Lidar-based classification and detection system for drivable area on roads

Separating the drivable and non- drivable areas on semi-structured and unstructured roads is an important task for autonomous vehicles to safely and avoid obstacles. Semi structured and unstructured roads have different intensities, normal vector angles, and curvature information than the background, and this paves the way for the design and development of an efficient detection system for drivable areas on this roads. In this paper, an effective method for detecting drivable areas is proposed that is based on important indicators of an experimental vehicles. This method calculate the information gain of features is calculated firstly to determine the sequence of feature processing. On the basis of this sequence calculate the maximum inter-class variance of features, and combined with the specific indicators of the experimental vehicle to realize the detection of drivable areas. Finally, the performance of the method is evaluated in terms of average precision, recall, and detection accuracy, and compared with the performance of existing road detection methods, including the K-nearest-neighbors classifier and the random forest classifier methods. The experimental results show that the average precision, recall, and detection accuracy of the system are 96.19%, 96.89%, and 96.72%, respectively. The method proposed here can effectively identify and classify drivable areas on semi structured and unstructured roads.