[3G1-OS-14a] OS-14

9:30 AM - 9:45 AM JST(12:30 AM - 12:45 AM UTC)

[3G1-OS-14a-03] Pedestrian Flow Estimation Based on Human Observations

〇Dichi Sakuma¹, Miyoshi Yuki¹, Masaki Inoue¹, Kanato Takeuchi¹, Hiroshi Watanabe¹ (1. Keio University)

Keywords:

Kalman filter,sensor fusion,human observation

Real-time monitoring of large-scale systems, such as social infrastructures, is becoming increasingly critical. However, comprehensively covering an entire target domain with high-precision physical sensors is often impractical due to high installation costs and communication constraints. To address this issue, we propose a state estimation method based on the Kalman filter that fuses physical sensors, which are precise and quantitative but sparse, with human sensors, which are noisy and qualitative but widely distributed.
We evaluated the proposed method through numerical experiments on a human flow system characterized by complex nonlinear behaviors. Specifically, we identified a linear approximation model from data generated by the crowd simulator CrowdWalk, which was used as the internal model of the estimator. Crucially, the observation model for human sensors explicitly accounts for quantization errors arising from subjective categorical reporting, in addition to perception noise.
Experimental results demonstrate that the proposed method achieves high accuracy and robustness, even in the presence of modeling errors caused by linear approximation and the significant uncertainty inherent in human sensory data.