Process of Person following

By integrating the methods described previously, a process of person following is designed to detect an unexpected object in corridors and identifying the object as a

Figure 6.2 An example to illustrate the position of tracking target.

human or not. Furthermore, the human tracking process is performed after a human is detected. As illustrated in Figure 6.4, the algorithm of the person following process is described as follows.

Algorithm 6. Person following process.

Step 1. Perform the above-mentioned image analysis algorithm to each captured image, and compute the feature Object Foot in each navigation cycle.

Step 2. Perform the above-mentioned human identification algorithm when a set of Object Foots is collected.

Step 3. If an human is detected, then do the following steps.

a. Announce a warning message to the system if this human appears for the first time.

b. Calculate the steering angle θi according to the tracked target in each navigation cycle. If the angle is positive, turn the vehicle leftward for the angle; if the angle is negative, then turn the vehicle rightward for the angle.

c. Calculate the distance di between the tracked target and the vehicle, and if the distance is smaller than a predefined threshold, end the person following process and announce a message to the system.

d. Perform the fuzzy guidance process.

e. Repeat Step 1 through Step 3, until the human disappears in the eyesight of the vehicle.

Figure 6.3 A process of person following

Too small Fuzzy guidance

process

Chapter 7

Experimental Results and Discussions

7.1 Experimental Results

In this section, we will show certain navigation results in a complicated room environment. All experiments of this study were conducted in a laboratory in a building in National Chiao Tung University. In Section 7.1.1, a navigation result in the single path mode is shown, and in Section 7.1.2, a navigation result in the area mode is shown.

7.1.1 Navigation in Single Path Mode

In the learning stage, we controlled the vehicle to learn a path from one spot to another in our laboratory. The learned path is marked as a yellow dotted line and an example is shown in Figure 7.1. After the manual learning process, the proposed path map creation process was performed to create a path map with a set of nodes and directed edges. The set of nodes for the previous example are also shown in Figure 7.1, in which the directed edges (marked by blue or green arrows) are connected.

In the navigation stage, the vehicle begins to navigate along the learned path autonomously. Two kinds of navigation strategies are performed in corresponding navigation sections. Line following was performed in straight-line sections, which are

which are marked as green lines in Figure 7.1. And certain representative images are also shown in Figure 7.1 (numbered from one to sixteen), each of them showing the corresponding situation in the navigation session.

Besides, collision avoidance is another important capability of the proposed vehicle system, and also can be seen in this experimental result. In images (6) and (11) in Figure 7.1, two obstacles were put on the ground and obstructed the learned path in advance before navigation. Several suitable steering angles were calculated by the proposed fuzzy guidance techniques to avoid the obstacles.

The speed of the vehicle used in this navigation is set to be 10 cm/sec. Moreover, the computing time of each navigation cycle is about 0.5 seconds.

7.1.2 Navigation in Area Mode

At the learning stage, we first selected an initial node as a start point, and controlled the vehicle to move from the start point to two spots in our laboratory.

These two spots are defined as room 1 and room 2, respectively. The positions of the room nodes, the initial node, and the two learned paths are shown in Figure 7.2. The path map creation process was performed to create a map with a set of nodes and undirected edges.

At the navigation stage, a user command was given, which ordered the vehicle to move from room 1 to room 2 and back to room 1. At the beginning, the essential data generation process was performed to generate the weight of each edge in the learned path map, and the dynamic path planning process was also performed to plan a shortest path from room 1 to room 2. After these works were finished, the vehicle began to navigate along the path. When the vehicle arrived at room 2, the dynamic path planning process was performed to plan a shortest path from room 2 to room 1.

And certain representative images taken in the navigation are shown in Figure 7.3 through Figure 7.7; each of them showing a corresponding situation during the two navigation trips. In Figure 7.3, some pictures of the system interface are shown in Figures 7.3 (1) through (3). The image at the lower left part of the system interface was grabbed with the wireless camera on the vehicle at each navigation cycle. In the meanwhile, the image at the lower right part of the system interface shows a result of fuzzy guidance processing.

Besides, two situations of collision avoidance can be seen in Figures 7.3(8) through (10) and Figures 7.4(22) through (24). Especially in the latter situation, the vehicle was backing to room 1 and close to the destination. Since the position of the room node was very close to a piece of furniture (at about 40 cm aside), when the vehicle got close to the room node, it tried to turn its direction to avoid possible collisions. Therefore, the distance between room 1 and the final stop spot may be larger than expected. This kind of situation can be solved by adding a learning rule, which says that at least one meter between the position of a room node and its surrounding furniture should be maintained.

