An Open Source Analysis Toolbox For Street Network Comparison

thanks a lot so my name is on it because I had the pleasure of being able to presented in front of you today what I will be presenting some work we did in the beginning of this year so I have before a jury is 2 comma decimal 0 was out and it's about the comparison of is between OpenStreetMap and the official Austrian reference graph street network in Austria and I'll be showing you on the toolbox that we developed and so the results of the comparison for the city of Vienna I invite of we that that would be myself then come at this about who is our our OpenStreetMap by expert and Michael admitted that the passionate who is expert for DOS and street of Boston reference graph to give you a short

outline of what I was talking about there's a short introduction into the different concepts of street network modeling that those 2 straight graphs follow it's needed to understand a few of the results that I will be talking about later then I will show up which a comparison indicators have been used so far it there's quite a lot of studies on street network comparison notably a lot in the UK here but also in Germany and the US and France for example so they have already introduced some indicators which I have implemented also in the tool box and I will be showing the results will be then we introduce 2 new indicators which are specific for vehicle rooting so we look at the quality of street that but for vehicle group think it's important that the 1 way street directions are correct and that the mold and that you have the turn restrictions which apply for that we had a course industry that works so we have developed indicators to measure the correctness of those I will end with the case study for VN before the conclusions I'm quite sure for most a few the modulation is not something I have to dwell on for a long time student that works I just a really really important input data set for all kinds of analysis and at IT where I work at the mobility department which is of course a lot of analyzes with traffic data and you always knew the street you almost always need a street network to do any kind of a real word so of course in a research setting it's great to have OpenStreetMap because it's global you can collaborate with any entity all over the world on a project you can just share debate but it's free you don't have to mess around the strange licenses can I give my street that to my partner in university why set but of course there's always been concerned especially if the vehicle rooting part how far advanced is OpenStreetMap already and can we use it for a project so I found this quote from 2 us 12 applying to Germany which basically says that well yeah last night not that great yet there are not that monitoring restrictions in it like in the commercial datasets and it will still take many years so but decidedly just have a look at it ourselves and we can go on from there short sneak peek what was there in the end but if you know the want and she dubious and formal is expanding our to processing so you can do all you can create your own tools were to put in the tool box and then you can chain them together to get analyzes workflows and what I will presenting it is those that the tools of that where necessary to do the other comparison between the OpenStreetMap and the core bolstering graphs and they're all these tiny entries in the graph operations and graph comparisons you can get them on the top they all out there and straighten out yourself if you're interested I so back to the street mathematics OpenStreetMap I'm sure you're all familiar with everyone can edit everyone can use on the other hand there is the Austrian street network graph it's created by the authorities and currently there is really limited access to it you cannot even go there is a high I want to buy it no they don't have a business model yet so if you have to know someone who knows someone who can probably get you a copy of it but that will hopefully change anyways we we got our hands on a copy and that's but I will be talking about another thing I'm sure many of you are a bear of that for use of mobility and intruding applications-India rule the bone afterward and by default OpenStreetMap and it is not true the ball you cannot just take any export and run your bikes that holding algorithm on it it will not work in at the preprocessor to and that means on the lowest level that you have to splits they edges at the intersections because Institute and OpenStreetMap but it's perfectly fine to have an intersection model just with 2 edges instead of for which you would normally have in across intersection like this 1 by just having this 1 node in the middle shared between the 2 links that says there's an intersection and you can turn and up everything is fine but in to use normal eroding algorithms you will have to sleep the edges here make 4 edges out of it which share the note another aspect of a street

that as the classification of the roads which are important and which are less important and in OpenStreetMap there's the highway tagged with this sort names for different wrote classes and in many commercials street that directs the user concept called functional rolled class if R a C but also in those trends treatment work they use this and it's usually from 0 to a plus they introduced some others so 11 280 which are pedestrian ways or something like that which is not relevant for the study that I will be presenting so we put them side by side and decide which which classes match up approximately and that's the The result of this match matching process I will use these groups later on because they're comparable to show you some results

