A 2D-Lidar-Equipped Unmanned Robot-Based Approach for Indoor Human Activity Detection

Area of Science:

• Gerontology
• Robotics
• Artificial Intelligence

Background:

• Monitoring elderly individuals living alone is crucial for detecting hazardous events like falls.
• Traditional 2D Light Detection and Ranging (LIDAR) systems face limitations due to furniture obstructing line-of-sight (LOS) and inability to detect falls after the event.
• Static sensor placement restricts detection capabilities, especially if a fall occurs and is not immediately identified.

Purpose of the Study:

• To propose and evaluate a novel fall detection system utilizing a 2D LIDAR sensor integrated with an autonomous cleaning robot.
• To overcome the limitations of static LIDAR systems in realistic home environments with occlusions.
• To enable detection of falls even after a delay by leveraging the robot's mobility.

Main Methods:

• A 2D LIDAR sensor was mounted on a cleaning robot to continuously collect distance measurements while roaming.
• LIDAR data was transformed, interpolated, and compared against a reference environmental state.
• A convolutional long short-term memory (LSTM) neural network was trained to classify processed measurements for fall event detection.

Main Results:

• The proposed system achieved 81.2% accuracy in fall detection.
• The system demonstrated 99% accuracy in detecting individuals in a lying position.
• Compared to static LIDAR, the mobile system showed significant accuracy improvements (69.4% to 81.2% for falls, 88.6% to 99% for lying detection).

Conclusions:

• Integrating 2D LIDAR with a cleaning robot offers a robust solution for elderly fall monitoring, overcoming static sensor limitations.
• The autonomous mobility of the robot enhances the ability to detect falls and lying postures, even with environmental occlusions.
• This approach significantly improves the reliability and accuracy of fall detection systems for independent elderly living.