Barrier Detection Using Sensor Data from Multiple Modes of Transportation with Data Augmentation

To present routes that are traversable for modes of transportation such as wheelchairs or strollers requires accessibility information, i.e., detailed information regarding the type and location of barriers. Earlier research recognized the types of barriers using sensor data from able-bodied people and people in wheelchairs, but the number and range of travel of people in wheelchairs is limited. Also, small barriers could not be detected by able-bodied people, as they are able to absorb small barriers. In our research, the goal was to detect details regarding barriers using sensor data from various modes of transportation. However, there are issues with the selection of barrier recognition models for each mode of transportation and with the cost of collecting data. To overcome model selection problem, we propose a model that recognizes modes of transportation and barriers in two stages. We also propose a method for reducing the cost of collecting data, in which we prepare a course with a smooth surface, collect data, and simulate rough surfaces by adding noise. In experiments, we achieved accuracy of 91.5% recognizing modes of transportation in six classes, and showed that adding noise increased accuracy by 3.7 percentage points for rough surfaces. When recognizing eight classes of barrier, we achieved accuracy of 87.7% when using a stroller, and showed that adding noise increased accuracy by 6.8 percentage points for rough surfaces. We have shown that the proposed method is effective and implemented recognition of barrier details using multiple modes of transportation.