In this video, we propose a classification algorithm of road surface conditions using LiDAR with spatio-temporal information. In general, the camera vision system is widely utilized to detect road conditions. However, the camera sensor might have low accuracy due to environmental effects such as illumination changes or shadows of objects. To solve this problem, we utilize reflectivity of the road surface from LiDAR sensor so that the robustness could be obtained against misclassification from the vision system. For designing feature vectors, the reflectivity and total data points number from LiDAR, and vehicle speed from the in-vehicle sensor are selected. We propose the spatio-temporal structure for the Deep Neural
Network(DNN). The front road is divided into four regions to utilize the spatial information according to the regulation of the road. Then the data is stacked for temporal information by using the time windowing method. The final result is obtained by utilizing the spatio-temporal information with the
DNN result of each region. To validate the effectiveness of the proposed method, we compared it with three other algorithms, and we obtained the highest accuracy of 97.8%, 98.7%, and the lowest risk situation of 2.3%, 1.6%.

свернуть