Classifier fusion for outdoor obstacle detection

Abstract

This paper describes an approach for using several levels of data fusion in the domain of autonomous off-road navigation. We are focusing on outdoor obstacle detection, and we present techniques that leverage on data fusion and machine learning for increasing the reliability of obstacle detection systems. We are combining color and infrared (IR) imagery with range information from a laser range finder. We show that in addition to fusing data at the pixel level, performing high level classifier fusion is beneficial in our domain. Our general approach is to use machine learning techniques for automatically deriving effective models of the classes of interest (obstacle and nonobstacle for example). We train classifiers on different subsets of the features we extract from our sensor suite and show how different classifier fusion schemes can be applied for obtaining a multiple classifier system that is more robust than any of the classifiers presented as input. We present experimental results we obtained on data collected with both the eXperimental Unmanned Vehicle (XUV) and a CMU developed robotic tractor.

A study of visual and tactile terrain classification and classifier fusion for planetary exploration rovers

“…Kelly et al has demonstrated the effectiveness of multispectral imaging, specifically at the near-infrared range, for terrain classification (Kelly et al, 2004). Dima et al has employed the LUV color space and computed distribution statistics of each channel over ground patches as color features (Dima et al, 2004).…”

“…Ramussen (2001) used Gabor filters to detect "denseness" of textured surfaces to distinguish roads from surrounding vegetation. Dima employed a Fast Fourier Transform representation of terrain surfaces for texture feature extraction (Dima et al, 2004). Angelova utilized a histogram-based method where terrain classes are represented by textons and terrain patches are identified through occurrence statistics of these textons .…”

“…Meta-classifier fusion is very similar to stacked generalization (SG) presented by Wolpert (1992) and applied for road detection in (Dima et al, 2004). In our method, low level classifiers described in Section 2 correspond to the "level-0 generalizer" where meta-classifier corresponds to "level-1 generalizer" of SG architecture.…”

Vision-based Motion Planning for an Autonomous Motorcycle on Ill-Structured Road

“…Due to the inherent difficulties in understanding natural objects and changing environments, autonomous driving is still in its infancy. However, existing results such as motion planning with 3D vision and the use of multiple classifiers [10], [14] shed light on a different class of problems, where roads do not disappear completely.…”

“…Motion blurring and vibration caused by a fast moving vehicle further degrade image quality. To address these issues, researchers approach the problem using different strategies such as color vision [10], [16], prior knowledge [6], pixel voting [15], classifier fusion [14], optical flow [21], neural networks [3], and machine learning [20], [21].…”