Submitted version of the thesis - Airlab, the Artificial Intelligence ...

More documents

Recommendations

Info

56 Chapter 5. Vision To calculate PR ball which is 3D homogeneous coordinates of the point in world, the transformation matrix camera to robot (TR C ) is used. P R ball = T R C ∗PC ball The second configuration is for the object on the ground. This case is simpler since it requires a 2 D to 2 D transformation. The transformation between camera and robot is used again and the calculation is the same as the previous configuration. The homography is defined for this case as: H = K ∗T C ⎡ ⎤ 1 0 0 ⎢ ⎢0 1 0 ⎥ R ∗⎢ ⎥ ⎣0 0 0⎦ 0 0 0 The third row of the matrix is all zeros since the we assume that the objectisatgroundwithZ=0, hencethisreducesthemappingfrom3Dto2D. To calculate the object position Pobject, we will use the inverse homography on the image point Pimage. Pobject = H -1 ∗Pimage. Pobject is then normalized. The resulting point is the distance of the robot to the target object. This information is later used to calculate the motor contributions. Among the mentioned approaches, we started with the first configuration (for the ball). The idea was to detect a ball, no matter whether it is on the ground or up in the air. We started by testing with a red ball, since the color red, blue, green were initially defined for the camera. The results were not sufficient to follow the ball for greater distances because of several effects. One of them is related with the camera resolution. Normally the camera can work with 160x120 resolution for the blob search. This does not cause a problem if the ball is close to the camera, but as the ball goes further from the camera, the ball represented in the image is getting smaller and smaller. Due to the low-cost optics, at the worst case the ball becomes 1 pixel in the picture. Using only this pixel, it is impossible to detect the ball. Even though, we can recognize the ball for some cases, in most of the cases we cannot. Due to the low-cost optics, it is possible to have several other 1 pixel size red blobs that are not our ball. In order to detect the ball, inside the blob search algorithm we need to eliminate the blobs that are not the ball we are searching.
5.1. Camera Calibration 57 Later, in order to solve this problem, we came up increasing the size of the ball. Again, this trial did not resulted as we expected, since when the ball gets further from the robot, the diameter information is not reliable. This results to calculate a wrong value of the ball position. In this trials, we realized also the effects of the white balance and exposure compensation for the color information. The camera tries to automatically adjust the color information as the ball gets further from the light source. Since the blob elimination is dependent on both the shape of the ball and the color of the blob, the information changed by the camera results to definition of wrong blob information. To clarify with an example, the ball shape is checked by the circularity information calculated by the camera, but the whole ball is not composed of the same color, since the light is affecting the color, so the center can be more bright or the corner more dark etc. The camera with the automatic setting of the white balance and exposure compensation tends to change the color in order to have a better picture, but by doing so the color temperature is changed. So the blob search eliminates the non-color points as well as the non-shape objects and detects the ball in another shape, such as an open rectangle. This whole change also results to a change of the diameter of the ball and the distance information is not calculated correctly. Thediameter of theball is a problematic issuedueto the low-cost optics. Instead, we decided to change our ball into a cylindric shape and assume that the object will always be on the ground. This reduces the mapping to a 2 D to 2 D mapping problem, since Z (the height) will be always zero. Instead of calculating the diameter of theball, we determinethedistance between the end of the image and the intersection point of the object with the ground. This distance is easier to measure, independently from the shape of the object, and will always change as the object’s position changes in the world. The mentioned mapping also makes easier the blob search, since it is not very important to detect the object clearly as a cylindric. If we can recognize some part of the object we can ensure that the object is more or less close to the calculated point. This also gave us the flexibility for the colorinformation, sincewedonothavetheobligationtodetectalltheobject, we can defineamore generic color code that is working for most of the cases. We used the second configuration in both camera positions. Even the internal camera parameters are always same in our camera, the external parameters are effected with a translation or rotation. Indeed, when the camera head position is changed, it results with the usage of a H matrix for each transformation.
