Pedestrian Detection

Pedestrian Detection


Antonio M. López, David Gerónimo, Javier Marín, David Vázquez, Yainuvis Socarrás, Alejandro González, Yiaolong Xu.

Associated Projects
Spanish Ministry of Education and Science (MEC) Research Projects TRA2004-06702/AUT,  TRA2007-62526/AUT, TRA2011-29454-C03-00.

Pedestrian accidents are the second source of traffic injuries and fatalities in the European Union. In this sense, advanced driver assistance systems (ADAS), and specifically pedestrian protection systems (PPS), have become an important field of research to improve traffic safety. Of course, in order to avoid collisions with pedestrians they must be detected, being camera sensors key due to the rich amount of cues and high resolution they provide.

Currently there are two main lines of work, one based on images of the visible spectrum, and the other, mainly motivated by nighttime, based on thermal infrared. The former has accumulated more literature because the easier availability of either CCD or CMOS sensors working in the visible spectrum, their cheaper price, better signal-to-noise ratio and resolution, and because most of the accidents happen at daytime. At the moment, we work with this visible spectrum images.

In this context, difficulties of the pedestrian detection task for PPS arise both from working with a mobile platform in an outdoor scenario, recurrent challenge in all ADAS applications, and from dealing with a so aspect-changing class like pedestrians. Difficulties can be summarized in the followings:

  • targets have a very high intra-class variability (e.g., clothes, illumination, distance, size, etc.);
  • background can be cluttered and change in milliseconds;
  • targets and camera usually follow different unknown movements;
  • and fast system reaction together with a very robust response is required.

The following figures correspond to the proposed pedestrian detection architecture we propose. It is divided in six different steps, each one with its own responsibilities inside the system. At the moment we just address the first three ones, i.e., pre-processing, foreground segmentation and classification. The center figure illustrates the plane computation technique which is aimed at defining a set of pedestrian-sized windows laying on the 3D ground. The idea is to estimate the road pitch orientation related to the camera by using RANSAC on the 3D stereo points Y-Z projection. The right one is the classification stage, where the selected windows are classified as pedestrian/non-pedestrian by using AdaBoost and two combined sets of features.


Color-based pedestrian detection

In this project we evaluate the opponent colors (OPP) space as a biologically inspired alternative for human detection. In particular, by feeding OPP space in the baseline framework of Dalal et al. for human detection (based on RGB, HOG and linear SVM) or in the framework of Part Base Model for object detection of Pedro Felzenszwalb et al., we will obtain better detection performance than by using RGB space.

Camera-based person detection is of paramount interest due to its potential applications. The task is diffcult because the great variety of backgrounds (scenarios, illumination) in which persons are present, as well as their intra-class variability (pose, clothe, occlusion). In fact, the class person is one of the included in the popular PASCAL visual object classes (VOC) challenge. A breakthrough for this challenge, regarding person detection, is due to Felzenszwalb et al. These authors proposed a part-based detector that relies on histograms of oriented gradients (HOG) and latent support vector machines (LatSVM) to learn a model of the whole human body and its constitutive parts, as well as their relative position. Since the approach of Felzenszwalb et al. appeared new variants have been proposed, usually giving rise to more complex models. We focus on an issue that has not attracted suficient interest up to now. In particular, we refer to the fact that HOG is usually computed from RGB color space, but other possibilities exist and deserve the corresponding investigation. In this paper we challenge RGB space with the opponent color space (OPP), which is inspired in the human vision system.We will compute the HOG on top of OPP, then we train and test the part-based human classifer by Felzenszwalb et al. using PASCAL VOC challenge protocols and person database. Our experiments demonstrate that OPP outperforms RGB. We also investigate possible differences among types of scenarios: indoor, urban and countryside. Interestingly, our experiments suggest that the beneficts of OPP with respect to RGB mainly come for indoor and countryside scenarios, those in which the human visual system was designed by evolution.

Synthetic pedestrian models
Check associated research line

Pedestrian segmentation
Information to be uploaded

Part-based pedestrian detection
Information to be uploaded

Motion-based pedestrian features
Information to be uploaded


Associated Publications