In developing sidewalks for urban planning, it is necessary to investigate how and by whom people use the sidewalks. Traditional methods estimate the number of pedestrians based on the images from visible light cameras installed on the roadway, but the accuracy may be reduced in low-light conditions such as rainy weather or at night. In such low-light conditions, existing research has shown that infrared cameras can detect pedestrians.
For a more accurate estimation of the number of pedestrians, a visible light camera and a far-infrared camera are installed above the bus stop and have one year's worth of data. The far-infrared camera cannot detect pedestrians correctly depending on the temperature because it captures the temperature information of objects. Therefore, this research analyzes the conditions under which pedestrian count estimation fails using captured video and metadata, such as temperature and weather. By combining visible light and far-infrared images, we aim to estimate the number of pedestrians throughout the day regardless of temperature and illumination.
Publication
- Takumi Fukuda, Ismail Arai, Arata Endo, Masatoshi Kakiuchi, Kazutoshi Fujikawa, “Benchmark of Deep Learning Visual and Far-Infrared Videos Toward Weather-tolerant Pedestrian Traffic Monitoring,” IEEE International Conference on Smart Mobility (SM) (SM'2023), Mar. 2023.