Intelligent TransportationAnalytics With Google Cloud

Patrick Dunn Customer Engineer Years of steady improvement in highway safety is over

60,000 4.00

3.50 50,000 37,461 3.00 40,000 2.50

30,000 2.00

1.50 20,000 1.00 10,000 0.50

0.00 0

1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015 Fatalities and Fatality Rate per 100 Million VMT, by Year, 1975 - 2016

Sources: FARS 1975 - 2015 Final File, 2016 ARF, Vehicle Miles Travelled (VMT): FHWA

Roads are busier, heavily congested...

...and dangerous

Current trends show that by 2030, road The average commuter in metropolitan areas experience traffic injuries will become the seventh 1 4 hours of road congestion every day. leading cause of death globally.2

U.S. Department of Transportation, Federal Highway Administration Congestion Trends Report 2. CDC The solution isn’t building more roads

State and local government construction costs rose 13% in the last five years1

It’s harnessing your $31.65 billion data to optimize those roads

1. U.S. Census Bureau, Seasonally Adjusted Data But that data exists in silos, making it difficult to use Disconnected data is costing us too much

NSC estimates the cost of motor-vehicle More than 37,000 people died in motor deaths, injuries, and property damage vehicle crashes in 2017.2 in 2016 was $416.2 billion. 1

1. National Safety Council 2. US Department of Transportation’s National Highway Traffic Safety Administration What if your data could be the driver of traffic Enhance situational awareness management? based on traffic patterns

Coordinate response based on real-time insights

Predict road and device maintenance Your data can drive important decisions...... now and in the future

Where do I need to reroute traffic?

Where are crashes occurring most often?

Which roads will need the most repairs in the future? Data-driven insights make for a safer, more informed community Managing data volume and speed on traditional platforms results in...

Higher up-front, operational Higher risk of failure and maintenance costs

*Gartner How Capable is Your Data Enterprise?

1 2 3 4

Storage Program Data Warehouse Batch Processing ML for Real-time Reporting for Big Data Decisions Standardized retention Analysts work out of one or Distributed processing Machine Learning models windows across data more data warehouses frameworks allow actively deployed to stores; granular, historic used to build reports from organization-wide data to be increase enterprise data can be retrieved in a aggregated data used in complex analytic efficiency few days processing

11 Organizations are Struggling to Move 1 2 3 4 Storage Data Batch ML for Past the Traditional Program Warehouse Processing Real-time Reporting for Big Data Decisions Warehouse

12 The Traditional Warehouse is not Enough

● Minimal support for realtime data ● Coarse-grain aggregates used to compensate for scaling complexities ● Prohibitive licensing costs and terms ● Multi-tenancy issues lead to new data silos

13 Supporting Sustainability

Google datacenters already have half the overhead of typical datacenters

Applying Machine Learning produced 40% reduction in cooling energy

Confidential & Proprietary Building the Data Platform for Modern Problems

Endpoint clients Data Studio

Apache BigQuery Beam 3rd-party Data Mobile BI Tools consumers PubSub Dataflow User & device data Or Or ML

Web Apache Apache Kafka Spark IoT

BigTable

IoT Ingest Transform Analyze Why Google Cloud for Solving this Problem? 15 Years of Tackling Big Data Problems

Open Source

Map Google GFS BigTable Dremel Flume Java Spanner Millwheel Dataflow Tensorflow Papers Reduce

Google Cloud Products BigQuery Pub/Sub Dataflow Bigtable ML

2002 2004 2005 2006 2008 2010 2012 2014 2015 2016

17 Our data analytics design principles

Focus on analytics Develop comprehensive End-to-end ML Innovation and not infrastructure solutions lifecycle proven results Serverless analytics for complete ML lifecycle Innovators in Transportation

Google Maps Waze Waymo Offers visualization, navigation, Exchanges publicly-available Aiming to bring fully self-driving and analytics incidents and slow-down data technology to improve mobility What is DAISy?

