Navigation Functionalities for an Autonomous UAV Helicopter

More documents

Recommendations

Info

98 APPENDIX A. Fig. 1. The WITAS helicopter descend- Fig. 2. Landing pad with reference pating to the landing pad. tern seen from the on-board camera. proper trade-off between accuracy, range, latency, and rate has to be found optimizing the overall performance of the system. Our method requires a special landing pad (Fig. 2). As unmanned helicopters usually operate from a designated home base this is not a real constraint. A precise and fast pan/tilt camera is used to extend the range of the vision system and decouple the helicopter attitude from the vision field. We developed a single camera solution, as multi-camera systems make the system more complex and expensive and don’t offer significant advantages when using known landmarks. For the experimentation we used a helicopter platform (Fig. 1) which had been developed in the WITAS project [2,1]. Vision-based control of small size autonomous helicopters is an active area of research. A good overview of the state-of-the-art can be found in [7]. Our contribution to the landing problem consists of: (a) many demonstrated landings with a system only based on images from a single camera and inertial data using off-the-shelf computer hardware; (b) a wide envelope of starting points for the autonomous approach; (c) robustness to different weather conditions (wind, ambient light); (d) a quantitative evaluation of the vision system and the landing performance. 2 Vision System The vision system consists of a camera mounted on a pan/tilt unit (PTU), a computer for image processing, and a landing pad (a foldable plate) with a reference pattern on its surface. In this section, we explain the design of the reference pattern, describe the image formation, and present the image processing algorithm. The reference pattern is designed to fulfill the following criteria: fast recognition, accurate pose estimation for close and distant range, minimum size, and minimal asymmetry. We have chosen black circles on white background as they are fast to detect and provide accurate image features (Fig. 2). From
A.3. PAPER III 99 the projection of three circles lying on the corner points of an equilateral triangle the pose of an object is uniquely determined, assuming all intrinsic camera parameters are known. Circles are projected as ellipses, described by the center point ue, the semi-major axis la, the semi-minor axis lb, and the semi-major axis angle θe The pose of the landing pad with respect to the camera coordinate system is estimated by minimizing the reprojection error of the extracted center points and semi-axes of the three ellipses. We use five circle triplets of different size (radius 2 to 32 cm, distance 8 to 128 cm) with common center point to achieve a wide range of possible camera positions. Each triplet is uniquely determined by a combination of differently sized inner circles. A point p ˜x in the landing pad frame is projected on the image plane as follows: ũ = P p ⎛ ⎞ αu 0 u0 0 � cpR ˜x = ⎝ 0 αv v0 0 ⎠ 0 0 1 0 ctp 0T � p˜x 2 p 3 ũ ∈ P ˜x ∈ P 3 1 The extrinsic camera parameters are given by the three Euler angles of the rotation matrix c pR and the three components of the translation vector c tp. We use a camera model with the following intrinsic parameters: ”focal lengths” αu and αv in pixels, principal point (u0, v0), and four lens distortion coefficients . All intrinsic parameters are calibrated using Bouguet’s calibration toolbox [3]. A conic in P 2 is the locus of all points ũ satisfying the homogeneous quadratic equation ũ T C ũ = 0. The transformation of a circle Cp on the landing pad into an ellipse Ci in the image plane is given by[4]: Ci = (H −1 ) T Cp H −1 The homography matrix H is the projection matrix P without