3D object detection using point cloud is an essential task for autonomous driving. With the development of infrastructures, roadside perception can extend the view range of the autonomous vehicles through communication technology. Computation time and power consumption are two main concerns when deploying object detection tasks, and a light-weighted detection model applied in an embedded system is a convenient solution for both roadside and vehicleside. In this study, a 3D Point cLoud Object deTection (PLOT) network is proposed to reduce heavy computing and ensure real-time object detection performance in an embedded system. First, a bird’s eye view representation of the point cloud is calculated using pillar-based encoding method. Then a cross-stage partial network-based backbone and a feature pyramid network-based neck are implemented to generate the high-dimensional feature map. Finally, a multioutput head using a shared convolutional layer is attached to predict classes, bounding boxes, and the orientations of the objects at the same time. Extensive experiments using the Waymo Open Dataset and our own dataset are conducted to demonstrate the accuracy and efficiency of the proposed method.

It is a challenging task for a human operator to manipulate a robot from a remote distance, especially in an unknown environment. Excellent teleoperation provides the human operator with a sense of telepresence, mainly including real-world vision, haptic perception, etc. This paper presents a novel virtual environment building method using the red–green–blue (RGB) colour information, the surface normal feature-based 3D-point-cloud registration method and the weighted sliding-average least-square-method-based real-world dynamic modelling for teleoperation. The experiments prove the method to be an accurate and effective means of teleoperation.

Three-dimensional (3D) data processing has applications in solving complex tasks such as object recognition, environment modeling, and robotic mapping and localization. Because using raw 3D data without preprocessing is very time-consuming, extraction of geometric features that describe the environment concisely is essential. In this sense, a plane can be a suitable geometric feature due to its simplicity of extraction and the abundance in indoor environments. This paper presents an online incremental plane extraction method using line segments for indoor environments. Our data collection system is based on a “nodding” laser scanner, so we exploit the incremental nature of its data acquisition in which physical rotation and 3D data processing are conducted in parallel. Line segments defined by two end points become supporting elements that comprise a plane, so a large proportion of scan points can be ignored once the line segments are extracted from each scan slice. This elimination of points reduces the algorithm complexity and computation time. Experiments with the tens of complete scan data sets which were acquired from a typical indoor environment demonstrated that our method was at least three times faster than the state-of-the-art methods.