An Improved and Enhanced Depth Map Approach for Obstacle Detection in Unknown Environments for Reactive Autonomous Mobile Robots
Vomsheendhur Raju, Majura F. Selekwa · Volume 5: Dynamics, Vibration, and Control · 2024
Abstract Interest in autonomous vehicles has attracted research into developing better navigation systems. Autonomous navigation systems guide unmanned vehicles to traverse any environment without human intervention; they require clear knowledge of the environment to make correct decisions. Information about the environment can be obtained from either ranging or vision sensors. Although ranging sensors such as LiDARs, radars, or sonars can detect the presence and location of obstacles, they don’t provide any information on whether the obstacles are traversable. On the other hand, vision sensors produce images similar to how human eyes perceive the environment, which can be processed to determine whether the sensed obstacle is traversable. Since vision sensors are light, less costly, and consume low power compared to LiDARs or Radars, there is an increasing interest in using them in autonomous navigation of unmanned ground vehicles. The main limitation of vision sensors for this purpose is their failure to directly identify the locations of the obstacles. Often, the location is determined by using a depth map, which is not always accurate and can be very computational. This paper proposes a new method of processing a depth map with improved accuracy. The depth of the closest objects to the stereo camera is extracted using the horizontal and vertical field of view of the sensor. It is further enhanced using morphological operations and the P-Tile segmentation technique to segment the obstacle(s) from its background and noise. This approach generates a denser map without any discontinuities, which leads to improved accuracy in locating close-by objects at less computational cost. Information from this depth map can be used as a low-cost alternative to ranging information in the navigation system for ground vehicles.