Open AccessPreLoc: predicting pedestrian's location and learning unmapped pathways by long short‐term memory model
Author(s) -
Zhong Haolin,
Tian Jacky,
Yao Chang
Publication year - 2019
Publication title -
iet communications
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.355
H-Index - 62
eISSN - 1751-8636
pISSN - 1751-8628
DOI - 10.1049/iet-com.2018.5881
Subject(s) - computer science , global positioning system , term (time) , pedestrian , popularity , point (geometry) , location based service , artificial intelligence , machine learning , data mining , real time computing , transport engineering , mathematics , telecommunications , engineering , psychology , social psychology , physics , geometry , quantum mechanics
With increasing popularity of smartphone apps such as fitness assistants, location‐based services, more and more pedestrians now travel in short distances with purposes of leisure, running exercise or social activities. As users move from place to place, the mobile apps capture the data showing the history of change in location in a definite time interval. Given the set of global positioning system (GPS) location points a user has travelled along, this study presents, PreLoc , a system that accurately predicts the user's future location. By preprocessing the raw data, the authors eliminate existing noises and extract valid trajectories from pedestrians or runners. Then the long short‐term memory model is applied to predict the next GPS point that the user will pass through. Meanwhile, PreLoc is able to learn an unmapped pathway based on the trajectory data of mass users that have travelled over this pathway. Experimental results over real‐world dataset show the distance error could achieve about 12 m in residential area, and achieve acceptable distance errors (about 30 m) for other larger and complicated sites. Moreover, they provide real scenarios for unmapped pathway learning.