Abstract
This paper presents a method of map matching that is a scheme for accurate guidance of an electric wheel chair. Indoor navigation seems to require more accurate guidance than outdoor areas, the location estimation by the rotary encoders embedded in the wheel chair cannot be satisfy the accuracy requirement, because estimation error is accumulated as the wheel chair travels. The authors propose the map matching that uses the building structure in order to compensate for accumulated position error. The corner detection and its position information are used for replacement as a correct position. The methods of the corner detection and the calculation model for the correct position are shown, and the validity of the proposed methods are confirmed by the experiment using a laser range finder.
You have full access to this open access chapter, Download conference paper PDF
Similar content being viewed by others
Keywords
1 Introduction
Many schemes have been proposed for the operation of an electric wheelchair for aged or physically disabled persons [1, 2]. The authors are now investigating automatic guidance for such a wheelchair that will allow its operation in an extensive indoor area. Spacious shopping malls, large hospitals and the like are complicated, making it challenging to reach a particular goal even for the physically healthy. Relevant studies of automated traveling technology have been pursued in recent years [3], including attempts to create a driverless car. Indoor navigation seems to require more precise guidance than outdoor areas, because corridors are narrower, and rooms are smaller than road. The room plates are used for landmarks to recognize a mobile robot location [4]. This scheme requires high accurate image recognition in any illumination environment. This paper presents a scheme for the accurate guidance of an electric wheel chair as it travels within an extensive indoor area. Map matching has been proposed to compensate for accumulated position error, using corner detection in corridors. This scheme does not require any addition of landmarks, and it is desirable from aspect of aesthetic purposes and workload of setting landmarks. This technique’s effectiveness has been verified by experiment, with an automatically controlled wheelchair traversing a corridor, including a corner.
2 Sequence to Target Position and Necessity of Error Correction
Figure 1 shows the sequence of events as an automatic wheelchair follows a route to the target spot. We assumed that the route to the target spot was composed of straight travel and 90°turns, which is normal in most buildings. The main actions to reach at target spot include straight travel, turns, and map matching to correct accumulated position error.
The route is determined in advance by designating the start and target spot using the Dijkstra method and map information. We require arrival accuracy at the final target spot of better than about 10Â in, which is based on the width of a door.
An electric wheelchair itself generally has two rotary encoders, and these sensors are used for automatic travel and position estimation. One encoder is installed in the left motor and one in the right motor, to detect each motor’s rotation speed. The other sensor, a laser range finder (LRF), is used to detect the distance to nearby obstacles such as wall. The rotary encoder is also used for curved path travel, in which the right and left wheel rotation must adjust to the radius of curvature, and the LRF is used for straight travel detecting the distance to left- and right-and walls, and keeping the ratio of the distance constant. Figure 2 shows the electric wheelchair with its LRF installed on a pole to maintain a height sufficient to avoid obstacles such as a pedestrian in the corridor. The main specification of the LRF used in this investigation is summarized in Table 1.
The position of the electric wheelchair can be estimated by integration of the angular velocity of the two motors. However, this method is not precise, because of variables such as wheel slippage on the floor and pressure differences in the tires. This makes it necessary to compensate for position-estimate error to arrive precisely at the target spot.
3 Error Correction by Map Matching for Precise Travel
The authors propose error correction by map matching. Map matching itself is not a new technology [5]. It is widely used, especially for driving navigation to replace a car’s road position when GPS yields an improbable position. Some markers are proposed for map matching that indicate the position in an indoor area. However, this method requires the installation of place markers, which becomes a burden and can be unsightly. We propose to use the structure of the building for reference points, specifically its corners, so that no markers need to be installed. The position estimated by the encoders is updated based on the position of a nearby corner. We propose this method for position estimation error correction.
3.1 Basic Principle of Corner Detection
To begin with, detecting the corner is required in order to use the proposed method. The LRF is used for corner detection. Figure 3 indicates the criteria for detecting at corner in the corridor. The distance detection limit of the LRF used in this investigation is 5.8 m. Therefore, anytime that at wall cannot be detected within 5.8 m is taken to indicate a corner, as shown in Fig. 3. The four possible patterns are indicated in this figure: no corner (straight corridor), L-shaped corner, T-shaped corner and Cross-shaped corner.
The angle range for corner detection depends on the structure of the corridor, as shown in Fig. 3, especially on the width of the corridor and the position of the LRF on the wheelchair. The wheelchair is assumed to travel along the center of the corridor in this investigation, and the angle ranges are examined in advance by using the LRF on the wheel chair. The range for the corridor (no-wall detection range) is set from 14.3° to 38.1°. The type of corner is determined from the characteristics of this no-wall region as shown in Fig. 3, from these choices: no region = L-shaped corner, one = no corner (straight corridor), two = T- shaped corner, three = Cross-shaped corner.
