Overview of 3D Object Detection for Robot Environment Perception

Environmental perception is one of the key issues in robot research. To complete more precise tasks, robots need a recognition ability to recognize all dynamic and static objects in the environment accurately. Currently, the prevailing approach in robotics is to employ object detection methods for perceiving stationary objects in the robot's vicinity. This primarily entails the use of 2D object detection based on images and 3D object detection based on depth images or point cloud data. From an input data perspective, it is evident that depth images and point cloud data offer a substantially richer set of three-dimensional spatial information compared to conventional image data. Therefore, 3D object detection is more suitable for robot environmental perception. This article provides an overview of 3D object detection methods for facial robots. This article mainly reviews methods based on radar point cloud information, introduces the historical evolution of 3D object detection, and elaborates on the performance and limitations of different methods. At the same time, it lists commonly used datasets for related tasks.