Research On Estimating Illumination Direction From Cast Shadow In Outdoor Scenes

In computer vision field, many important applications can get more robustness ofthe results with known illumination conditions. Thus, estimating scene illuminationcondition is an important research topic in computer vision field. Illuminationconditions include three components: the illumination source direction, illuminationintensity and illumination color. There are already a lot of literatures for estimatingillumination direction in the specific indoor scene. However, there are only fewliteratures for estimating illumination direction in outdoor scene. Due to the scenegeometric model complexity and illumination condition is uncontrollable in theoutdoor scene. There are some difficulties for estimating illumination directionaccurately in the outdoor scene. Thus, estimated light source direction for the outdoorscenes is very challenging research topic. In this paper, this difficult problem issolved by using shadow information. The main work of this paper includes:For the single method of shadow detection get lower accuracy rate, an adaptivethreshold shadow detection method is proposed combination of the bright channel andbrightness rate features. First, the object pixels in the foreground are separatedinitially by using the features of bright channel, brightness ratio and chromaticitydifference, then shadow and object areas in the foreground are discriminated by usingthe adaptive threshold of chromaticity difference and brightness ratio. Experimentswere carried out based on a public dataset, the effectiveness of the method evaluatedby using the factors of shadow detection rate and shadow distinguish rate. Bycomparison with the single adaptive threshold shadow detection method, the resultsshowed that our shadow detection method is more efficiently.A method is proposed to estimate illumination direction from the cast shadow inimages of outdoor scenes.The brightness rate (BR), which is independent of reflectionproperties of the material but dependent on illumination intensity onto the material, iscomputed in shadow areas. A BR line is fitted onto the points that have minimal BRvalues. Based on the property that illumination intensity at the points in a cast shadowarea varies depending on their locations, illumination direction is estimated bymodeling the height of the object, incident rays from the source of the illumination,and the BR line. Experiments with known illumination direction in indoor scenes,experiments in outdoor scenes and comparative analyses were carried out. The results showed that our method enables illumination direction to be estimated efficiently,without complicated image segmentation.