Skip to main content

Big Data for Mobility Tracking Knowledge Extraction in Urban Areas

Periodic Reporting for period 2 - Track and Know (Big Data for Mobility Tracking Knowledge Extraction in Urban Areas)

Reporting period: 2019-07-01 to 2020-12-31

The business impact of Big Data is starting to take form in many fields, however, innovation is held back by limitations of current Big Data processing methods and infrastructures. Track&Know researched and developed a framework that aims to increase the efficiency of Big Data. A variety of toolboxes, containing specific methods & algorithms for various types of data aggregation, manipulation & further analysis were developed within the project. The Big Data Processing toolbox implements data acquisition technology that captures data from heterogeneous data sources, extending the current solutions for efficient access, indexing, partitioning and load balancing for Big spatiotemporal data. The Complex Event Recognition toolbox detects complex event occurrences by analyzing patterns in simple events by using contextual information and results from the Big Data Analytics toolbox. The toolbox may infer a complex event (such as dangerous driving) by analyzing patterns based on vehicle speed, direction, driver events, fuel consumption and contextual information (e.g weather).
The BDA toolbox analyzes heterogeneous data, drawing conclusions about spatiotemporal distribution of mobility patterns by applying scalable data mining techniques (such as clustering, sequence mining, hotspot analysis) for voluminous trajectory data. The Visual Analytics toolbox develops interactive and scalable methodologies to efficiently visualize heterogeneous spatiotemporal data of varying levels of resolution and quality. The toolboxes are included in the Big Mobility Data Integrator platform and tested against three pilot cases, organized with two common links: service optimization and driver behavior. Applied to the Insurance sector: (a) to gain in-depth and accurate crash probability estimation; (b) Electric Cars adoption by studying the cost-benefit of switching to electric vehicle, matching charging times and points to drivers’ habits; (c) Car Pooling opportunities by analyzing for reduced parking on sharable routes with a cost-benefit of switching to a sharing mobility paradigm, and likelihood of finding a proper sharable route that matches time and geographical zone. Within the Health Service: (a) improve the response times (increasing new patients, follow-up patients; (b) reduce unnecessary travel (reduce patient travel distances, courier costs, CO2 emissions); (c) generate Cost efficiency gains; (d) New methods of OSA diagnoses based on driver behavior. For Fleet Management, Track&Know’s Business objectives include Predictive maintenance, anomaly detection, reduction of false alarms, correlation of data with weather services, fleet costs reduction, fleet downtime reduction, fleet response time improvement, improve driver behavior and reduce accidents.
The project achieved:
-Establishment of an Online Observatory comprising a literature survey, open-source software, and datasets for Big Data from a mobility context
-Creation of a highly scalable interoperable Big Data platform for streaming and at rest data. The platform is able to interface with all required database systems and has an easy plug-and-play approach to integrating toolboxes
-Data cleaning, map matching and enrichment pipeline to deal with streaming and historic heterogeneous mobility data, with different spatiotemporal resolutions, noisy signals
-Creation of data access operators over NoSQL stores enabling interfacing using just a common framework
-Mobility analysis methods for individual long-term crash prediction using individual mobility models and processes them in combination with individual driving behavior traces, as well as individual crash data
-Developments in future location predicting in real-time the positions of a fleet of moving objects in some user-specified timepoint. Predictions of FLP are for a longer time period in the future. More efficient for big fleet management
-The complex event recognition algorithms are able to leverage the enriched data pipeline to identify high-speed driving, dangerous or non-eco driving, and refueling opportunities using inputs such as road network information, abrupt acceleration/deceleration/cornering, points of interest information, and weather information
-Contextual Analysis of Movement Events through visual analytics enabling greater insights and understanding of driving data, resulting in a Best Paper award at the EuroVA 2019 conference
- Automated mobility analysis methods on costs/benefits for switching to EV or Carpooling by taking the individual mobility models, plotting them onto the road network, and the EV charging network
-Mobility driven service optimization model that can be used together with expert knowledge for (any) health service redistribution at different geographical levels.
-A Data transformation service, which takes patient appointment booking data and outputs journey data, could be used for any similar analysis
-OSA Driving profile is a set of features captured via accelerometer reading, identifying deviations from normal driving that are due to sleepiness
-Driving Safeguard is a service providing car collision risk prediction based on driver history and additional environmental information to reduce driving risk and save money from less crashes
-E-mobility Services software providing visibility of mobility patterns of vehicles (at group and at individual level) vs. the geographical locations EV stations.
-Fleet & Pooling software that visualizes and matches similar mobility paths of individuals in a private social network to propose pooling options
-Deep Fleet Analytics software for Fleet Data analytics with ML features to manage in real time a fleet of vehicles
-Customizable analytics dashboard to support cross-scale analysis and location-allocation analysis has been designed for the pilots
-Creation of an app for OSA patients to use during their diagnosis phase to determine the extent their condition affects driving ability
-Publications related to scientific work carried out in Track&Know have been accepted and presented at 32 academic conferences and 19 papers have been published in academic journals
-Track&Know held a Special Session alongside the ANT 2019, held several demos and talks at the 2020 EU Big Data Value Forum, and showcased results and provided input at 40+ distinct conferences, workshops, podcasts, webinars, and liaison activities.
Track & Know delivered impact across several socio-economic factors. Through the pilots:
- reduced patient travel times, resulting in CO2 reductions, and operating cost reduction
- reduction in accidents by approaching all sectors: Insurance, Health, and Fleet Management
- reduction of CO2 in everyday mobility applications
Track & Know has made several key innovations that have been or are being taken to market.
Implemented Prototypes:
-Big Mobility Data Integrator Platform – the Track&Know platform capable of processing Big Data streams and achieves
-Data Cleaning, map matching and Enrichment Pipeline – integrated into the Track&Know platform to pre-process at scale large amounts of data and augment with POI, Weather, and Map data
-Future Location Prediction components for online / streaming processing
-Big Data Analytics and Complex Event Recognition toolboxes using deep/machine learning to provide beyond the start of the art insights into large scale Big mobility data
-Dashboard with integrated visual analytics methods to identify complexities with data at scale.
Track & Know Big Data Architecture
Track & Know Dashboard for Healthcare Service Optimisation
Track & Know Enriched Individual Mobility Network
Track & Know Individual Mobility Network Extraction
Track & Know Data Enrichment Pipeline
Track & Know Vehicle Complex Event Extraction