Smoothing Range Image using Reflectivity

Post on 17-Oct-2021

6 views 0 download

Transcript of Smoothing Range Image using Reflectivity

Smoothing Range Image using ReflectivityShuji Oishi∗, Ryo Kurazume∗, Yumi Iwashita∗

∗Kyushu University, Fukuoka, Japan

Abstract— This paper proposes a new smoothing technique forrange images taken by a Time-of-Flight range finder. The most ofTime-of-Flight range finders measure not only the range data butalso the power of the reflected light as a side product. The basicidea of the propose technique is that the reflectance and rangeimages are utilized for smoothing the range image corrupted bynoise so that the detailed shape in the range image is preserved.

I. I NTRODUCTION

In recent years, a variety of low cost and real time Time-of-Flight (RT-TOF) range scanners have been commerciallyavailable. On the other hand, high-precision (but very expen-sive) three dimensional laser range finders have been developedfor landscape surveying or digital 3D modeling. We alsohave been developing a low cost and high resolution lasermeasurement systems using a laser scanner and a rotary tablefor 3D environmental map building[1]. However, range imagesacquired by these sensors are suffered by small noises dueto the reflectance property of objects’ surfaces or electricaland mechanical errors. Therefore the denoising problem forrange sensors (RT-TOF range scanners and LIDAR sensors)still remains as an essential problem.

In this paper, we propose a new smoothing technique focus-ing on the reflectivity. This method allows us to smooth a rangeimage and remove noises while preserving detailed shapes.

II. SMOOOTHING USINGREFLECTIVITY

A. Reflectance Image

Optical range sensors such as a laser range finder obtainrange data by measuring a round-trip time of a laser pulsereflected by an object. On the other hand, most of opticalrange sensors can measure the intensity of the reflected laserpulse (reflectivity) with its round-trip time. The intensity valueis determined uniquely according to the pixels in the rangeimage. In other word, the range image and the reflectance imageare fundamentally aligned precisely. Therefore, we will usenot only the range image but also the reflectance image forsmoothing the range image.

B. Edge-preserving Smoothing filter

We propose a new edge-preserving smoothing filter whichutilizes reflectance and range images for smoothing a rangeimage as follows:

gi =

∑j w(xi, xj)w(fi, fj)w(di, dj)fi∑j w(xi, xj)w(fi, fj)w(di, dj)

(1)

w(x, y) =1√2πσ

e−|x−y|2

2σ2 (2)

(a) Original range image (b) Gaussian filter

(c) Bilateral filter (d) Proposed filter

Fig. 1. Experimental results for a complex environment

Wherefi anddi are the range and reflectance value in pixelxi,andw(xi, xj), w(fi, fj), andw(di, dj) are Gaussian functionsin the spatial, range, and reflectance domains with standardvariations ofσx,σf,andσd, respectively.

III. EXPERIMENT

We applied the proposed method to the data acquired byCPS-V[1]. Figure 1(a) shows original data, and the results ofdenoising by Gaussian filter, Bilateral filter and the proposededge-preserving smoothing filter are also shown in Figure 1(b),(c) and (d). Gaussian and Bilateral filter enable to smooth thesurface of walls, and especially Bilateral filter can preserve theshape of a monitor (the arrow). However, the roof edges ofwalls, the face of a person, and the wrinkles of the clothesare also becoming dull. On the other hand, we can say thatproposed edge-preserving smoothing filter smoothes the rangeimage successfully while preserving the jump and roof edgesappropriately.

IV. CONCLUSION

This paper proposed the new edge-preserving smoothingfilter for range images using reflectance values acquired as theby-product of the range value with Time-of-Flight sensors.

REFERENCES

[1] Ryo Kurazume, Yusuke Noda, Yukihiro Tobata, Kai Lingemann, YumiIwashita, and Tsutomu Hasegawa, Laser-based Geometric Modeling usingCooperative Multiple Mobile Robots, in Proc. IEEE International Con-ference on Robotics and Automation, pp.3200-3205, May 12-17, 2009.