Optical consistency between the real world and virtual objects is an important issue in augmented reality (AR). We propose a method to estimate illumination from an object shadow and incomplete object shape information captured by an RGB-D camera. The environmental illumination can be estimated without any prior knowledge of the object shape. The radiance of each light source is computed by solving linear equations derived from color and depth images. Since prior knowledge of the object shape is not necessary, we can flexibly use/combine any objects. Thus, we can use/combine multiple objects in the target scene to increase the accuracy and flexibility. Our experimental results show the characteristics of our proposed method.