An Improved Map-Matching Technique Based on the Fréchet Distance Approach for Pedestrian Navigation Services
Wearable and smartphone technology innovations have propelled the growth of Pedestrian Navigation Services (PNS). PNS need a map-matching process to project a user's locations onto maps. Many map-matching techniques have been developed for vehicle navigation services. These techniques are inappropriate for PNS because pedestrians move, stop, and turn in different ways compared to vehicles. In addition, the base map data for pedestrians are more complicated than for vehicles. This article
... s a new map-matching method for locating Global Positioning System (GPS) trajectories of pedestrians onto road network datasets. The theory underlying this approach is based on the Fréchet distance, one of the measures of geometric similarity between two curves. The Fréchet distance approach can provide reasonable matching results because two linear trajectories are parameterized with the time variable. Then we improved the method to be adaptive to the positional error of the GPS signal. We used an adaptation coefficient to adjust the search range for every input signal, based on the assumption of auto-correlation between consecutive GPS points. To reduce errors in matching, the reliability index was evaluated in real time for each match. To test the proposed map-matching method, we applied it to GPS trajectories of pedestrians and the road network data. We then assessed the performance by comparing the results with reference datasets. Our proposed method performed better with test data when compared to a conventional map-matching technique for vehicles. Sensors 2016, 16, 1768 2 of 18 users because the map-matching techniques are inherently for vehicles, and therefore are unsuitable for PND. Moreover, pedestrians have different patterns of movements compared with vehicles. Movements of pedestrians have a much higher probability to stray from the road lines because they are much less dependent on the road lines than those of vehicles. Figure 1 shows the examples of GPS signals and network datasets. Therefore, it is difficult to predict the changes in the state of motion such as start, stop, acceleration, turn, etc. Instead, pedestrian users show relatively low velocity so that we do not need to compensate for the delay that occurred in processing time of matching. Table 1 shows the differences between navigation services for a vehicle and a pedestrian.