Multi-Sensor Fusion Adopted 2-D Laser Rangefinder and Camera for Pedestrian Detection

The purpose of this research is to detect pedestrians in an outdoor scene by a multi-sensor with a 2-D laser rangefinder (LRF) and a pin-hole camera. Our project used LRF information to detect the environment of vehicles ahead and collect the highly risky candidate LRF points. The data collected would be transformed to image frame and then the region of interest (ROI) was decided in an image plane. The ROI of a synthetic image would be adopted to detect pedestrians. Histograms of oriented gradients and local weighted pattern feature descriptors were employed later to describe the shape and textural information for pedestrians. At last, the linear support vector machine would be used as the classifier to recognize pedestrians. A car to pedestrian stationary scenario was selected, and the maximum speed of demonstrated vehicle was 30 kilometres per hour. The average detection rate of our proposed system was 99.06%.