人の流れに基づく危険予知システムに関する一提案

摘要

The people flow data is defined as the time series of events each of which is composed of spatio-temporally interpolated attributes set for every single minute such as "identification number, date, time, position, sex, age, address, job, purpose and traffic means", those are acquired by the public investigations. This kind of people flow data is initiated in 1998 at Tokyo and after that, Japanese major cities followed. Recently, the activities are expanded abroad. By utilizing such large databases, if it becomes possible to predict when and where to come across the pedestrians, the risk prediction will become possible and stable, even if the sensing and the communication are extremely degraded. In this article, we firstly focus on the "people flow" as the most important exterior factor for a vehicle which exists behind the human error. Secondly, we propose the concept of the risk prediction system which works through the three stages of operations such as acquisition, learning and prediction of the distributed cooperation databases.%人の流れデータとは，公的調査で得られた「識別番号，日時，位置，性別，年齢，住所，職業，目的，交通手段」を1分間隔で時空間内挿した時系列データであり，98年東京を皮切りに全国主要都市や海外にも展開されている.この大規模データべースを利用して，いつどこで歩行者と出くわすかを予測できれば，センシングや通信が劣悪でも安定して危険予知が行える.本稿ではヒューマンエラーの背後に存在する車室外要因として最重要である「人の流れ」に注目し，分散協調型データべースの獲得・学習・予測の3段階により危険予知を行うシステムのコンセプトを提案する.