For highly automated autonomous driving, accurate and detailed perception of the vehicle’s surrounding environment is crucial. lidar is widely applied for sensing information about the surroundings. Addressing the issue of low accuracy in traditional lidar point cloud processing methods, this paper proposes an improved PointNet++ algorithm to construct a lidar point cloud obstacle detection network. The training strategy of the network is enhanced by employing point re-sampling and adding independent noise for data augmentation. The optimization algorithm is also improved, leading to a 2.5% increase in the average detection accuracy of the network model. To tackle the problem of significant information loss within the network, a residual structure is introduced into the PointNet++ network, enabling the network to retain information from the original input during the learning process. The proposed algorithm is validated on the KITTI dataset, and the results show that the improved network achieves a 4.5% higher detection accuracy compared to the original network.