Data Analytics Intelligence System (DAISy) is a cloud-based data analytics platform that brings intelligence, efficiency, and interoperability to CDOT’s existing transportation network while enabling world-leading roadway operations for a safer, more reliable, connected, and autonomous future.

Enabling New Applications

Public Traffic Safety Winter Weather Messaging Management Patrol Operations Center

V2X Mobility on Freight Signals Applications Demand Platooning & Movement

Management Variable Intelligent Future Dashboards Speed Limits Transportation Applications Systems Building DAISy with Data Two major categories

Geospatial Streaming Video Organizing & Enriching Datasets

Speed Weather Incidents

✓ Speed Readings ✓ Roadside Weather ✓ DOT Road Events Historical ✓ Speed Aggregates ✓ Highway Messaging Legacy Datasets ✓ Speed Radars ✓ Roadside Weather ✓ DOT Events Real-Time ✓ Construction Data IoT Data

✓ ✓ ✓ Waze Jams NCAR Pikalert Waze Accidents New ✓ Weather Data ✓ Snow Plow locations 3rd Party Datasets Making Data Accessible

Zonar Events Speeds Road View Level Conditions

DAISy Mart Logical Level

Waze Alerts Internal Level (Source Systems) Zonar Pikalert Data Warehouse Waze Jams

* example Bringing it all together Geospatial Data

Open source tool for massive Open source tool for sharing geospatial querying. geospatial data.

● Storage in ● Uses GeoMesa datastores ● Processors in Spark ● Runs on ● Transforms, indexes, and stores ● Publishes data to any geographic geography data for rapid access. data standard, including GeoJSON Road Segment: set of coordinates representing a stretch of road Joining Waze Incidents to Road Segments

● 600,000 road segment coordinates ● Millions of Waze Incidents Event: DOT-defined event, ● Join using distance between Waze Incident and incident, or road closure nearest road segment coordinate

● Group together Waze Incidents and DOT Events ● Better overall picture of highway system health

Waze Incident: Accident or other traffic impeding event Enabling Situational Awareness & Replay

Video Intelligence & Analytics Cloud AI Video Intelligence

Improving traffic management by adding intelligence to CDOT’s legacy cameras. Two types of AI building blocks

Video Intelligence API AutoML Pre-trained ML models Custom ML models

Leverage Google’s predefined dataset Train your own custom model with an to automatically detect a vast number of easy-to-use graphical interface. scenes, objects, etc. No coding required Coding required Video Intelligence Streaming API Real-time video analysis for live video and archived data

01 02 03 New

Batch Data Streaming Live Streaming

Annotate large video archives stored Annotate videos by splitting the Annotate live video feeds to take in Google Cloud Storage. video data into chunks and action immediately. Current support streaming each chunk using gRPC. for HLS, RTSP, and RTMP. Video Intelligence Streaming Features

Detect/track objects Monitor road conditions Detect scene changes Recognize road signs

snow

Specify region of interest 24x7 live analytics Hybrid cloud solution Ingestion Library Streaming Video Intelligence Hybrid Architecture

1 Capture

2 Ingest Analytics 3 Analyze / Store 4 Visualize / Activate Pre-Trained Models Video Intelligence API Cloud / On-Premises Application / Visualization Live Object Tracking Annotations Application Live Label Detection Cloud Pub/Sub App Engine Ingestion Live Shot Change AIStreamer Custom Models AutoML

Application Storage Visualization Video Storage Cloud Storage Solving real-world DOT problems

Cost Road Predictive Safety Reduction Optimization Maintenance Improvements Video Intelligence Demo Video Intelligence Streaming Service

Estimate vehicle speed

Identify vehicle types

Observe traffic abnormality

Monitor weather conditions

Video Intelligence Analytics Camera Detected Vehicle Counts ● Hourly Totals By Day ● Hourly Average by Weekday Data points * 4 cameras over 13 days ● >60M Labels ● >21M Objects Supporting CDOT’s key performance measures

Increased safety Improved social Highly responsive Collaborative & mobility equity, roadway information sharing affordability & air operations & communication quality with the public Questions?