In this paper, we propose a novel end-to-end localization method on topological maps using the recurrent neural network model. In the era of big data and deep learning, we are motivated by availability of large amount map data that constantly gets updated. When a pedestrian moves on a map, such as the road network on OpenStreetMap (OSM) that is an abstract graph with nodes and edges, motion features, such as a sequence of turning angles can be used to represent the trajectory. Proposed recurrent neural network takes the trajectory as its input and locates trajectory. In order to learn potential motion patterns on a map, training dataset is generated on OSM. In our experiments, we captured real-time pedestrian trajectory using an inertial navigation system (INS) to test the trained network in a realistic case. When sufficient amount of nodes are traversed, the users can be efficiently localized and tracked on OSM by using the proposed network. We have observed that the final classification accuracy reaches up to 98%.
Paper
Full text
Trajectory Mining for Localization Using Recurrent Neural Network
Semantic Scholar · Computer Science · 2019
Abstract
In this paper, we propose a novel end-to-end localization method on topological maps using the recurrent neural network model. In the era of big data and deep learning, we are motivated by availability of large amount map data that constantly gets updated. When a pedestrian moves on a map, such as the road network on OpenStreetMap (OSM) that is an abstract graph with nodes and edges, motion features, such as a sequence of turning angles can be used to represent the trajectory. Proposed recurrent neural network takes the trajectory as its input and locates trajectory. In order to learn potential motion patterns on a map, training dataset is generated on OSM. In our experiments, we captured real-time pedestrian trajectory using an inertial navigation system (INS) to test the trained network in a realistic case. When sufficient amount of nodes are traversed, the users can be efficiently localized and tracked on OSM by using the proposed network. We have observed that the final classification accuracy reaches up to 98%.