A Navigation and Guidance System for Autonomous Flights of MAVs into Buildings

Manuel Popp, Silvia Prophet, Georg Scholz, Gert F. Trommer

Abstract: In the event of accidents or natural disasters fast reconnaissance for the purpose of disaster control is essential. In order to avoid unnecessary risks for the rescue forces it is desirable to use autonomous vehicles to explore hazardous areas. Micro Aerial Vehicles (MAVs) with hovering capabilities and high maneuverability are predestined for reconnaissance missions in urban terrain especially inside buildings. However, the conditions for the autonomous access to buildings through windows, doors or breaches in the wall are highly demanding. The guidance has to achieve a save and collision free flight through the passageway into the building. Besides that it has to maneuver the MAV in such a way that the target never gets out of sight of the MAV’s detecting sensors. In order to enable the MAV to accurately follow the guidance, the presence of a precise and reliable navigation solution is vital for the flight control. However, close to buildings the GNSS signals are not always available or the information is corrupted due to multipath and shading. Thus, additional sensors are needed to assist the inertial navigation system. Furthermore, the available computational power is limited because of the restricted payload of the MAV. The actual building approach flight and the subsequent flight into the building are rarely addressed in literature. Commonly, GPS measurements, external cameras that are observing the MAV and the target or maps are used to assist the target approach guidance. However, the usage of external cameras in real life conditions is not entirely possible. In addition there commonly exists no prior knowledge about the building structure, potential obstacles or about the position of the entrances. Therefore, the guidance we are presenting in this paper is not depending on such resources. Instead, we are using a combination of an image based guidance technique, called visual servoing and artificial potential fields to guide the MAV through a building’s window. Before the approach flight starts, the target has to be detected. As long as the MAV is still far away from the target, the chances are higher to detect the target in the camera image than to detect it in the laser scan. Thus, at first the target is detected in the camera image. Therefore, advantage is taken of the distinctive square shape of a regular window. To find the right window among all visible windows, the operator has to choose the target either in the camera live image, or in an aerial image that was taken beforehand. For the latter case a wide baseline correspondence finder was developed and will be presented in this paper. As soon as the window is detected an optical flow algorithm is used to track the striking corners of the window in the camera image. The information of the window’s position in the image is subsequently used to assist the laser based target detection. Thus, the position of the window is transformed from the image into the laser rangefinder’s scan to reduce the search area. The detection of the passageway in the laser range finder’s scan is done using an adaptive line extraction algorithm. The whole approach and fly-through can be divided into two phases. During the first phase, the information about the position of the window corners in the image is used by a visual servoing guidance to maneuver the MAV towards the target window. Our hybrid visual servoing guidance is combining an image and a position scheme in order to take advantage of their complementary properties. As soon as the MAV is close enough to the window to detect it in the laser scan, the second phase of the approach is starting. With the help of the laser rangefinder’s measurements and the camera image, a local artificial potential field is set up. Walls and other obstacles are represented by repulsive potentials in the form of radial defenders, the frame of the target window is described by an elliptic rotating potential. Whereas the repulsive potentials are keeping the MAV away from obstacles, the rotating potential is pulling the MAV through the entrance. Additionally a predictive vector field algorithm is used to further improve the accuracy and robustness of the guidance system and to avoid local minima in the potential field. The resulting gradient of the potential field serves as the desired velocity of the MAV and is controlled by a PID controller. In order to calculate the controller error, the navigation solution is used. Commonly applied multi sensor navigation systems are using laser range sensors, stereo or monocular cameras to substitute GNSS measurements. Especially in indoor areas with short distances to walls and other objects, laser range scanners are predestined to aid the navigation filters. Outside of buildings, measurable objects are not permanently available within the scan area of the laser range scanner. In this case, the laser range measurements are of limited use. Another problem is the different appearance of the indoor and the outdoor area, which leads to a high variety of object distances and scaling. Therefore stereo cameras with fixed baselines are also not suitable for our purpose due to their limited depth range. In order to determine an accurate and precise navigation solution for both, indoor and outdoor area, we presented a navigation system on the ION GNSS+ 2014. This integrated navigation system is combining a laser range finder and a dual camera system with non-overlapping view. Relative positon and attitude measurements are calculated with the help of laser based scan matching, by decomposing Homography and Essential matrix and with the help of vanishing points, detected in camera images. In this paper we are presenting additional usage of the relative pose between the window and the MAV to correct the navigation solution and to further increase the robustness of the navigation filter. The MAV, we are using for our experiments is a quadrocopter with a diameter of 0.74 m. Thus it is small enough to fly through a window but it is able to carry the necessary payload. In order to calculate the algorithms in real time, a combination of a self-designed circuit board and an embedded computer is used. Whereas the circuit board is used to calculate a simple backup guidance and navigation solution, the embedded computer is used to calculate the navigation and guidance algorithms, presented in this paper. In the end, both simulated and real MAV flights are used to experimentally evaluate the guidance and navigation system. Altogether a novel guidance and navigation system was developed to enable reconnaissance flights in urban terrain by maneuvering the MAV inside a building. Therefore a robust navigation system is presented for both indoor and outdoor areas enabling outdoor - indoor transition flights. A target detection and tracking system is providing the guidance with information about the target’s position. In order to maneuver the MAV in a way to prevent losing the target out of the camera image and to avoid any collisions, a combination of an image based and a potential field based guidance is presented.
Published in: Proceedings of the 28th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2015)
September 14 - 18, 2015
Tampa Convention Center
Tampa, Florida
Pages: 784 - 816
Cite this article: Popp, Manuel, Prophet, Silvia, Scholz, Georg, Trommer, Gert F., "A Navigation and Guidance System for Autonomous Flights of MAVs into Buildings," Proceedings of the 28th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2015), Tampa, Florida, September 2015, pp. 784-816.
Full Paper: ION Members/Non-Members: 1 Download Credit
Sign In