System performance

The proposed system is implemented in Java language, compiled by Eclipse and ex-ecutable on Android platform. The Android NDK and Google APIs are also applied to

(a)

(b)

(c)

Figure 4.10: The demonstration of speech recognition. (a) Speech ask. (b) Speech recognition. (c) Speech match success.

(a)

(b)

(c)

Figure 4.11: The following demonstration of computing the direction. (a) Taking a forward-facing picture. (b) Image matching processing. (f) Speech guidance to turn direction.

(a)

(b)

Figure 4.12: The following demonstration of the path monitoring and notification. (a) Path monitoring. (b) The notification of station, destination , and the remaining distance.

our system. The system performance is recorded by timer which counts the processing time of matching images on the mobile phone in log files. We construct the proposed vision-based navigation system on the mobile phone with 1 GHz CPU, 576 megabytes ram, and five mega pixel color image. The resolution of each query image is 320 × 240, and 1440 × 244 for the panoramic image in the database. The image matching time for each panoramic image is shown in Table 4.2. The off-line means only counting the image matching time, and the on-line means counting the image matching time and running the system meanwhile.

Matching Time

Off-line 4 ∼ 5 seconds On-line 8 ∼ 10 seconds

Table 4.2: The image matching time for each panoramic image.

4.4 Stability

The most important advantage in our system is robust stability. Figure 4.13 shows the variance of GPS in the same position at different times.

Figure 4.13: The variance of GPS.

Figure 4.14: The variance of using Compass

As shown in Figure 4.14, the variance of using compass. Also we put mobile phone stable as far as possible in our hand while walking, it still has big variance. However,

our system provides more robust and stable results by using computer vision method as shown in Figure 4.15. The red line means the matching line of the facing direction. We compute the standard deviation, and the average, and the variance of computing facing degree between the proposed system and digital compass, as shown in Figure 4.16 and Table 4.3. The standard deviation of the proposed system is lower than digital compass.

Figure 4.15: The matching result for the proposed system.

Figure 4.16: The matching result for the proposed system.

VBNS Compass

STD. 1.98 4.08

Average 44.19 39.09

Table 4.3: The comparison between the proposed system and the digital compass.

Chapter 5

Conclusions

5.1 Discussions

Issues for matching accuracy As mentioned previously in Section 4.2.1, the accuracy in region A, B, and C are: A < B < C. We observe that the geographical environment in region A is more complicated than B and C. The pathway in region A has two intersection, only near one building (MISRB), illumination change, and many non-static region such as cars, moving people and trees (Figure 5.1).

The above reasons cause the difficulties of recognition. The region B is a straight road, and has more buildings along the path. Hence, it is easier to match the database.

The reason for high accuracy of region C is the distinctiveness of FSD2, therefore it has the highest accuracy.

Figure 5.1: The example of non-static region and illumination change. (a)Cars (b)Moving people and cars (c)Trees (d)Illumination change.

Causes of inaccuracy In our system, we have previously established in the panoramic database, and the starting position and destination of the user must near stations in the panoramic database. Causes of inaccuracy: GPS, calibration, and human factors.

5.2 Conclusions

In this work, we have proposed a vision-based navigation system for visually impaired people using computer vision method. In initialization, we construct the database which includes GPS information, map, and panoramic images. Then our system provides a friendly user interface to support speech input and output. Our system first record the GPS information of the user. According to the destination of user, we plan a shortest path. The most important need for visually impaired people is the information of orientation. We use computer vision methods, SURF algorithm and least square method, to calculate the orientation which the user needs to turn. Besides, we provides the distance information for the user to know how far to reach each station and destination. On path monitoring, we give issue an alarm if the user’s orientation is over 10 degrees. The user needs to turn back to the correct orientation until the alarm silences. Finally, our system will notify the user when they arrive each station or the destination. The advantage of our system is stable and robust. Although the accuracy is not high, we provide the chances to take a picture again when the user needs to calculate his orientation. In the future, we want to expend the region of the database and to improve the matching accuracy. We hope our system can improve the daily lives for visually impaired people.

Bibliography

[1] The architecture of androidar. Website. http://developer.android.com/

guide/basics/what-is-android.html.

[2] Speech input. Website. http://developer.android.com/resources/

articles/speech-input.html.

[3] H. H. Alan L. Liu, H. Kautz, G. Borriello, P. A. Brown, M. Harniss, and K. Johnson.

Indoor wayfinding: Developing a functional interface for individuals with cognitive impairments. In Proceedings of Computers Accessibility, ASSETS, pages 95–102, 2006. .

[4] H. Bay, A. Ess, T. Tuytelaars, and L. V. Gool. Surf: Speeded up robust features. In Computer Vision and Image Understanding (CVIU), volume 110, pages 346–359, 2008.

[5] M. G. Dodson AH, Moore T. A navigation system for the blind pedestrian. In GNSS 99, page 513–518, 1999. .

[6] R. G. Golledge, R. L. Klatzky, J. M. Loomis, J. Speigle, and J. Tietz. A geographical information system for a gps based personal guidance system. In Int J Geograph Inform Sci, page 727–749, 1998. .

[7] B. JA. A review of mobility aids and means of assessment. In In: Warren DH, Strelow ER (eds) Electronic spatial sensing for the blind. Martinus Nijho, page 13–27, 1985. .

[8] K. L. Air sonar with acoustical display of spatial information. In Busnel RG, Fish JF (eds) Animal sonar system. Plenum, New York, page 769–816, 1980. .

[9] S. H. Lisa Ran and S. Moore. Drishti: An integrated indoor/outdoor blind naviga-tion system and service. In Proceeings of the Second IEEE Annual Conference on Pervasive Compuing and Communications (PerCom.04), pages 23–30, 2004. .

[10] Microsoft. Image composite editor. Website. http://research.

microsoft.com/en-us/um/redmond/groups/ivm/ICE/.

[11] E. E. Sabelman, C. G. Burgar, G. E. Curtis, G. Goodrich, D. L. Jaffe, J. L. McKinley, M. V. der Loos, and L. G. Apple. Personal navigation and wayfinding for individuals with a range of disabilities. In Project report: Device development and evaluation, 1994. http://guide.stanford.edu/Publications/dev3.html.

[12] T. Strothotte, S. Fritz, R. Michel, A. Raab, H. Petrie, V. Johnson, L. Reichert, and A. Schalt. Development of dialogue systems for the mobility aid for blind people:

initial design and usability testing. In ASSETS ’96, page 139–144, 1996. .