Stereo- and neural network-based pedestrian detection

Abstract

Abstract-Pedestrian detection is essential to avoid dangerous traffic situations. In this paper, we present a fast and robust algorithm for detecting pedestrians in a cluttered scene from a pair of moving cameras. This is achieved through stereo-based segmentation and neural network-based recognition. The algorithm includes three steps. First, we segment the image into sub-image object candidates using disparities discontinuity. Second, we merge and split the sub-image object candidates into sub-images that satisfy pedestrian size and shape constrains. Third, we use intensity gradients of the candidate sub-images as input to a trained neural network for pedestrian recognition. The experiments on a large number of urban street scenes demonstrate that the proposed algorithm: 1) can detect pedestrians in various poses, shapes, sizes, clothing, and occlusion status; 2) runs in real-time; and 3) is robust to illumination and background changes.Index Terms-Driver assistance system, neural networks, object detection, pedestrian detection, range image segmentation, stereo vision.

Multi-cue Pedestrian Detection and Tracking from a Moving Vehicle

“…Table 1 summarizes the main pedestrian detection systems, distinguished by image sensor type, area of coverage, detection performance, use of tracking, processing speed and test set, where-ever specified by the respective authors. As we will discuss in Section 6, performance comparisons are hazardous, since data sets are typically small (and diverse) (Mohan et al, 2001), Papageorgiou and Poggio (2000), Viola et al (2005), Zhao and Thorpe (2000) and often relate to single computational components (e.g. classification).…”

“…The brute-force approach in combination with powerful classifiers (i.e. Mohan et al, 2001;Papageorgiou and Poggio, 2000;Zhao and Thorpe, 2000) is currently computationally too intensive for realtime application. Recently however, Viola et al (2005) demonstrated an efficient variant of the sliding window technique, which involves a detector cascade using simple appearance and motion filters (similar to the Haar-wavelets).…”

“…Zhao and Thorpe (2000) obtain a foreground region by clustering in the disparity space. Broggi et al (2003) and Grubb et al (2004) consider the x-and y-projections of the disparity space following the 'V-disparity' technique (Labayrade et al, 2002).…”

“…For example, Broggi et al (2004) employ vertical symmetry features. Zhao and Thorpe (2000) apply a high-pass filter and normalize the ROI for size, thereafter applying a feed-forward neural network. Papageorgiou and Poggio (2000) pioneered the use of Haar-wavelet features in combination with a Support Vector Machine (SVM); this approach was subsequently adapted by Elzein et al (2003) and others.…”

People Detection Using Color and Depth Images

“…For instance, Zhao and Thorpe [34] use a stereo system to segment the silhouettes that are fed to a neural network that detects pedestrians. Xu and Fujimora [32] also extract body silhouettes but with a time-of-flight device.…”