IR Pedestrian Detection for Advanced Driver Assistance Systems

Abstract

Abstract. This paper describes a system for pedestrian detection in infrared images implemented and tested on an experimental vehicle. A specific stabilization procedure is applied after image acquisition and before processing to cope with vehicle movements affecting the camera calibration. The localization of pedestrians is based on the search for warm symmetrical objects with specific size and aspect ratio. A set of filters is used to reduce false detections. The final validation process relies on the human shape's morphological characteristics.

Survey of Pedestrian Detection for Advanced Driver Assistance Systems

“…In the former case, the algorithms are mainly based on the study of visual features. In [20], [21], Broggi et al correct the vertical image position by relying on the detection of horizontal edges oscillations: The horizon line is computed according to the previous frames. A comparative study of different monocular camera pose estimation approaches has been presented in [22].…”

“…Miau et al [36], [37] use a biologically inspired attentional algorithm that selects ROIs according to color, intensity, and gradient orientation of pixels. In several works from Parma University, the vertical symmetries in the visible [38], [39], [40], [41] and TIR spectra are used alone [42], [20] or as a complement to stereoimaging [39]. In this case, ROIs are adjusted around each symmetry axis maintaining the PSC.…”

“…Some techniques that utilize TIR images are 2D model matching [20], 3D [42], [113] model matching, symmetry [114], and a multiple filter approach, based on the area overlap between positively classified ROIs and group multiple detections in a single window [51].…”

Stereo Vision Camera Pose Estimation for On-Board Applications

“…Notice that since the vision system is rigidly attached to the vehicle, camera pose and vehicle pose are indistinctly used through this work. In order to tackle urban scenarios, some monocular systems have been proposed to automatically compute camera pose by using the prior knowledge of the environment (e.g., Franke et al, 1998;Bertozzi et al, 2003a;Suttorp & Bücher, 2006). However, scene priorknowledge not always can help to solve problems, in particular when cluttered and changing environment are considered, since visual features are not always available.…”

A multi-resolution approach for infrared vision-based pedestrian detection

“…While the flat road assumption can be supposed to hold in the area close to the vehicle (up to 20 meters) even in presence of hills or bumps, in the faraway area (more than 20 meters) less confident results may be obtained. Therefore, vertical oscillations need to be compensated and a specific software image stabilizer able to reduce these effects has been developed [14].…”