Frustum-LDGCNN: A High Efficient 3D Object Detection Network with Frustum Proposal

In the field of autonomous driving, the object detection task faces complex and diverse scenes during the driving process. Depending solely on an RGB image for object detection lacks security and stability. Therefore, this paper proposes a object detection method based on the fusion of RGB images and point clouds. This paper combines frustum fusion method with the Link Dynamic Graph Convolutional Neural Network (LDGCNN) to propose a Frustum-LDGCNN 3D target detection method. LDGCNN can learn the spatial features between point clouds, which allows the network to more accurately locate object. In order to verify the performance of the algorithm, multiple comparative experiments were conducted on the KITTI dataset. Compared with other 3D object detection method, Frustum-LDGCNN has batter performance on the detection accuracy of cars, pedestrians, and bicycles.