Multi-branch Deep learning Based Transport Mode Detection using Weakly Supervised Labels
This is a modal window.
The media could not be loaded, either because the server or network failed or because the format is not supported.
Formal Metadata
Title |
| |
Title of Series | ||
Number of Parts | 351 | |
Author | ||
License | CC Attribution 3.0 Unported: You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal purpose as long as the work is attributed to the author in the manner specified by the author or licensor. | |
Identifiers | 10.5446/68968 (DOI) | |
Publisher | ||
Release Date | ||
Language | ||
Production Year | 2022 |
Content Metadata
Subject Area | ||
Genre | ||
Abstract |
| |
Keywords |
FOSS4G Firenze 2022274 / 351
1
7
13
22
25
31
33
36
39
41
43
44
46
52
53
55
58
59
60
76
80
93
98
104
108
127
128
133
135
141
142
143
150
151
168
173
176
178
190
196
200
201
202
204
211
219
225
226
236
242
251
258
263
270
284
285
292
00:00
Group actionMultiplicationEndliche ModelltheorieTrailTrajectoryPoint (geometry)AreaComputer animation
00:10
Endliche ModelltheorieObservational studyTrajectoryTask (computing)Set (mathematics)Computer animation
00:25
Data modelAndroid (robot)ArchitecturePredictionDivisorPhysical systemKolmogorov complexityFunction (mathematics)Mathematical modelComputer networkCAE Inc.Transportation theory (mathematics)Hausdorff dimensionPopulation densityDirection (geometry)MathematicsMultiplicationData structurePoint cloudProcess (computing)Set (mathematics)Observational studyEndliche ModelltheoriePoint (geometry)Process (computing)Wave packetComputer architectureMobile appMaxima and minimaSupervised learningBus (computing)Row (database)Branch (computer science)ImplementationSoftwareGene clusterFunctional (mathematics)Utility softwareSmartphoneStapeldateiFunctional (mathematics)MultiplicationTable (information)outputResultantAndroid (robot)Public domainMathematical modelSoftware testingService (economics)Basis <Mathematik>TrajectoryInsertion lossOpen sourceComputer animation
03:20
Data modelComputing platformPlastikkartePredictionNumberAndroid (robot)MehrplatzsystemAndroid (robot)NumberoutputRight angleEndliche ModelltheorieComputer animation
Transcript: English(auto-generated)
00:00
Mobility data based on global positioning system tracking have been widely used in several areas. A transport mode model assigns a GPS point or a GPS trajectory to a particular transport mode based on user's activity. Unavailability of transport mode training datasets make it a challenging task to build a TMD model.
00:23
Hence, this study proposed a weakly supervised model using snorkel to generate synthetic training data. Further study proposed a two-branch deep learning architecture that well suited for transport mode detection. Results are evaluated generously with previous machine learning model as well as with the iOS and Android inbuilt tool.
00:46
In any supervised learning approach, label data availability is critical to maximize model performance. Hence, this study proposes a transport mode label generation approach using snorkel with programmatic labeling as there is no ground truth data available for this study.
01:05
We considered publicly available datasets related to road network, railway network, bus routes, etc. These datasets were used to create road bus train labels by overlapping GPS points on associated transportation network.
01:22
These labels are called soft labels. Transportation network based soft labeling and other mobility features are used to define multiple labeling functions in snorkel. Study proposed a two-branch deep learning architecture where road GPS latitude and longitude values
01:42
are used in one branch and derived mobility features are used in the other branch. We used three fully connected hidden layers for road GPS data with latitude and longitude values. For mobility features, we used four fully connected hidden layers.
02:01
Features derived from the two branches are concatenated in feature domains. Furthermore, three fully connected hidden layers and softmax cross-entropy were used as a loss function. This study utilized various free and open source solutions to implement the deep learning model using a scheduled pipeline for estimating TMD on a daily basis.
02:26
Multiple dedicated mobile applications collect big GPS data and the pipeline will gather daily data and process using function as a service. The jobs are scheduled and triggered as daily batch using Apache Airflow.
02:41
The implementation utilized Apache Hadoop clusters to process big GPS data and Apache Hive tables are used to store the daily batch TMD results. We compared the proposed model with our own previous model AXG Boost Classifier and improved accuracy by nearly 2x.
03:03
Moreover, we collected smartphone-based GPS trajectories for multiple modes of transportation collected by testers in Tokyo, Japan. And resulted in higher average accuracy compared to the Android and iOS inbuilt.
03:22
Overall, 17.86% better than iOS Android inbuilt tool. Proposed model improved number of modes and user coverage. Top right animation demonstrates the transportation detection for a single user in a day.
03:44
Thank you for listening.