本論文では，寒冷地ITSの取り組みの1つとして，「プローブパーソンによる集合知を用いて路面凍結情報を網羅的に提供する方式」を提案し，評価する．近年，ITS分野ではプローブ情報システムへの関心が高まっている．路面の凍結状況をプローブするためには，移動速度・路面形状・路面状態・天候・昼夜を問わず，路面凍結を精度良く検出できることが望ましい．しかしながら，これまでに提案されてきた路面凍結センサはこれらの条件の充足が十分とはいえず，プローブカーに設置するものとして採用するために十分とはいえない．プローブ情報システムには，自動車をセンサとして用いるプローブカーのほかに，人間をセンサとして用いるプローブパーソンというアプローチがある．また，近年Web2.0と呼ばれる潮流の中で，多数の人間が持つ知識を融合して1つの知識として活用する「集合知」への評価が高まっている．そこで本論文では，「プローブパーソンによる集合知を用いて路面凍結情報を提供する方式」を提案する．そして，試作システムを構築したうえで，冬期に実際に稼動実験を行い，その有効性を評価する．評価実験の結果，定性的評価では精度で75.4%，有用性で72.1%の閲覧者から好意的な評価を得ることができた．また定量的評価では，厳密精度は46.0%にとどまったが，ε精度は92.3%が得られた．「走行環境情報の提供」や「危険警告」を主目的としていることを考慮して本論文で定めた達成度基準によれば，提案方式は及第レベルにあり，今後さらなる工夫によって有効性向上の期待が持てる．

In this paper, we propose and evaluate a method: "continuous provision of frozen road surface information using collective intelligence by probe persons", as one of the approaches to ITS for colder areas. Recently, lots of researchers in ITS area are attracted to utilizing probe information system effectively. In order to probe frozen road surface, we need a precise sensor of frozen road surface which can detect in any moving speed, any road shape, any road condition, and any weather without limited to either daytime or night. However, since the conventional sensors for frozen road surface are not always sufficient in satisfaction of these conditions, we can not always say that they are adequate for adopting as setting in a probe car. In probe information system, addition to "probe car" approach using a car as a sensor, "probe person" one using a person as a sensor has been tried. Furthermore, recently, in a trend called "Web2.0", evaluation to "collective intelligence" using a knowledge by fusion of knowledge from many persons has been grown. Therefore, in this paper, we propose a method: "continuous provision of frozen road surface information using collective intelligence by probe persons". We also implement a pilot system, carry out a public demonstration experiment in winter season, and evaluate its effectiveness. According to its results, in qualitative evaluation, we have obtained good evaluation from 75.4% audience concerning preciseness and from 72.1% ones concerning usefulness. On the other hand, in quantitative evaluation, although we have achieved only 46.0% strict precision, we have achieved 92.3% ε precision. The proposed method is now in pass level as achievement one defined in this paper taking into account the main purpose which is "provision of driving environment information" and "hazard warning", and has possibility to improve its effectiveness by some modifications.