Search Results

Autonomous vehicles are currently a subject of great interest and there is heavy research on creating and improving algorithms for detecting objects in their vicinity. A ROS-based deep learning approach has been developed to detect objects using point cloud data. With encoded raw light detection and ranging (LiDAR) and camera data, several basic statistics such as elevation and density are generated. The system leverages a simple and fast convolutional neural network (CNN) solution for object identification and localization classification and generation of a bounding box to detect vehicles, pedestrians and cyclists was developed. The system is implemented on an Nvidia Jetson TX2 embedded computing platform, the classification and location of the objects are determined by the neural network. Coordinates and other properties of the object are published on to various ROS topics which are then serviced by visualization and data handling routines.

People through many years of experience, have developed a great intuitive sense for navigation and spatial awareness. With this intuition people are able to apply a near rules based approach to their driving. With a transition to autonomous driving, these intuitive skills need to be taught to the system which makes perception is the most fundamental and critical task. One of the major challenges for autonomous vehicles is accurately knowing the position of the vehicle relative to the world frame. Currently, this is achieved by utilizing expensive sensors such as a differential GPS which provides centimeter accuracy, or by using computationally taxing algorithms to attempt to match live input data from LiDARs or cameras to previously recorded data or maps. Within this paper an algorithm and accompanying hardware stack is proposed to reduce the computational load on the localization of the robot relative to a prior map.