another important aspect in street network modeling of OpenStreetMap and be there there's a big difference is in how the driving permissions are modeled in OpenStreetMap you will laugh default driving permissions so on a highway it's usually not possible that you can cycle there that will be forbidden implicitly if you don't have to specified anywhere I and on and they're all string graph which is called the that's abbreviation I use every render that's because the real name is way too long they have at model that driving commissions using binary code so there's a bit for every mode of transport like for war going for cycling for a for cars for taxes there's because its own people and you will check for every direction of the road is a 0 or 1 can I go there cannot go there is the 1 way natively 1 and the 1 their actions here in the other direction and you have to kind of figure that out if you want to group and that graph so that's 2 completely different ways of modeling the same thing it's kind of similar was the turn restrictions by default in OpenStreetMap if there's an intersection you can turn that's the default behavior and then you can have explicit restrictions which say you cannot go from a to B but you can do everything else why you half commands which basically said here you can only turn from a to see no girls I in give it again everything is explicit they have a lot of entries in the list of turning relations and everything that's not in this list is forbidden so you can easily go for that list and I will show you

later how we use them so that was the basics of the street networks and not having know those we can compare the 2 and the 1st part will be the indicators which are used in the literature so far One basic indicator

concerns itself with positional accuracy in its 1st up being used in the UK study by Mr. Hatley and it calculates the percentage of the lower Juris representation so the net work you want to pass of if there are distance in the high curious ship of representation of moving a specified just as I will explain it using the image that's probably easier so you have a high accuser representation disciplinary lines and you specify certain of buffer distance in chair speech and you check from the network that you want to test how big is the percentage that falls within the buffer and is therefore positionally accurate and how much is outside the office so I've implement that in a tool for us extend the program processing for month and year and you can basically see the parts coming together here you can read the models of the top down there's always the the reference craft that would be OpenStreetMap in our case of the buffer size which you can vary depending on how generous you are with your definition of positionally accurate and then there's the 2nd graph which is compared to and I introduced the concept of regions so you can do 1 in 1 go analyze different regions at once it's just an area a layer which you wouldn't and stand it does although the buffering that to differences are at a faster rate to dissolve currently I found the dissolve operation takes just forever but if the 2 differences that runs much faster so we dissolve all the buffers and then we get the the line length within the buffer and the total line length and we can calculate the ratio that's basically the result of the of this too the a 2nd indicator

that is also values that just blew a fuse that is on there there even more they compare simply that then that America length per region so they make cells and then they checked columnist OpenStreetMap and FERC in it and how long is the other at a road network which we think is better or simply different and so here's for for nothing I found that in the high place that even the wasn't patterns and you can see there's some regions there always is is lower than that at the Ordnance Survey graph they used and the other way round implemented in the tool

box you have again 2 graphs that you put in you have those regions again and you compare the toll the graph lengthened out comes the difference between the 2 and the resulting layer will then be those cells with an additional attitude of the road links difference What's also

been used in some studies is just to look at educated completeness which of course is another important aspect of the street that works so they check out a whole big is the share of street which have a name applied to them or big the percentage the names are still missing on that's also quite easy in this case you only need 1 graph because it's not a direct comparison the are you input the classification field and then the tool will check which streets of there's a null in this in that field of for which there's some value in it and calculate the links for these 2 groups and compare them when it comes to the legal rooting features like restrictions and 1 way is what I found was all it 1 comparison by the German study which compared to the number of turn restrictions so you company account and there you say when there is less than that's of worse than when there's more but it's for the Austrian street network that will never gonna register explained how the modeling approaches are so did not work everything as explicit they would occur explain this is Rabin this is Rabin this is forbidden while in open street I OpenStreetMap you could just say at this intersection it's only allowed to you're only allowed to turn left so there would be only 1 entry instead of possibly 3 or more but it will have the same information it will be equally correct so just counting them does not work you have to find a different solution so before the sum of more rooting based graph comparison of approaches there's a 1st preparations that which is currently done with our in-house rooting software which extracts the other 1 how 1 way streets and the turn restrictions from both graphs and which is also used to calculate the shortest paths I will explain the concept in further detail on the next few slides and therefore of evaluation which follows is then again done in quantum Genesis extent them so the 1st