The speed of the vehicle used in this navigation is set to be 10 cm/sec. Moreover, the computing time of each navigation cycle is about 0.5 seconds.

Figure 7.1 An experimental result of navigation in single path mode.

Figure 7.2. An experimental result in area mode (1).

(1) (2) (3)

(4) (5) (6)

(7) (8) (9)

(10) (11) (12)

(13) (14) (15) Figure 7.3 An experimental result in area mode (2).

(16) (17) (18)

(19) (20) (21)

(22) (23) (24)

(25) (26) (27) Figure 7.4 An experimental result in area mode (3).

7.2 Discussions

We discover certain problems by analyzing the experimental results. And these problems are described as follows.

(1) The floor of the indoor environment has to be as flat as possible. If certain little bulges or hollows appear on the floor that the vehicle has to pass, then the moving direction of the vehicle will diverge. And this kind of situation cannot be detected in the proposed system, so if this situation occurs too often, then the vehicle will get lost gradually.

(2) Because of the small height of the camera on the vehicle, the eyesight of the vehicle is from about one to five meters. In other words, if an unexpected object appears suddenly in front of the vehicle and the distance between the object and the vehicle is smaller than one meter, then this object cannot be seen by the vehicle and the proposed fuzzy guidance technique cannot work to avoid the collision, either.

(3) The color of the floor in a navigation environment must be different from those of other objects. If the color of a certain object (e. g., a piece of furniture) is too similar to the color of the floor, then this kind of object will be regarded as part of the route area. Therefore, the proposed fuzzy guidance technique may not lead the vehicle to avoid probable collisions correctly.

Chapter 8

Conclusions and Suggestions for Future Works

8.1 Conclusions

Several techniques and strategies have been proposed in this study and integrated into an autonomous vehicle system with fuzzy guidance and simple learning capabilities. Satisfactory navigation results have been.

At first, a simple nonvisual learning strategy with two modes has been proposed for different navigation needs. Less data are acquired in the proposed learning process without causing instable and imprecise navigation results. Next, fuzzy guidance techniques which are based on the fuzzy theory and computer vision techniques have been proposed for guiding a vehicle safely in indoor environments with complicated scenes. The navigation is based on a collision avoidance concept using fuzzy-control -rules and image analysis techniques. Two kinds of navigation strategies, namely, line following and curve following, have been proposed to guide smoothly a vehicle in two different types of navigation sessions. And the experimental results showed the feasibility of the proposed method.

In addition, a person following process has been proposed to track the detected human appearing in navigation paths. A sequence of processes, namely, object detection, human identification, and human tracking, have been proposed to detect an

unexpected object and to identify it as a human or not. After an occurrence of a human is confirmed, a human tracking process is performed to track it continuously until the human disappears in the eyesight of the vehicle.

8.2 Suggestions for Future Works

The proposed strategies and methods, as mentioned previously, have been implemented on a small-vehicle system. Several related interesting issues are worth further investigation in the future. They are described as follows.

(1) Finding a method to calibrate the odometer during the navigation process, using certain external devices like a compass. Besides, fixed cameras in the navigation environment can be used to locate the vehicle accurately.

Therefore, odometer calibration can also be achieved using the location data given by the fixed cameras.

(2) Adding the capability of starting navigation from arbitrary start points when users demand the vehicle to move to other room spots.

(3) Adding the capability of voice control when users want to issue navigation orders to the vehicle.

(4) Analyzing the gait of a human from his/her back view to identify him/her.

(5) Adding the ability of detecting the situation that an unexpected human looks on the vehicle and trends to move closer to the vehicle.

(6) Finding a wireless network camera to replace the wireless 2.4G camera, aiming at reducing electric wave interference.

(7) Designing a camera system with a capability of panning, tilting, and

swinging.

References

[1] A. J. Davison and D. W. Murray, “Simultaneous localization and map-building using active vision,” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, Issue 7, Jul 2002, pp. 865-880.

[2] J. B. Hayet, F. Lerasle and M. Devy, “A visual landmark framework for indoor mobile robot navigation,” IEEE International Conference on Robotics and Automation, Vol. 4, 2002, pp. 3942-3947.

[3] Gaussier, P., Joulain, C., Zrehen, S., Banquet, J.P., Revel, A., “Visual navigation in an open environment without map,” Proceedings of 1997 IEEE/RSJ International Conference on Intelligent Robots and Systems, 7-11 Sept. 1997, pp.

