Road green belt is an important part of urban green-land system, which not only can beautify the urban environment, but also has the function of organizing and maintaining urban traffic as well as other eco-environmental services. Fine classification and extraction of road green belt information and the dynamic analysis of green belts are of great significance to road management. Based on the vehicle LiDAR technology, this study proposd a algorithm for automatic extraction and fine classification of road green belts. To verify the effectiveness of the algorithm, a road section of Fengtai District, Beijing was selected as the experimental area. The data collection time of the first and second phase tests were June 2015 and September 2015, respectively. The vehicle LiDAR points cloud data were taking as the original data. To improve the speed of the algorithm, data points within a certain range on both sides of the road were reserved according to the GNSS track line. After removing the distant points cloud data and compressing the data amount, the reserved points cloud was preprocessed by clipping and partitioning. Firstly, ground, low ground features, and high ground features were classified for each section of the road points cloud data, and then low ground features and ground points were combined. Secondly, the green belts in each segment of the points cloud data were extracted according to the points cloud features and spatial characteristics of the green belt, and the extracted green belts were identified twice to improve extraction accuracy. Based on the extracted green belts, the classification scope was determined. Based on the different characteristics of the points clouds of various ground features, the high and low ground features in green belts were classified in detail. Finally, the data of multiple green belts in the same area were compared, so as to determine whether the green belt area and the types and quantities of various ground features in the green belts have changed, which provided data support to the garden and city management departments. To verify the accuracy of the proposed algorithm, the green belts were extracted by means of manual interaction, and all kinds of ground objects in the green belts were manually classified. Using these as reference, the artificial statistical information was compared with the automatically extracted green belts and the information of each classified ground features. The total green belt areas extracted by manual and automatic extraction in the experimental area were 13 027 and 12 749 m ², respectively, with a difference of 278 m ² between the two groups of data and a relative error of 0.02. In the scene of the experimental area, the detectivity of pole-like objects, trees, and shrubs by the automatic classification algorithm were 83.52%, 81.81%, and 73.91%, respectively. By comparing the two phases of green belt data, it was found that the area was reduced by 129.5 m ², and three new shrubs were added. Our experimental findings suggest the high accuracy of the proposed algorithm.