Paper Info
Title
Transition-points-based Segmentation and Hierarchical Classification for Locomotion and Transportation recognition on Radio-data
Abstract
ABSTRACT Technological advancements these days allow people to handle the locomotion and transportation recognition from mobile phone data, which plays an important in human behavior analysis. Motion exploration from accelerometer or gyroscope data is commonly used to achieve high performance in this recognition task. However, the radio data collected from mobile phones supposedly can take on this role while providing more useful information about the location and mobile connection. Our team (Hu-Bi) proposes a pipeline based on this type of data, includes transition points-based segmentation and hierarchical classification to recognize 8 modes of transportation and locomotion: Still, Walk, Run, Bike, Car, Bus, Train, and Subway on The 2021 Sussex-Huawei Locomotion-Transportation (SHL) recognition challenge Dataset which involved the GPS location, GPS Reception, Wi-fi Reception, and GSM cell tower scans. The hierarchical classification includes two stages, firstly divide the time-series data into two groups based on speed-related features, and then the second classifier provides the target labels by extracting robust features from the segmentation results. Unlike other works, we put the bus mode along with non-motorized mode into the low-speed group, and other motorized modes are in the high-speed group. The transition-points-based segmentation aims to overcome the time-window segmentation limitations, correctly partitions the time-series into an optimized length of each group of transportation modes. We trained our model on the Train data and evaluate it on the Validation data, our approach achieved 94% accuracy in the initial Classification and Segmentation, reached 88% accuracy for each group prediction, and 75% accuracy on each data points for the whole process.
Year
DOI
Venue
2021
10.1145/3460418.3479380
Ubiquitous Computing
Keywords
DocType
Citations 
SHL dataset, hierarchical classification, radio-data, transition points based segmentation
Conference
0
PageRank 
References 
Authors
0.34
0
3
Name
Order
Citations
PageRank
Nhat-Tan Le100.34
Nazmun Nahid200.34
Sozo Inoue317658.17