idea of what turned restrictions what we can take the list of turn restrictions and a Austin street that graph and the construct all those forbidden turns the forbidden turns in this graph are the black arrows so that's but you're not allowed to do it always starts in the middle of 1 worldling and ends in the middle of the next rolled things so you have an easier time matching between different street network representations because it is you put a test in an intersection it will just not know where to put the starting point for the route thing that we need to do and the 2nd the colorful error that you see that is the result of what happens when you route between the start and the end point of the Forbidden geometry and in the case where the other wrote that work has the same knowledge up above the turning restrictions it finds a way around the UPS obstacle and if it doesn't know about the Turn restriction will just go happily Gojira left because it doesn't know that it's not allowed to do it so we generate both those of geometric representations and then be compared and Miss if their tools similar it means that the Turn restriction was ignored it wasn't available in the dataset and 2 similarities decided to define using stuff distance which is in turn the tear in the model but which basically just says there maximum of the minimum distances between the 2 data sets the it's easiest up to imagine on the graphic I think it's the typical to explain anyway you get the maximum distance between the 2 and be used a threshold of 15 meters there's so if it's further than 15 meters apart that's different enough it probably went a different route it didn't begin order 1 with other turn restrictions so for the 1 way streets we used a similar approach may again extract the one-way geometries in this case just for a short part in the middle of the lines again that's the forbidden geometry so the site and the rule that again on the left side you see finds a way around the 1 they because it knows about it and on the right side it doesn't know about that 1 vendor just goes through that's the model that best that it just goggle at export and geometry columns calculates the length of the lines the chip line geometries and they compared because they extract always just tell me their short and if the of root resulting and a graph is much longer than we know that it's on the way around and that's the 1 way was modeled correctly in OpenStreetMap so that all

the things that we didn't theory if that's the

case study for the and if the bigger me region so there's a lot of smaller towns around the world and we look at it on that December 2012 the them for the position

Lecturer received that something I you can only do for select the probe so the look that like in the British study we looked at the highways in that region and bivariate that the buffer size so the definition of what this position on the accurate and you can see that the positional accuracy of most of that was pretty good even if it goes 0 only 10 meters of from the center line of the geometry and there's a few of figure deviations those are mostly only reaching into the analyzes area for really really short of parts and steric it's difficult to distinguish between all those ramps and how we call their classified and it just gets messy there and so I wouldn't give too much and those well use and then

if you remember I said that vector lengths that's what everyone does so we all the big network length but I will you can see here are the classes that I introduced before so there's the highways and trunk and secondary and a lot of ideas and to the tracks which they later told us that they didn't include into the export that we got from the Austin street network is everything that's unpaved just wasn't in the expert we received so we'll meet at the Group B for all further analysis and anyway OpenStreetMap this still longer in general all the classes show the same thing it's a bit longer than the Austin reference graph

and that's because it is of those reference graph is more generalized did get you can see here they would have just 1 edge where OpenStreetMap has so it's just on a different level of generalization so comparing the length is again something that shows an indicator book but it doesn't tell you better OpenStreetMap is better or worse than the comparison that word better get complete

lasts only texts names and speed values so the max speed in OpenStreetMap and as you can generally see that so they're at the authority of network is is more complete of though it doesn't mean that is correct still and they have a kind of veered names like und unnamed street they would put in a street doesn't have a name wrestles OpenStreetMap but probably wouldn't put anything and in general was that the higher last roads are better coverage that's just to get an overview of the missing names while that of 4 street network is pretty much complaint completely named in this in that the city of him it will still get some unnamed streets in in OpenStreetMap OpenStreetMap especially those which the god to those areas our industrial zones where it's probably difficult to find the work street name

