Conclusion

A simple, early method for background removal would be a useful step in many object recognition and tracking problems. We have demonstrated such a method based on the joint use of range and color data. This approach is quite compelling since fast, cheap (R,G,B,Z) sensors will be available commonly in the near future.

There are several advantages of this particular segmentation approach. The use of color and range together reduces the effect of classic segmentation problems in each data source when taken separately including: 1) points with similar color background and foreground, 2) shadows, 3) points with invalid data in background or foreground range, and 4) points with similar range background and foreground.

Background estimation in joint range and color space also presents several advantages. Higher dimensional histograms allow better separation of background and foreground statistics, resulting in a cleaner estimate at each point. The special interpretation of background as the farthest range event implies that at each point the background has to be visible in fewer frames for accurate background estimation. Background estimation in a scene which always contains some foreground elements is, in itself, a useful tool in site modeling and graphics.