Virtual trip: omni-directional photo database using probe cars

GPS World, Nov, 2006 by Kenji Muto, Kazuoki Matsugatani, Masumi Egawa, Hideaki Nanba

Search Procedure

We then improve the search mechanism. FIGURE 7 compares two images taken at almost the same location on different days. Image (a) was taken on a fine day; image (b) was taken on a cloudy day. Though they are taken at the same location, each of these images creates a different impression.

When browsing a series of images, these differing impressions may cause a problem (see FIGURE 8). As shown in (a), the probe car ran on route 1 during fine weather and the series of images were stored in the database. On another day, when the weather was cloudy, the probe car ran on route 2 and collected a different series of images. Using this simple procedure, the image at the intersection is overwritten as shown in (b).

Based on this mapped data, the series of images along route 1 will be shown as in A, B and C on the right of Figure 8. These images are unnatural because the impression for image B differs from the others. To solve this problem, we developed a search method that makes use of the probe car's environment or situation. When the probe car takes images, several additional pieces of information--time, date, location, and speed of the car--are also recorded.

From this information, we defined performance function as [S.sub.i](t, s,...). Here, i represents the index of the image, parameters t represent the time, and s represents the speed. This function is designeqd to extend a variety of the parameters. We calculated this function for all images and stored each of them in the database. When the user sends a request command to the server, the server searches for the most suitable image using the following formula:

F(i) = min(||[S.sub.o](t, s,...) - [S.sub.i](t, s,...)||) Here, ||a|| is a norm of a and [S.sub.o](t, s,...) is the situation of the current browsing image. From this formula, the server finds the index i image for the next scene.

FIGURE 9 shows the effect of this procedure. The data is not overwritten, and all of the photos taken at the same location are stored in the database. When the server sends the images from A to C, it uses the formula to select image D instead of B. The user will then browse images in the order A, D, and C. This series has a more natural relationship than the A, B, C sequence.

In this prototype, we elected only to use time t as a parameter, and we were able to confirm its effectiveness. We are planning to extend the parameters to include weather w using data from the weather database.

Application

Our last step was fabricating the client application software or browser. FIGURE 10 shows our application, which allows users to view a realistic image and landmark information on their desired location, as shown in (a). Using the control button, as shown in (b), users can look around in the desired direction. The arrow on the map shows the direction and location, as shown in (c). The circles on the map indicate the location where the photos are stored, as shown in (d). These photos, transmitted from the probe cars, are automatically updated by the database control software.