information that turn restrictions we also count data so here's the other comes around 700 in OpenStreetMap and around 2 files and 500 in it so I already explained to you by my that is another really good comparison when we apply our rooting

based comparison then we get these results we see that in 95 per cent of the cases the one-way streets in OpenStreetMap and deep would match so if there is 1 in those reference graph refined that represented in OpenStreetMap and in 68 per cent of terror restrictions that's true as well we also went out for a small region and checked on site because my we only know the difference is we don't know who's right and who's wrong it could be the way so we checked in the 9th district of Vienna which is rather small area and there we found that the old OpenStreetMap errors while they were more common they were not novels errors rain OpenStreetMap there also arose in the official reference on that so in this case the official 1 had 20 to 30 per cent failure rate of those differences of course I also wanted to

look at it had looks in space and its in the print version you can then hopefully read numbers but there's a lot of what 1 way streets in the I always has 3 digits and it's a 1 by 1 commuter a greater but the darker the better the match so you see the light of cells around comparison shows that they don't agree those 2 street networks while in the city of the Allies spread a dark and they're great wealth same for 10 restrictions there's

generally fewer than 1 ways but also hear you say see the further you go outside the city that the less they agree bolstered networks and this is already going further than what you will find in the paper that goes with this talk but the continued over it is simple feature comparison to how do actually rooting results on these 2 whenever graphs compare and we chose the center of of those 10 by 10 cells in the center and the cup of calculated roots in both graphs so 10 cell beer-making 99 flowers fruits and with an average length of almost 7 kilometers and what you can see here is the distribution of length differences you would see if bundles of racial Austin at work but it's the angular meters long any of this distribution of length in open to the world and in general though as OpenStreetMap broads tend to be short term not much there's a means difference of 70 meters that could be because there's more sh more shortcuts are because it lacks some restrictions so you can just find truck the rules which in practice and not allowed but we need to look into more detail there of death

if you wanted to define how many roots are similar and difficult different thresholds how many would match and if you're very strict with all the 10 meter fresh old allowed it would be 16 per cent of those but in average seven-kilometre wrong long groups and did they go up to 60 per cent almost 400 meter fashion I'm rubbing up and

so we did a comparison of OpenStreetMap and the Austrian reference graph vehicle rooting for of focus on Eitel rooting we found that the OpenStreetMap Bourgain this analyzes area was of 6 17 per cent longer than the Official route network because it's less generalized but that appeared completeness there is some catching up to do the time restrictions and one-way street information match by 95 per cent and 70 per cent approximately and so you really gets already a good match off the root root length so if your applications only dependent on the route length a problem it doesn't make much difference which street that we use when you go down to the actual geometry of the rules that might differ still of course so that's a

toolbox there's a link to to keep up there you can check it out it's currently 406 stand this all we've quantum chairs 1 comma decimal 8 5 of present planning to port it to 2 comma decimal

0 processing and it's an open source of process for all the comparison parts but for the preparation there's still a bit lacking because that was fun of some in-house tools so there could be some chance for collaboration of someone would like to look into that and then of course we also interested in looking at other modes of transport besides cars so if anyone sees wants to talk about that later you can find me

around thank you but the

and very much and we got about this questions that you have and we I don't merge attributes I for that would have to do a one-to-one matching of the street links we haven't implemented that yet but it can be found in the literature it's explained how to do it I am very interested in doing that so yeah I it is definitely possible to some degree you will have to check how good the results will be thus the and you could of that of a host of distance and and the 1 of the worst there's still geometries basically and you compare between that geometries where's the biggest so minimum distance so you get from all the points in geometry on aid to all the points in geometry B and eject which is the closest 1 and you do that for all combinations and then for at least to take the maximum this is half of it and if they are part of thank you very much I'm just system