COMBINING DISTANCE AND MODULATION INFORMATION FOR DETECTING PEDESTRIANS IN OUTDOOR ENVIRONMENT USING A PMD CAMERA

摘要

This paper describes a pedestrian detection system which integrates modulation amplitude information with distance information, received from a PMD camera. A computationally efficient approach is proposed to achieve a relatively accurate depth image that enables some 3D reconstruction and real-time object detection methods in outdoor environment, which were not feasible previously. For that aim and in order to eliminate noise, intensity information is used, applying a rectangular low-pass filter. The outcome is given in form of regions of interest, containing query objects, which gives a way to approximate real distances from far objects to the PMD camera. Alternatively, we introduce a new key feature, so called Haar Wavelet of Gabor Filter (HWoGF), extracted from each region of interest, which is used as the discriminate feature for the Support Vector Machine training procedure. Finally, the paper also presents a mapping technique between 2D and 3D images, which allows overcoming the limitations of the low resolution imaging 3D sensor system.