While several methods have been proposed for detecting three-dimensional (3D) objects in semi-real time by sparsely acquiring features from 3D point clouds, the detection of strongly occluded objects still poses difficulties. Herein, we propose a method of detecting strongly occluded objects by setting up virtual auxiliary point clouds in the vicinity of the target object. By generating auxiliary point clouds only in the occluded space estimated from a detected object at the front of the sensor-observed region, ie, the occluder, the processing efficiency and accuracy are improved. Experiments are performed with various strongly occluded scenes based on real environmental data, and the results confirm that the proposed method is capable of achieving a mean processing time of 0.5 s for detecting strongly occluded objects.