In the 21st century, aging is becoming an increasingly significant global reality for all countries. As the elderly population grows, there is a need to redesign existing services to meet their companionship, care, and support needs. According to the World Health Organization, approximately 0.42 million people die each year as a result of falls. In this project, an autonomous mobile robot system was used to detect falls in elderly individuals and provide voice-assisted rapid intervention. The system includes a Raspberry Pi, a camera, and a data communication module mounted on an autonomous mobile robot (Evarobot). An ESP-32 embedded board and an accelerometer were used for fall detection. The fall detection system is a portable device, which is attached to the individual's belt. In this study, a threshold-based algorithm was used for fall detection. When the device detects a fall, it sends a message to the Raspberry Pi using the MQTT protocol. The Raspberry Pi then sends a message via ROS to start the roaming of the robot. It is assumed that the elderly individual lives in a 1+0 home environment. The robot knows the map of the home environment. The robot autonomously moves to predetermined points within the home and captures images of the surroundings. In this study, YOLO model pretrained with 15,247 images was used. The images captured by camera were fed to YOLO model to determine whether a person is lying on the floor. If a person lying on the floor is detected, the robot stops roaming and moves toward the detected direction. During this movement, distance sensors are used to ensure the robot approaches the fallen person as closely as possible. The robot uses a microphone and speaker to ask the person questions about their condition and listens to their responses. If the person is deemed to be in good condition, no emergency notification is made; otherwise, caregivers are alerted. The notification message is sent by PushBullet platform. The tests for the robot's roaming in the home, capturing images from the camera, and YOLO-based fall detection from image data were conducted in the Gazebo environment. Additionally, the ESP-32 based fall detection device and the YOLO-based fall detection application were tested in a laboratory setting. All the results show that the voice-assisted robot can also serve as a fall detection observer in elderly’s life.

Keywords: Elderly, autonomous robot, human detection, fall detection, YOLO, voice command

This work has been supported by TÜBİTAK 2209B Industrial Oriented Research Project Support Program in 2023/2 under project number 1139b412394689.