Utilization of optimally selected features for car detection in calibrated camera and LRF system

This paper proposes a method for car detection in a calibrated system of camera and Laser Range Finder (LRF). The LRF sensor is used to extract the car candidates by a clustering method. Adjacent points sensed by the LRF are grouped together to form a car candidate. Those car candidates are then filtered out based on their length. Using the property of the calibrated camera and LRF system, region of interests (ROI) on the camera image are defined by the car candidates from LRF data. Histogram of Oriented Gradient (HOG) features are extracted from each ROI. A genetic algorithm (GA) based approach is performed to select the optimal subset of features. Finally, a machine learning based approach is performed to do the validation process. From the experiments, it is shown that the GA based approach enables feature size reduction up to 75% while maintaining the detection performance.