Robust Perception and Control for Humanoid Robots in Unstructured Environments Using Vision

Chapter 4
Robust Light Stripe Scanning

Light stripe ranging is an active, non-contact surface measurement technique that has been studied for several decades. However, conventional scanners rely on the brightness of the light stripe exceeding all other features in the image for reliable results. This assumption is easily violated in robotic applications that require the scanner to operate in ambient light and deal with sensor noise, secondary reflections and cross-talk from other robots. Robust light stripe scanning methods have been proposed in previous work, but suffer from issues including assumed scene structure, lack of error recovery and acquisition delay. This thesis proposes an actively calibrated stereoscopic light stripe scanning technique that allows the stripe to be reliably identified in the presence of noise and distractions. The novel method exploits redundancy from stereo measurements and is optimal with respect to sensor noise.

Mirror Experiment: Interference Rejection (Section 4.6.1)

The performance of the robust stereoscopic scanner was compared with two other established methods: single-camera scanning, and double-camera scanning [Nakano et al., 1988; Trucco and Fisher, 1994]. The three methods were tested on a scene containing typical domestic objects and a mirror to simulate interference caused by cross-talk and specular reflections. To establish a fair comparison, the three methods were implemented to operate simultaneously on the same raw measurements during a single scan. Sub-sampled VRML models of the 3D colour/range data sets for each scanner (half and one-quarter the of the actual scanner resolution) are presented below. MPEG "fly-thoughs" allow the models to be viewed without a VRML browser.

Single-camera scanner result (Figure 4.9(a))

Double-camera scanner result (Figure 4.9(b))

Robust scanner result (Figure 4.9(c))