third column (z = 0). We calculate the ellipse center and axes from Ci and represent the parameters in a common feature vector c. Fig. 3 shows a data flow diagram of the vision system. Round-edged boxes represent image processing functions, sharp-edged boxes indicate independent processes, and dashed lines show trigger connections. Closed contours are extracted from gray-level images using a fast contour following algorithm with two parameters: edge strength and binarization threshold. The latter is calculated from the intensity distribution of the reference pattern. In the contour list we search for the three biggest ellipses belonging to a circle triplet. Ellipse parameters are estimated by minimizing the algebraic distance of undistorted contour points to the conic using SVD [4,6]. After having found three ellipses, the corresponding contours are resampled with sub-pixel accuracy. A coarse pose is estimated based on the ratio of semi-major axes (1) (2)
Page 1:
Linköping Studies in Science and T
Page 5:
Acknowledgements This work would ha
Page 9 and 10:
Contents 1 Introduction 1 2 Overvie
Page 11 and 12:
Chapter 1 Introduction An Unmanned
Page 13 and 14:
of the environment in order to avoi
Page 15 and 16:
Chapter 2 Overview This chapter pre
Page 17 and 18:
2.1. UAV SOFTWARE ARCHITECTURE 7 wh
Page 19 and 20:
2.2. THE UAV HELICOPTER PLATFORM 9
Page 21 and 22:
2.2. THE UAV HELICOPTER PLATFORM 11
Page 23 and 24:
Chapter 3 Simulation 3.1 Introducti
Page 25 and 26:
3.2. HARDWARE-IN-THE-LOOP SIMULATIO
Page 27 and 28:
3.3. REFERENCE FRAMES 17 3.3 Refere
Page 29 and 30:
3.4. THE AUGMENTED RMAX DYNAMIC MOD
Page 31 and 32:
3.4. THE AUGMENTED RMAX DYNAMIC MOD
Page 33 and 34:
3.5. SIMULATION RESULTS 23 3.5 Simu
Page 35 and 36:
3.5. SIMULATION RESULTS 25 Figure 3
Page 37 and 38:
3.5. SIMULATION RESULTS 27 Figure 3
Page 39 and 40:
Chapter 4 Path Following Control Mo
Page 41 and 42:
4.2. TRAJECTORY GENERATOR 31 Figure
Page 43 and 44:
4.2. TRAJECTORY GENERATOR 33 Fig. 4
Page 45 and 46:
4.2. TRAJECTORY GENERATOR 35 be use
Page 47 and 48:
4.2. TRAJECTORY GENERATOR 37 4.2.3
Page 49 and 50:
4.2. TRAJECTORY GENERATOR 39 radius
Page 51 and 52:
4.2. TRAJECTORY GENERATOR 41 Calcul
Page 53 and 54:
4.2. TRAJECTORY GENERATOR 43 Figure
Page 55 and 56:
4.2. TRAJECTORY GENERATOR 45 by the
Page 57 and 58: 4.2. TRAJECTORY GENERATOR 47 Figure
Page 59 and 60: 4.3. OUTER LOOP CONTROL EQUATIONS 4
Page 61 and 62: 4.4. EXPERIMENTAL RESULTS 51 Up [m]
Page 63 and 64: 4.5. CONCLUSIONS 53 1 + s 1 C(s) =
Page 65 and 66: 4.5. CONCLUSIONS 55 Vel [m/s] 20 18
Page 67 and 68: Chapter 5 Sensor fusion for vision
Page 69 and 70: composed of three accelerometers an
Page 71 and 72: 5.1. FILTER ARCHITECTURE 61 filter
Page 73 and 74: 5.1. FILTER ARCHITECTURE 63 north,
Page 75 and 76: 5.2. EXPERIMENTAL RESULTS 65 Figure
Page 77 and 78: 5.2. EXPERIMENTAL RESULTS 67 Figure
Page 79 and 80: 5.2. EXPERIMENTAL RESULTS 69 by the
Page 81 and 82: 5.3. CONCLUSION 71 vision system st
Page 83 and 84: BIBLIOGRAPHY 73 [9] E. Frazzoli. Ro
Page 85 and 86: Appendix A 75
Page 87 and 88: A.1. PAPER I 77 is a three-dimensio
Page 89 and 90: A.1. PAPER I 79 The reference point
Page 91 and 92: A.1. PAPER I 81 Path Av Err Max Err
Page 93 and 94: A.2. PAPER II 83 From Motion Planni
Page 107: A.3. PAPER III 97 A.3 Paper III Aut
Page 111 and 112: A.3. PAPER III 101 yes Tbo > Tlim2
Page 113 and 114: A.3. PAPER III 103 immediately afte
Page 115 and 116: A.3. PAPER III 105 altitude [m] 20
Page 117: A.3. PAPER III 107 Fig. 10. Flight
Page 121 and 122: Department of Computer and Informat
Page 123 and 124: No 598 Rego Granlund: C 3 Fire - A
Page 125 and 126: No 1024 Aleksandra Tešanovic: Towa
show all

Navigation Functionalities for an Autonomous UAV Helicopter

You also want an ePaper? Increase the reach of your titles

Delete template?

Save as template?