Framework of real-time car detection using calibrated camera and LRF

This paper proposes a framework for a real-time car detection method using calibrated system of camera and Laser Range Finder (LRF). Car candidates are extracted from the LRF data using a gridding method. The points sensed by LRF are grouped into 2D grid. Two adjacent occupied grid elements are marked with same label, forming an object. The objects formed by the labeling method are filtered out based on their size. A region of interest (ROI) in camera image is generated for each object located in 2D grid using the property of the calibrated camera and LRF system. From each ROI, Histogram of oriented gradient (HOG) features are extracted. In order to achieve a faster computation time, the dimension of the HOG feature is reduced using genetic algorithm approach, with a machine learning approach as the validation method. Experiments result shows that the proposed framework achieves around 68 fps of processing speed.