An image processing apparatus which employs a basic configuration including an image processing controller which processes images captured by a stereo camera and a recognition processing controller which recognizes an object based on information from the image processing controller includes a target object area specifying unit, feature amount extracting unit and smoke determining unit as functions of enabling recognition of a smoky object. The target object specifying unit specifies an area of an object which is a detection target, by canceling the influence of the background, the feature amount extracting unit extracts an image feature amount for recognizing a smoky object in a target object area and the smoke deciding unit decides whether the object in the target object area is a smoky object or an object other than the smoky object, based on the extracted image feature amount.