Artificial Intelligence and Machine Learning with ArcGIS

[Music] artificial intelligence and machine learning are some of the most important and fascinating concepts in today's world many organizations and most disciplines are exploring how these concepts can help us expand the reach of our capabilities and as we explore how those capabilities help us solve spatial problems it can be a difficult task to unpack all of the new terms and concepts so today we're gonna break it down into three main items at the broadest level you have artificial intelligence which describes how you can have a computer or a machine do a task that requires some level of human intelligence now one type of engine that can make this possible is machine learning and that refers to data driven algorithms or really techniques that learn from data to get you the information that you need and one type of machine learning that has emerged in the past few years is deep learning and that refers to code structures that loosely resemble how the human brain is adaptive when you solve problems so there's a variety of use cases for these concepts some of these you may have heard about like self-driving vehicles or recommendation engines in ArcGIS we apply machine learning to do spatial analysis we have a history of applying machine learning algorithms and to our geoprocessing tools to solve problems and three broad categories with classification you can use support vector machine algorithms to create land cover classification layers another example is clustering we have a new density based clustering tool allows you to process large quantities of input point data into meaningful clusters from sparse noise and with prediction geographically weighted regression allows you to use geography to calibrate those factors that help you predict now with respect to the recent external interesting deep learning frameworks like tensor flow and scikit-learn or entire platforms like IBM Watson and Amazon Sage Maker the strength of our API is like the arches API for Python allow integration to happen and your capabilities to expand so today son and I would love to share a few examples of ArcGIS and machine learning in action here's Kristen Hogan this data reflects over 700,000 traffic accidents in Virginia DC and Maryland in 2017 so how do we make sense of all this data let's try to identify the most dangerous traffic intersections one way to do this is by using the new density based clustering tool this tool will help us find where our loudest signals are and our noisy data by separating our accidents into meaningful clusters and sparse noise the two algorithms we'll focus on today are DB scan and h DB scan let's tune into some of these clusters with DB scan we get the top 100 worst intersections these clusters correspond with many of the intersections in Baltimore of which North Avenue seems to be an area of concern now let's bring this closer to where we are here at the conference center these are some of the dangerous intersections in Washington DC now it makes sense that these clusters occurred intersections right but I'm not just concerned with intersections I'm also concerned with other places with different densities of events HDB scan can help us with this now here is where machine learning gets very explicit this algorithm requires little user input and is the most data-driven of the clustering algorithms to the point where it learns to define its own cluster and we could see this along Massachusetts Avenue with a cluster now spans several city blocks versus just at the intersection finding several clusters in a city is expected but what does it mean to be a cluster in a suburban town let's travel to Hagerstown Maryland it's reasonable to find the cluster in downtown but my eye gets drawn to these teal clusters with HDB scan the clusters coincide not only along the interstate but within the park laughs if I want to understand the clusters of accidents only within the parking lots I can use another machine learning method called image classification by means of support vector machine or random trees utilizing Chesapeake conservancies image classification layer we could separate those clusters of accidents in all parking lots versus along the interstate to produce a new list of the top 25 worst parking lots this is just a quick tour of some of the machine learning tools within ArcGIS now I'm going to pass it over to Alberto with an example of integration Alberto so let's take a short trip to a relatively safer intersection in Jackson Hole Wyoming we're at the top left you see a live video feed of this traffic intersection in the middle of town and at the bottom left you see the interpretation of this activity happening by a deep learning model that was trained to detect cars buses trucks and people so let's take a few seconds just to observe what's happening in Jackson Hole today and it's morning rush-hour I do have to warn you this is a live video feed and we're not fully sure what we're gonna see in the next few seconds thankfully it's a it's quite a few vehicles and if we get lucky we may catch a few people crossing the intersection or maybe we can catch another bus cross that would be interesting and there's a few people jaywalking actually very nice so it's a small town and it's a single