Page 1:
POLITECNICO DI MILANO Corso di Laur
Page 5:
To my family...
Page 9:
Summary Aim of this project is the
Page 13 and 14:
Contents Sommario I Summary III Tha
Page 15 and 16:
List of Figures 2.1 The standard wh
Page 17 and 18:
Chapter 1 Introduction 1.1 Goals Th
Page 19 and 20:
1.4. Thesis Structure 3 tailed desc
Page 21 and 22: Chapter 2 State of the Art Advances
Page 23 and 24: 2.1. Locomotion 7 provides low resi
Page 25 and 26: 2.1. Locomotion 9 Figure 2.4: Diffe
Page 27 and 28: 2.1. Locomotion 11 Omnidirectional
Page 29 and 30: 2.2. Motion Models 13 2.2 Motion Mo
Page 31 and 32: 2.4. Games and Interaction 15 a 2.5
Page 33 and 34: 2.4. Games and Interaction 17 We in
Page 35 and 36: 2.4. Games and Interaction 19 lot o
Page 37 and 38: Chapter 3 Mechanical Construction W
Page 39 and 40: 3.2. Motors 23 The initial design o
Page 41 and 42: 3.2. Motors 25 Figure 3.3: The calc
Page 43 and 44: 3.3. Wheels 27 target surface, maxi
Page 45 and 46: 3.5. Bumpers 29 Figure 3.8: Three w
Page 47 and 48: 3.6. Batteries 31 itself also works
Page 49 and 50: 3.7. Hardware Architecture 33 In sh
Page 51 and 52: 3.7. Hardware Architecture 35 The s
Page 53 and 54: 3.7. Hardware Architecture 37 With
Page 55 and 56: Chapter 4 Control The motion mechan
Page 57 and 58: 4.1. Wheel configuration 41 Figure
Page 59 and 60: 4.2. Matlab Script 43 This is case
Page 61 and 62: 4.2. Matlab Script 45 Mixed Angular
Page 63 and 64: 4.3. PWM Control 47 acollision, suc
Page 65 and 66: 4.3. PWM Control 49 mentioned previ
Page 67 and 68: Chapter 5 Vision The vision system
Page 69 and 70: 5.1. Camera Calibration 53 the prin
Page 71: 5.1. Camera Calibration 55 Figure 5
Page 75 and 76: 5.2. Color Definition 59 warmer (ye
Page 77 and 78: 5.2. Color Definition 61 Anotheraut
Page 79 and 80: 5.2. Color Definition 63 9000 8000
Page 81 and 82: 5.3. Tracking 65 any color values.
Page 83 and 84: Chapter 6 Game The robot will be us
Page 85 and 86: • Initialize sensors • Initiali
Page 87 and 88: If the camera is set to SERVO MID,
Page 89 and 90: Chapter 7 Conclusions and Future Wo
Page 91 and 92: ior, which will be useful in the re
Page 93 and 94: Bibliography [1] Battle bots-http:/
Page 95 and 96: BIBLIOGRAPHY 79 [25] Han et al. A n
Page 97 and 98: Appendix A Documentation of the pro
Page 99 and 100: Figure A.2: The flow diagram of the
Page 101 and 102: Appendix B Documentation of the pro
Page 103 and 104: B.1. Microprocessor Code 87 /* Conf
Page 105 and 106: B.1. Microprocessor Code 89 #ifdef
Page 107 and 108: B.1. Microprocessor Code 91 Set_Spe
Page 109 and 110: B.1. Microprocessor Code 93 TIM_DeI
Page 111 and 112: B.1. Microprocessor Code 95 /* Expo
Page 113 and 114: B.1. Microprocessor Code 97 RGBCube
Page 115 and 116: B.1. Microprocessor Code 99 /* Incl
Page 117 and 118: B.1. Microprocessor Code 101 #endif
Page 119 and 120: B.1. Microprocessor Code 103 } void
Page 121 and 122: B.1. Microprocessor Code 105 GPIO_I
Page 123 and 124:
B.1. Microprocessor Code 107 /* Sta
Page 125 and 126:
B.1. Microprocessor Code 109 } Inve
Page 127 and 128:
B.1. Microprocessor Code 111 #inclu
Page 129 and 130:
B.1. Microprocessor Code 113 /* * M
Page 131 and 132:
B.1. Microprocessor Code 115 b[1] =
Page 133 and 134:
B.1. Microprocessor Code 117 } * (B
Page 135 and 136:
B.1. Microprocessor Code 119 #defin
Page 137 and 138:
B.2. Color Histogram Calculator 121
Page 139 and 140:
B.3. Object’s Position Calculator
Page 141 and 142:
B.3. Object’s Position Calculator
Page 143 and 144:
B.4. Motion Simulator 127 0 0 0 0 0
Page 145 and 146:
B.4. Motion Simulator 129 if (plot
Page 147 and 148:
B.4. Motion Simulator 131 fill (m a
Page 149 and 150:
B.4. Motion Simulator 133 end text(
Page 151 and 152:
Appendix C User Manual This documen
Page 153 and 154:
C.1. Tool-chain Software 137 Figure
Page 155 and 156:
C.2. Setting up the environment (Qt
Page 157 and 158:
C.3. Main Software - Game software
Page 159 and 160:
Appendix D Datasheet The pin mappin
Page 161 and 162:
Figure D.2: The schematics for the
show all

Submitted version of the thesis - Airlab, the Artificial Intelligence ...

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

Delete template?

Save as template?