Natalie Frietsch

Vision-based attitude estimation for indoor navigation using Vanishing Points and lines

A novel method for vision-based indoor attitude aiding is described in this paper. A strapdown Inertial Navigation Systems (INS) with low-cost Micro-Electrical Micro-Mechanical (MEMS) sensors is augmented by a mono-camera. Line features are detected and analyzed using the concept of Vanishing Points (VPs) and Vanishing Lines (VLs). By exploiting geometrical constraints inside buildings, long-term stable attitude information is extracted. The performance of this aiding method is analyzed in turntable experiments and demonstrated in field test within an Integrated Pedestrian Navigation System (IPNS). It is shown that the proposed method effectively eliminates attitude drift in indoor environments and is real-time capable.

Adaptive path planning for VTOL-UAVs

This describes the development of path planning algorithms of a small unmanned four-rotor helicopter. A powerful simulation environment of the whole UAV system - including the characteristics of the important ranging sensors for collision avoidance was developed. This is essential for developing, testing, and verifying of the algorithms. Different collision avoidance strategies for VTOL-UAVs are presented. Enhancements and miniaturization will offer more powerful sensor technologies regarding size, range, and power in the future. Very promising are improvements of sensor modules and new technologies like three-dimensional LASER range-finder, PMD sensors, RADAR range-finder, and stereo camera tracking system. Because of the general high level simulation tool introduced herein they can be easily validated and tested without the need and effort of a real hardware implementation. The results showed that adaptive path planning, including collision avoidance, is already applicable on-board small UAV vehicles. With the mentioned new sensor technologies and more calculation power, further improvements like advanced collision notice and global path planning on-board small UAVs are attainable.

Development of a GPS/INS/MAG navigation system and waypoint navigator for a VTOL UAV

Unmanned aerial vehicles (UAV) can be used for versatile surveillance and reconnaissance missions. If a UAV is capable of flying automatically on a predefined path the range of possible applications is widened significantly. This paper addresses the development of the integrated GPS/INS/MAG navigation system and a waypoint navigator for a small vertical take-off and landing (VTOL) unmanned four-rotor helicopter with a take-off weight below 1 kg. The core of the navigation system consists of low cost inertial sensors which are continuously aided with GPS, magnetometer compass, and a barometric height information. Due to the fact, that the yaw angle becomes unobservable during hovering flight, the integration with a magnetic compass is mandatory. This integration must be robust with respect to errors caused by the terrestrial magnetic field deviation and interferences from surrounding electronic devices as well as ferrite metals. The described integration concept with a Kalman filter overcomes the problem that erroneous magnetic measurements yield to an attitude error in the roll and pitch axis. The algorithm provides long-term stable navigation information even during GPS outages which is mandatory for the flight control of the UAV. In the second part of the paper the guidance algorithms are discussed in detail. These algorithms allow the UAV to operate in a semi-autonomous mode position hold as well an complete autonomous waypoint mode. In the position hold mode the helicopter maintains its position regardless of wind disturbances which ease the pilot job during hold-and-stare missions. The autonomous waypoint navigator enable the flight outside the range of vision and beyond the range of the radio link. Flight test results of the implemented modes of operation are shown.

Adaptive path planning for a VTOL-UAV

Unmanned aerial vehicles (UAV) can be used for versatile surveillance and reconnaissance missions. If a UAV is capable of flying automatically on a predefined path, the range of possible applications is widened significantly. This paper addresses the development of adaptive path planning algorithms for a small vertical take-off and landing (VTOL) unmanned four-rotor helicopter with a take-off weight below 1 kg. Because of the light weight and the small size of less than 1 m makes the use of compact and efficient sensor technology as well as small computer platforms is mandatory. The path planning for the UAV is processed in different phases. The global preflight planning phase calculates an optimized trajectory in consideration of boundaries. Afterwards, during the flight phase on-board ranging sensors are used to avoid interferences with unknown obstacles. The paper shows the details of the developed algorithms and the simulation framework allowing a verification and validation of the algorithms.