545 – 550, vol. 2

[4] K. P. Li and W. H. Tsai, “Autonomous land vehicle guidance in complex room environments by computer vision techniques,” Proceedings of 1996 International Conference on Image Processing and Character Recognition (part of 1996 International Computer Symposium), Kaohsiung, Taiwan, Dec. 1996, pp.9-16.

[5] P. L. Li and W. H. Tsai, “Path Learning, Planning, and Guidance for ALV navigation Inside Buildings,” M. S. Thesis, Department of Computer and Information Science, National Chiao Tung University, 1997.

[6] G. N. Desouza and A. C. Kak, “Vision for mobile robot navigation: a survey,”

IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, Issue 2, Feb. 2002, pp 237-267.

[7] Jung Tae Cho and Boo Hee Nam, “A study on the fuzzy control navigation and the obstacle avoidance of mobile robot using camera,” Systems, Man, and Cybernetics, 2000 IEEE International Conference on , Vol: 4 , 8-11 Oct. 2000 pp.2993 - 2997 vol.4.

[8] Howard, A., Tunstel, E., Edwards, D., Carlson, A “Enhancing fuzzy robot navigation systems by mimicking human visual perception of natural terrain traversability,” IFSA World Congress and 20th NAFIPS International Conference, 2001. Joint 9th , Volume: 1 , 25-28 July 2001 pp.7 - 12 vol.1

[9] Nielsen, J. and Sandini, G, ”Learning mobile robot navigation: a behavior-based approach,” 1994 IEEE International Conference on Systems, Men and Cybernetics, 2-5 Oct. 1994, pp.2809 – 2814, vol. 3.

[10] Nayar, S.K., Murase, H., Nene, S.A., “Learning, positioning, and tracking visual appearance,” 1994 IEEE International Conference on Robotics and Automation, 8-13 May 1994, pp.3237 – 3244, vol.4.

[11] Bianco, G., Zelinsky, A., Lehrer, M, “Visual landmark learning,” 2000 IEEE/RSJ International Conference on Intelligent Robots and Systems, 31 Oct.-5 Nov. 2000, pp. 227 – 232, vol.1.

[12] Mata, M., Armingo, J.M., de la Escalera, A., Salichs, M.A., “Using learned visual landmarks for intelligent topological navigation of mobile robots,” ICRA '03. IEEE International Conference on Robotics and Automation, 14-19 Sept.

2003, pp. 1324 – 1329, vol. 1.

[13] Y. Tsukamoto, “An approach to fuzzy reasoning method,” in M. M. Gupta, R.K.

Ragade, and R. R. Yager, Eds., Advances in Fuzzy Set Theory and Applications, Amsterdam: North-Holland, pp. 137-149, 1979.

[14] Baldassarri, P., Puliti, P., Montesanto, A., Tascini, G. “Visual self-localization using automatic topology construction,” In proceedings of Image Analysis and Processing, 2003, Proceedings. 12^th International Conference, 2003, pp.

368-373.

[15] Golovko, V., Ignatiuk, O., Sadykhov, R., “An approach to self-training of the mobile robot,” In proceedings of the Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, International Workshop on, , 2001, pp. 11-15.

[16] G. Y. Chen and W. H. Tsai, “A new approach to vision-based unsupervised learning of unexplored indoor environment for autonomous land vehicle navigation,” Robotics and Computer-Integrated Manufacturing, Vol. 15, Issue 5, October, 1999, pp. 353-364.

[17] Guan-Yu Chen and W. H. Tsai, “An incremental-learning-by-navigation approach to vision-based autonomous land vehicle guidance in indoor environments using vertical line information and multi-weighted generalized Hough transform technique,” IEEE Transactions on System, Man and Cybernetics-Part B: Cybernetics, Vol. 28, No. 5, October 1998, pp. 740-748.

[18] P. Gaussier, C. Joulain, A. Revel, S. Zrehen, J. P. Banquet, S. Moga, and M.

Quoy, “Autonomous robot learning-what can we take for free,” Proceedings of IEEE International Symposium on Industrial Electronics, Vol. 1, Portugal, Jul.

7-11, 1997, pp. SS1-SS6.

[19] Thomas H. Cormen, Charles E. Leiserson, Ronald L. Rivest, and Clifford Stein, Introduction to Algorithm, McGraw-Hill, Reading, London, England, 2001.

[20] J.-S. R. Jang, C.-T. Sun, and E. Mizutani, Neuro-Fuzzy And Soft Computing: A Computational Approach to Learning and Machine Intelligence, Prentice-Hall, Reading, London, International(UK), 1997.

[21] Rafael C. Gonzalez and Richard .E, Digital Image Processing, Addison-Wesley, Reading, MA, New York, U.S.A., 1993.