The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XLVIII-4/W1-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-4/W1-2022, 525–530, 2022
https://doi.org/10.5194/isprs-archives-XLVIII-4-W1-2022-525-2022
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLVIII-4/W1-2022, 525–530, 2022
https://doi.org/10.5194/isprs-archives-XLVIII-4-W1-2022-525-2022
 
06 Aug 2022
06 Aug 2022

MULTI-BRANCH DEEP LEARNING BASED TRANSPORT MODE DETECTION USING WEAKLY SUPERVISED LABELS

P. Vinayaraj and K. Mede P. Vinayaraj and K. Mede
  • Rakuten Institute of Technology, Rakuten Group, Inc., Tokyo, Japan

Keywords: Transport mode detection, Deep learning, Weakly supervised label generation, Global Positioning System

Abstract. Mobility data, based on global positioning system (GPS) tracking, have been widely used in many areas. These include, but not limited to: user direction guidance, analyzing travel patterns, and evaluating travel impacts. Transport Mode Detection (TMD) is an essential factor in understanding mobility within the transport system. A TMD model assigns a GPS point or a GPS trajectory to a particular transport mode based on the user’s current activity. However, the complexity of the prediction procedure increases with the number of modes that need to be predicted given the increasing overlaps in feature values between multiple transportation modes. Hence, this study proposes a two-branch deep learning-based TMD model that predicts multi-class transport modes to improve prediction accuracy. In addition, it proposed a weakly supervised labelling model using snorkel to improve the volume of labelled data and resulting TMD model prediction accuracy. We considered publicly available road networks, railway networks, bus routes, etc., for creating road, bus, train labels by overlaying GPS points on these transportation networks. We introduced a boolean (true/false) based soft-labelling function, where the same GPS point overlaid on road or railway network. The raw GPS data were used to generate point-level features such as speed, speed difference, acceleration, acceleration difference, initial bearing and bearing difference, all used as derived features for the TMD model. To construct the model we opted to use two branches where raw GPS latitude and longitude values were used in one and the derived mobility features are used in the other.