intersection but it's a real-time event feed and that can be a lot of information to process so in the far right of the screen we have an Operations dashboard that receives a snapshot of the activity every six seconds and behind the scenes we have the ArcGIS API for Python orchestrating the entire operation so it'd be interesting to expand this concept across the town or across the city right so we've brought this concept home and expanded the scale to all of Washington DC by partnering with traffic lands Network of a hundred and eleven traffic cameras across the district we're monitoring activity right now and the size of the point represents the amount of activity in the district where the color of the point represents the predominant type of activity so let's take a look at some of the busiest intersections in the past few minutes we're looking at Georgia Avenue and Arkansas Avenue right now and let's take a look at some of the activity at that intersection and actually just updated so let me refresh that let's go to our cars Scott Circle actually just update it with 16th Street let's let that refresh there it goes and we detected 17 separate vehicles at that intersection it's not too far from here so let's take a look at the second busiest location that's Pennsylvania Avenue and M Street near Georgetown it's not that refresh we're going to receive an object detection from this location briefly there we go and we detected a bus and 17 cars there's not a great deal of pedestrian activity across the district right now but coincidentally two weekends ago we found what seemed like an anomaly four separate locations along Constitution Avenue reported that the predominant type of activity were people and I had never seen this before it made me wonder is there like a protest are there a lot of tourists all of a sudden and no coincidentally it happened to be the same Patrick's Day Parade on that Sunday at that time and nicely labeled and green it was pretty interesting to see real-time event feeds alert us of an event but where it gets really interesting is when we start comparing real-time information with historic trends at each location this is a map of truck activity trends across the district where locations in green signify that there's normal truck activity and the locations in red signified that there's a spike in the amount of trucks at those locations so we're still learning about what normal activity for each of these locations means but with that information and a comparison against real-time event feeds we can produce information intelligence notifications and a call to action so this is a brief example of how the integration between deep learning and ArcGIS allows your capabilities to expand so next Kristin and I would love to share some of the work done by our colleague mansour rod so in the previous example we trained the deep learning model by explicitly labeling the patterns in the data there's also a class of algorithm that lets the Machine discover patterns in the data without human labeling though this is called unsupervised learning and we use this learning mode to solve the following problem an oil and gas operator that wants to generate optimized inspection routes to their remote oil wells but they're missing street segments to generate these routes and a solution is to manually digitize those missing street segments but as you could imagine that would be a very time-consuming task so another solution is to use the GPS breadcrumbs from previous visits to those remote sites now if we look at the breadcrumbs of just a few vehicles it's difficult to deduce a useful pattern right however if we look at the breadcrumbs from multiple vehicles from multiple days then our useful pattern starts to emerge and now the first thing thing may be to connect the breadcrumbs sequentially by time but if you go through that process you're actually going to get what is a mess so let's take a look at what that looks like not very useful so we solved this problem using an example of unsupervised machine learning algorithms first we find the breadcrumbs that are not on the existing roads second using density based clustering we group these breadcrumbs to form clusters and third and this is our favorite part for each cluster we apply a self-organizing algorithm to map a set of lines to points so let's demonstrate this in action do you see how the machine is mutating the lines to fit the points let's play that one more time so you can really catch it in action it's pretty amazing isn't it it even discovered the Turner ants and finally if we snap the new roads to the existing roads to form a complete network we complete this data set so in conclusion using unsupervised learning and self-organizing maps we took a collection of GPS breadcrumbs and converted it into navigable street segments what we demonstrated today are just a few of the many machine learning algorithms that you too can use with the ArcGIS platform today thank you you

Channel: Esri Events

Views: 26,155

Rating: 4.9455781 out of 5

Keywords: Esri, ArcGIS, GIS, Esri Events, Esri 2018 Federal GIS Conference, FedGIS 2018, maps, Artificial Intelligence, AI, Machine Learning, ML, Kristen Hocutt, Alberto Nieto

Id: Cm_oAaQVWZ8

Length: 11min 46sec (706 seconds)

Published: Wed Mar 21 2018