Abstract: In recent years, object proposals has become a major research area. Object proposals define
and train a measure of objectness generic over classes. But the current research about objectness is
based on RGB image. We give a measure of objectness via RGBD images. It combines current
state-of-the-art RGB objectness, and design two objectness cues based on depth image, then use a
Bayesian framework to combine them. At NYU Depth dataset we demonstrate that the combined
objectness measure performs better than any cue alone, and also outperforms traditional objectness
based on RGB image. It′s proven that the addition of depth map can better optimize objectness.

Key words: object proposals, RGBD, object detection, object recognition