Intelligent　vehicle　systems,　such　as　advanced　driving　assistance　systems,　are　relatively　popular　nowadays,　which　facilitate　the　timely　prevention　of　driving-related　accidents　and　human　injuries　caused　by　impaired　driving.　An　often　ignored　problem　in　existing　systems　is　the　heterogeneousness　across　multimodal　sensor　data,　which　is　becoming　more　crucial　due　to　the　widespread　use　of　more　different　sensors.　This　paper　proposes　a　novel　feature　fusion　based　detection　approach　using　deep　convolutional　neural　network　to　profile　driver-related,　vehicle-related,　and　road-related　features,　and　extract　collaborative　information　from　them.　Experimental　results　demonstrate　that　the　proposed　approach　is　capable　of　providing　more　accurate　detection　of　impaired　driving,　compared　with　that　achieved　by　other　state-of-the-art　classifier.