Teaming of an UGV with a VTOL-UAV in urban environments

This paper focuses on the teaming of a small unmanned ground vehicle (UGV) with an unmanned micro serial vehicle (MAV) in order to improve the navigation solution of the UGV e. g. in the case of GPS loss. The image data that is acquired by the on-board camera of the MAV is processed in order to detect the UGV marked by a colored pattern. After the identification of the UGVpsilas location in an image, the location of the UGV with respect to a geo-coordinate system and its yaw angle are estimated by taking the MAVpsilas position and attitude into account. Finally, this information is processed in the system on-board the UGV to improve the navigation solution. The performances of the particular parts of the setup and also of the whole system are illustrated with experimental data.

Development of a method for image-based motion estimation of a VTOL-MAV on FPGA

In this paper, the development of the vision based motion estimation for a small-scale VTOL-MAV as well as the implementation on FPGA are investigated. Especially in urban environments the GPS signal quality is disturbed by shading and multipath propagation and an augmentation with another sensor is inevitable. The vision system bases on the analysis of the sparse optical flow that is extracted from images taken by the onboard camera. From the extracted point correspondences projective transformations are estimated with a robust parameter estimation algorithm. As the underlying image processing routines are computationally expensive but can be processed in parallel they have been implemented on FPGA. The different parts of the algorithm as well as the implementation are covered in detail.

Improved object geo-location in airborne camera images using tight integration of vision and navigation data

A method for precise geo-location of objects that are observed by an airborne camera is described in this paper. The platform for image acquisition is a micro aerial vehicle (MAV) with an integrated navigation system. From the captured image sequences and MAV navigation data, the three-dimensional positions of objects of interest are retrieved. Different techniques for image feature tracking are compared. Combining measurements from multiple viewpoints in a Bundle Adjustment process yields optimal accuracy of the estimated object positions. The robustness of the optimization is enhanced by tight integration of data from both the vision and the navigation system.

Real Time Implementation of a Vision-Based UAV Detection and Tracking System for UAV-Navigation Aiding

This paper focuses on the real time implementation of cooperative navigation aiding for a small unmanned aerial vehicle (UAV) based on vision systems on board unmanned ground vehicles (UGV). In urban environments the signal quality of global satellite navigation systems (GNSS) often is bad or signals are even lost, so that the navigation solution of a UAV is aected. Especially in such situations the UAV’s geo-referenced position has to be known to ensure a safe guidance. A team of UGVs can overcome the problem of no GNSS position solution by the detection and tracking of the UAV in its on board images and the subsequent geo-localization. The determination of the UAV’s geo-referenced position can be achieved with several algorithms, depending on the number of UGVs. To integrate these position measurements in the UAV’s on board navigation computer the quality of the detection and geo-localization has to be estimated. All parts are implemented on embedded PC platforms. To allow real time usage on board the UGVs special algorithms have to be used for the most tasks, as image processing needs high computational power. The particular parts of this setup as well as the whole system are tested with experimental data.

Image based augmentation of an autonomous VTOL-MAV

In this paper, the development of a vision based system for a small-scale VTOL-MAV is presented. The on-board GPS/INS navigation system is augmented by further sensors in order to allow for an autonomous waypoint mode. Especially in urban environments the GPSsignal quality is disturbed by shading and multipath propagation. The investigated vision system based on algorithms analyzing the optical flow is essential to enable the helicopter to reliably hover even in these scenarios. Due to the integration of the vision based navigation information into the navigation filter, GPSsignal outages can be bridged. The necessary height above ground information is estimated from the relative altitude change given by the barometric altimeter and the optical flow.

Detection and tracking of objects in an image sequence captured by a VTOL-UAV

This paper focusses on the automated detection and tracking of moving objects in a camera sequence, that is provided by a small, electrically powered four-rotor helicopter in a hover-and-stare scenario. Two different algorithms for identifying independently moving areas are investigated and compared. The first approach bases on the previous compensation of the camera movement by estimation of homographies. Moving regions are extracted by robust background subtraction. The second approach bases on a dense optical flow field and needs no stabilization: Single points are identified that move not consistently with the background plane. These points are merged into objects by a cluster analysis algorithm. Furthermore, a strategy for tracking these objects over time is described including a Kalman filter. Due to several reasons, not every extracted area corresponds to an independently moving object and a heuristic rule set is used to sort artifacts out. Experimental results on in-flight images are presented and the performances of the developed algorithms are compared. Finally, first steps towards a geographic location of the tracked objects are described.