3.2 Evaluation of Detection Capability
Tests of corner detection were carried out in the area shown in Fig. 4. This area had The L-, T- and cross-shaped corners. Figure 5 shows one example of angle range variation of the no wall-region in a T-shaped corner, in which the travel direction of the wheelchair is shown in Fig. 4. The x axis indicates the travel time and y axis is the no-wall angle range. This no-wall region depends on the LRF measurement distance limit. Although only one no-wall region was found at the beginning, two no-wall regions appeared as the wheelchair traveled. The T-shaped corner was recognized at the moment of detected change, that is, from the one no-wall to the two no-wall condition. The first position satisfies the conditions for a no-wall region, i.e. from 14.3° to 38.1°, is regarded as the finding of the corner in this investigation. We call this point the node.
The evaluation test for corner detection was carried out in the area shown in Fig. 4. Sun illumination produces obstacle detection in the no-wall region, so a blind was used on the window so as to exclude direct sun-light in this experiment. The test result is shown in Table 2. Quite high detection performance was demonstrated for that corner in the usual room light environment.
3.3 Node Setting for Accumulated Error Correction
It is necessary to determine the position where corner detection occurs, and we call that point the node. Its position replaces the estimated position obtained by the rotary encoders, because the estimated position contains accumulated error. The geometrical model for determining the node position is shown in Fig. 6. The no wall detection range can be calculated from the building’s corridor structure and the position of the LRF by geometry. When is between 14.3° to 38.1°, the wheelchair is at the corner. In this investigation, at the instant when the above condition is satisfied for the wheelchair, map matching is carried out, i.e. replacement of the estimated position by the node position. The meaning of the terms is shown in the figure.
where,
The position of nodes calculated by this proposed method is stored in the database. When at corner is detected by the traveling wheelchair, the estimated position is replaced by this node position to eliminate the accumulated position error. The scheme can be regarded as map matching in the widest sense of the term.
3.4 Evaluation Test for Node Detection Point
In the proposed method, the node position decision is quite important, because it governs the accuracy of the wheelchair’s position by map matching. An experiment to evaluate the node position was carried out to confirm the validity of the proposed method and its accuracy. The experiment was carried out with the wheelchair traveling in the center of the corridor. The area is that shown in Fig. 3, and the traveling directions are shown in Fig. 7. Table 3 shows the experimental results and the theoretical values of the node position. The average and standard deviation for the experimental results are also shown. The maximum difference between experimental and theoretical values is 77 mm. The validity of the node determination was confirmed, and this result seems to establish that the position error of the wheelchair is less than tens of centimeters.
4 Automatic Travel Experiment and Position Estimation by Map Matching
An automatic travel experiment was carried out using the proposed scheme. The result is shown in Fig. 8. The black dots indicate the results with map matching. The grey dots are the results without map matching. The discontinuous points indicate map matching locations. Map matching was carried out three times at each corner. The positioning of the wheel chair is not at the center of the corridor because it records the location of the LRF where it is attached to the wheelchair. It was verified that position correction by map matching was carried out correctly, and at the same time, the wheel chair traveled parallel to the wall and turned appropriately.
5 Conclusion
This paper presents a map matching method that uses the detection of corners in a corridor. This method enables high accuracy travel for a wheelchair in an extensive indoor area without adding and landmarks. Methods were proposed for corner detection and the calculation of the node points at the corners to enable map matching, and their validity was experimentally verified. Position evaluation for travel to specific places, and investigation of a travel method that includes the use of an elevator remain for further studies.
References
Simpson, R.C., Levine, S.P.: Voice control of a powered wheelchair. IEEE Trans. Neural Syst. Rehabil. Eng. 10(2), 122–125 (2002)
Tanaka, K., Matsunaga, K., Wang, H.O.: Electroencephalogram-based control of an electric wheelchair. IEEE Trans. Robot. 21(4), 762–766 (2005)
Luettel, T., Himmelsbach, M., Wuensche, H.-J.: Autonomous ground vehicles - concepts and a path to the future. Proc. IEEE 100, 1831–1839 (2012)
Luo, R.C., Yu-Chih, L., Ching-Chung, K.: Autonomous mobile robot navigation and localization based on floor plan map information and sensory fusion approach. In: 2010 IEEE Conference on Multi sensor Fusion and Integration for Intelligent Systems (MFI), pp. 121–126 (2010)
Quddus, M.A., Ochieng, W.Y., Noland, R.B.: Current map-matching algorithms for transport applications: state-of-the art and future research directions. Transp. Res. Part C Emerg. Technol. 15(5), 312–328 (2007)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2015 Springer International Publishing Switzerland
About this paper
Cite this paper
Noriduki, Y., Shibata, H., Ioroi, S., Tanaka, H. (2015). Map Matching to Correct Location Error in an Electric Wheel Chair. In: Yamamoto, S. (eds) Human Interface and the Management of Information. Information and Knowledge in Context. HIMI 2015. Lecture Notes in Computer Science(), vol 9173. Springer, Cham. https://doi.org/10.1007/978-3-319-20618-9_24
Download citation
DOI: https://doi.org/10.1007/978-3-319-20618-9_24
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-20617-2
Online ISBN: 978-3-319-20618-9
eBook Packages: Computer ScienceComputer Science (R0)