3D DENSITY AND DENSITY MAPS FOR STEREO VISION-BASED NAVIGATION

摘要

Range data sensors for machine perception play an important role in allowing autonomous vehicles to detect their surroundings. Local information provides the level of accuracy and reliability needed for safe autonomous operations Among range data sensors, stereoscopic vision cameras (stereovision) offer rich visual perception in a three-dimensional format that is convenient for obstacle detection and safeguarding. This work describes several processing steps to extract the key information from stereo images and determine a vehicle's optimum path. The key concept conceived and developed for such purpose is 3D density and its practical application in density grids. A stereovision engine was implemented on a utility vehicle to detect the presence of obstacles and output a navigation map to be processed by a path-planner. The vehicle was capable of reaching a desired target point avoiding obstacles in the local environment. Results demonstrated the suitability of stereo for obstacle detection and autonomous navigation assistance.