of PM accidents. According to the Korea Road Traffic Authority’s annual PM accident statistics, 897 accidents occurred in 2020, rapidly increasing to 2,386 in 2022. Accordingly, PM-related deat...
The recent increase in personal mobility (PM)sharing services is associated with an increase in the number of PM accidents. According to the Korea Road Traffic Authority’s annual PM accident statistics, 897 accidents occurred in 2020, rapidly increasing to 2,386 in 2022. Accordingly, PM-related deaths and injury accidents also tended to increase. In addition, PM’s illegal parking, careless and reckless driving, failure to wear protective gear, and non-compliance with the ridership quota are causing situations that harm the environment and safety of the roads. As a related study to solve this problem, a system was implemented to detect PM passengers’ failure to wear helmets, violations of the riding quota, PM’s parking status, and normal driving conditions in closed-circuit television (CCTV) video-based images. However, due to the nature of CCTV, it was not fixed. The problem is that it can only be detected at a location, so it cannot be detected if it is outside the shooting range. Personal mobile device detection based on drone video data was proposed. However, since drones are used, there are problems with the drone’s flight restriction area and flight time depending on the battery. Many methods have been proposed to detect whether a passenger is wearing a helmet. However, these methods have the problem of not being able to detect other violations, such as non-compliance with the riding quota other than wearing a helmet. To overcome these limitations of existing research, this paper proposes a personal mobile device violation detection system using mobile AIoT devices. This system solves the problem of only being able to detect a fixed location and the flight restrictions of drones by installing an artificial intelligence of things (AIoT) device on the PM, which has a longer travel time than a drone, to provide mobility. Other detection items other than not wearing a helmet were added. The proposed personal mobile device violation detection system consists of a mobile AIoT device for detecting violations of PM passengers, a mobile Internet of Things (IoT) application for measuring location and network signal field strength, and an IoT server for data transmission and storage and violation. It comprises a web monitoring application for information and signal field strength. First, the mobile AIoT device transmits image and text data that detects conditions such as not wearing a helmet, non-compliance with the riding quota, and parking conditions to the IoT server through the PM violation detection model. Next, the mobile IoT application collects location, network, weather, and air quality data, such as PM passengers’ latitude and longitude, from the smartphone and transmits it to the IoT server. The IoT server stores all received data and uses a server that complies with the international IoT standard oneM2M. Lastly, the web monitoring application requests violation detection images and text data from the IoT server, the location at the time of violation detection, and the network receives field strength data at that location and receives the response data in real time through the map application programming interface (API). Mark it on the map. In particular, to implement a PM violation detection model in a mobile AIoT device, 3,301 image data were generated using image data collected by a camera. Three images were used: normal, which corresponds to the normal driving state of the PM rider, no helmet, which corresponds to the state of not wearing a helmet, and two or more people. A dataset of 7,990 objects was constructed by labeling the generated image data into four classes, including two persons, which corresponds to non-compliance with the passenger capacity due to abnormal boarding, and parking, which corresponds to the parking state. Through the constructed dataset, a PM violation detection model was learned using you only look once (YOLO)v5 structures. As a result of learning, we implemented a violation detection model with a training performance of 98% for test data in mean average precision (mAP), 97% for YOLOv5-n, and 99% for YOLOv5-x. Examples of how to utilize this system and its expected effects include PMsharing service providers, local governments, and police. First, PM-sharing service providers can encourage users to drive safely by providing mileage or rewards. Second, local governments can use it to install and reinforce parking infrastructure in areas where PM parking problems frequently occur. Lastly, police can be used to strengthen patrols in areas where violations are frequently detected. Keywords: Artificial Intelligence, Internet of Things, Mobile, Personal Mobility, Safety Violation,
The recent increase in personal mobility (PM)sharing services is associated with an increase in the number of PM accidents. According to the Korea Road Traffic Authority’s annual PM accident statistics, 897 accidents occurred in 2020, rapidly increasing to 2,386 in 2022. Accordingly, PM-related deaths and injury accidents also tended to increase. In addition, PM’s illegal parking, careless and reckless driving, failure to wear protective gear, and non-compliance with the ridership quota are causing situations that harm the environment and safety of the roads. As a related study to solve this problem, a system was implemented to detect PM passengers’ failure to wear helmets, violations of the riding quota, PM’s parking status, and normal driving conditions in closed-circuit television (CCTV) video-based images. However, due to the nature of CCTV, it was not fixed. The problem is that it can only be detected at a location, so it cannot be detected if it is outside the shooting range. Personal mobile device detection based on drone video data was proposed. However, since drones are used, there are problems with the drone’s flight restriction area and flight time depending on the battery. Many methods have been proposed to detect whether a passenger is wearing a helmet. However, these methods have the problem of not being able to detect other violations, such as non-compliance with the riding quota other than wearing a helmet. To overcome these limitations of existing research, this paper proposes a personal mobile device violation detection system using mobile AIoT devices. This system solves the problem of only being able to detect a fixed location and the flight restrictions of drones by installing an artificial intelligence of things (AIoT) device on the PM, which has a longer travel time than a drone, to provide mobility. Other detection items other than not wearing a helmet were added. The proposed personal mobile device violation detection system consists of a mobile AIoT device for detecting violations of PM passengers, a mobile Internet of Things (IoT) application for measuring location and network signal field strength, and an IoT server for data transmission and storage and violation. It comprises a web monitoring application for information and signal field strength. First, the mobile AIoT device transmits image and text data that detects conditions such as not wearing a helmet, non-compliance with the riding quota, and parking conditions to the IoT server through the PM violation detection model. Next, the mobile IoT application collects location, network, weather, and air quality data, such as PM passengers’ latitude and longitude, from the smartphone and transmits it to the IoT server. The IoT server stores all received data and uses a server that complies with the international IoT standard oneM2M. Lastly, the web monitoring application requests violation detection images and text data from the IoT server, the location at the time of violation detection, and the network receives field strength data at that location and receives the response data in real time through the map application programming interface (API). Mark it on the map. In particular, to implement a PM violation detection model in a mobile AIoT device, 3,301 image data were generated using image data collected by a camera. Three images were used: normal, which corresponds to the normal driving state of the PM rider, no helmet, which corresponds to the state of not wearing a helmet, and two or more people. A dataset of 7,990 objects was constructed by labeling the generated image data into four classes, including two persons, which corresponds to non-compliance with the passenger capacity due to abnormal boarding, and parking, which corresponds to the parking state. Through the constructed dataset, a PM violation detection model was learned using you only look once (YOLO)v5 structures. As a result of learning, we implemented a violation detection model with a training performance of 98% for test data in mean average precision (mAP), 97% for YOLOv5-n, and 99% for YOLOv5-x. Examples of how to utilize this system and its expected effects include PMsharing service providers, local governments, and police. First, PM-sharing service providers can encourage users to drive safely by providing mileage or rewards. Second, local governments can use it to install and reinforce parking infrastructure in areas where PM parking problems frequently occur. Lastly, police can be used to strengthen patrols in areas where violations are frequently detected. Keywords: Artificial Intelligence, Internet of Things, Mobile, Personal Mobility, Safety Violation,
AI-Helper
※ AI-Helper는 부적절한 답변을 할 수 있습니다.