| Wayfinding is a basic activity that we cannot avoid throughout our daily life.Navigation services,as a special form of location-based services,compute optimal routes and generate route instructions for these routes.In this study,the route instruction follower is pedestrian.As the most important cues in navigation systems,landmarks can support spatial orientation and guiding pedestrians.The primary objective of navigation service is to guide users to their destination with minimum actual cost and perceived cost during wayfinding.Landmarks are considered as external points of reference for human's mental spatial models.In pedestrian navigation systems,the importance of landmarks is shown as follows: 1)Typically,people describe places and routes by landmarks,thus it is more natural to express places and guide persons to their destination by using landmarks;2)In unfamiliar environment,the landmarks are leveraged to confirm directions,integrating salient landmarks in route instructions not only can support spatial orientation,but also can reduce uncertainty and wayfinding errors;3)The identification of prominent landmarks while navigating can assure pedestrians that they are following the correct path and that the navigation system is functioning correctly,routes with many landmarks are considered higher quality than routes with fewer landmarks.Recently,the car navigation services have been widely used,but there are several problems for pedestrian navigation:(1)There is no appropriate digital representation of human concepts of place.However,humans often describe places by landmarks in natural language,but not using the current location on navigation maps.However,there is no appropriate digital representation of human concepts of place – a so far elusive concept – for use in information systems.Currently,there is no appropriate digital representation of human concepts of place for use in navigation systems.(2)The impact of spatial cognition on landmark selection is not fully considered,and the gaze data cannot be captured effectively to predict actual heading.In order to accomplish wayfinding task successfully,the navigation users need to search the surrounding environment to identify landmarks.Typically,a wayfinder should use a visual search in the environment to find a landmark pre-specified by the navigation system.However,the landmarks were selected using landmark salience calculation model or the user-generated content neglect the impact of spatial cognition on landmark selection.In addition,pedestrians' heading prediction methods are mainly relying on electronic compass,magnetometer and the integraded camera in smart phone.Actually,the precision of the heading prediction depend on gaze data.Nevertheless,the gaze data cannot be captured effectively to predict actual heading.(3)Only a few landmark images and directional instructions in route instructions adapted to the variation of location and heading of pedestrian during navigation.In comparison with driving,walking is more flexible: 1)people are not just constrained to walk on road;2)the variation of heading is dramatic(people can change their heading freely during walking).If the landmark images and the directional instructions in route instructions are not adapted to the variation of location and heading of pedestrian during navigation,it will increase the error of route selection.(4)Few studies develope a routing approach to avoid CDPs for landmark-based pedestrian navigation.Previous studies have proposed approaches to minimizing actual or perceived cost in the routing problem.For example,1)calculating the fewest-turn routes and the simplest routes on a road network for pedestrian navigation and car navigation;2)an approach has been proposed to avoid CDPs,measured by the number of nodes,for car navigation.However,few studies develope a routing approach for landmark-based pedestrian navigation to avoid CDPs,which is quantified by landmark data in real scenes.This dissertation proposes the corresponding solutions to solve above mentioned problems,and validate these solutions by experimental studies.(1)This thesis proposes and tests a formal model to capture the concept of place in human perception by transferring the concept of fingerprints from robotics: a place being characterized by the landmarks around it.For the purpose of realistically testing whether places can be described by fingerprints,an algorithm for computing places from fingerprints and,independently,an in situ survey to capture these places and their fingerprints were designed.The results indicate that the union of all locations with the same fingerprint characterizes a ‘place' in human cognition.(2)This dissertation proposes a method to select landmark data according to visibility,nearness and eccentricity based on the cognitive experiment of place description and the previous research,and devises the specific algorithm.Then,this dissertation proposes a method for spatial orientation that predicts the heading of the observers by an interpolation heuristic.The method uses both the match degree of the images and the location relationships between the observers and panoramas.The match degree is measured statistically according to a percentage of matched scale-invariant feature transform(SIFT)features.The results show that the method significantly outperforms methods that ignore the relationships between the locations of the observers and panoramas.(3)This study proposes an adapted generation method for landmark-based route instruction to resolve this problem using panoramas provided by street view services.Then,the qualitative representation of landmark's and navigational direction are implemented.Finally,this study implemented the proposed method using C++ computer language.The results show that the proposed method can generate dynamically the user-centered route instructions which are adapted to the location and heading of the user.(4)This thesis proposes an approach to avoid complicated decision points(CDPs),with landmarks that are difficult to recognize,for landmark-based pedestrian navigation.This approach offers a model to identify CDPs based on the size and text of landmarks from panoramas provided in Street View maps.A modified genetic algorithm(GA)is used to implement the computation of a route that avoids CDPs.In the GA,distance is set as a criterion.The case study results show the flexibility of the GA approach,and that it has the capacity to avoid CDPs and simultaneously maintain a shortest travel distance within a route computation.The approach of this study is important for building a full pedestrian routing system that can include additional user group preferences such as avoiding nodes with many branches. |
PDF Full Text Request