Exploring Local
Mike Dobson of TeleMapics on Local Search and All Things Geospatial

Comments On The Routability of OSM Data

February 16th, 2010 by MDob

First, let me say that I am a fan of OSM (OpenStreetMap) and hope that the venture continues its past successes. I am writing this mini-blog to reply to the large number of people who wrote to let me know that OSM data is routable. Rather than reply to each of you, I thought I could reply once in response to all of your comments (and I appreciated hearing from so many of you).

In response to my last article, Richard commented that OSM (and he provided a link to the OSM website) had the most up-to-date map of the I-195 interchange in Providence that I had written about. I thanked Richard and noted, “Although the OSM data is not navigable, it appears up-to-date….blah, blah, blah.”

In my response, I was referring to the OSM page and data that could be accessed though the link Richard supplied. It is my understanding that OSM through its website does not provide routing or navigation services. My comment seems to have caught the attention of a number of readers and since the topic appears of more general interest than most of my other topics, I decided to write a short reply.

The terrific comments I received about OSM data being routable started me wondering. Since others route with OSM data, why doesn’t OSM provide that service? Anytime you see a carve-out, you should strategize that somebody smells money and try to figure out where (as in Where 2.0)! In this case, however, the money (and I suspect there is not much of it) seems to be in providing routing software and associated services, since any augmentations of the data to allow for routing would become part of the OSM map database. There must be more to the carve-out than apparent.

Next, as you may remember, recently I asked Mike Moore about the level of attribution in NAVTEQ’s database after he made a comment that I found it difficult to accept. In the same vein, I would like to ask, “What percent of the road segments in the OSM database are attributed for navigation?” Let’s look a little deeper.

My interest here is in understanding the competitive threat of OSM to commercial providers of navigable database. For example, presumably, one-hundred percent of the streets and roads in the databases of NAVTEQ and TeleAtlas have been attributed for navigation and, presumably, the same is true for the Google-Mapbase. Before anyone starts in, I agree that the depth of attribution is unclear for the databases provided by any of the companies mentioned.

My interest in asking this question of OSM is that anyone can calculate routes between locations of coordinates shown on a piece of graph paper, so being able to route across an unattributed map is not of particular interest. If I introduce directional constraints and turn restraints (in a way that would mirror traffic laws in a community, or barriers or other impediments to navigation), I make the routes less efficient, but more realistic. OSM’s data structure, however, contemplates navigation attributes. Therefore, the question that I would like to understand about OSM is, “What percent of their database is attributed for routing?” From the perspective of the user, this question may convert into “How reliable is routing using the OSM data?”

I realize that OSM data coverage is increasing and that the comprehensiveness of the OSM database will increase over time. However, my interest is its status now in respect to routing.

For many of us, it seems so difficult to discover things about OSM, its data, and the use of the data. Maybe I just need to spend more time reading their Wiki (guilty). However, I admit that I am confused about OSM licensing practices and liability issues. Every time that I start to research these issues, I get a headache. While I think I understand the limitations of the license to use OSM data and why these “carve-outs” are necessary, I find it difficult to understand how to use the data to any commercial advantage and wonder if that will limit the usefulness of OSM’s contribution.

As I understand it, you cannot make any changes/augmentations to the OSM map database unless you provide those augmentations/changes back to OSM. For example, it appears that a company willing to collect road attributes related to the routing of hazmat materials through a country, could not use OSM as a database, unless they were willing to provide their hazmat related attributes to OSM. Since specialized uses of data may require non-crowd sourced data acquisition and considerable expense, it appears that cases like this limit OSM.

Here is a more concrete example taken, from the OSM Wiki about using OSM data in a SatNav System. The question listed there is “The user wants to use OSM data in a satnav system. They need to be able to convert the data into a confidential proprietary format that works with their routing software which they don’t want to disclose.” The answer provided is “Due to the “technical measures” section, a non-confidential, non-proprietary version would have to be redistributed also.” I am not sure how this notion works in a commercial environment, but maybe it does for some.

As many of you who read this blog know, I am a great supporter of User Generated Content and believe it will change the nature of mapping. However, there are areas where User Generated Content may not work. At a presentation on User Generated Content that I made to a group from the automobile industry, I was stopped by this response “Well, I certainly would not want to drive a User Generated Car.” Of course, the audience responded in whoops of laughter, but it sounds reasonable, doesn’t it? Perhaps the question translates to, “Do I want to be in a car being routed using data that is entirely user generated and for which no one assumes liability?” Sounds like an interesting topic for a future blog.

Once again, thanks for all of the comments, links and help.

Finally, I apologize for the delay in my response to your comments, but I seem to get up later than most of you. Of course, it is likely that I am in a different time zone than many of you. However, the real reason I have been getting up late is that after finishing my days work, I have been keeping my XBOX 360 and PS3 very busy playing Mass Effect 2 on one and Bioshock 2 on the other. By the way, you can call me “Big Daddy!”

Bookmark and Share

Posted in Authority and mapping, Google, Mapping, Mike Dobson, Navteq, openstreetmap, Personal Navigation, routing and navigation, TeleAtlas, User Generated Content

18 Responses

  1. Shaun McDonald

    Most of the people that responded in your comments are based in the UK, which is where OpenStreetMap started out back in 2004.

    I regularly use the OSM data for routing using my phone and an app call TrackMyJourney, while walking and cycling. I do keep my wits about me, as they routing may have an error (which I can go and correct and regularly do), or it could simply be that there has been an accident or there are roadworks and the road is closed, so I have to find an alternative route. I even have several different options for the routing options, depending on the mode of transport, and type of journey that I want (for example fastestest, shortest, cycling, walking, etc.)

    Thanks Shaun. I appreciate your comments. Today’s blog has more on the topic and near the end is a comment indicating that I suspected I live in a time zone later than many who read Exploring Local. — Mike

  2. SnowDown

    To answer your car analogy:

  3. MDob

    Thanks for the link, I learned something new, which makes today a good day in my book. I took a look at the OScar website and downloaded some documents including the OScar design pdf. I liked the one with the fluffy fenders! –Mike

  4. Richard Fairhurst

    Some really interesting points.

    One of the perennial problems for OSM, as a project, is how much to put on the openstreetmap.org website. At its core, OSM is not about providing an end-user service. It’s about providing the data for others to use, whether they be providing map tiles, routing, basemaps for GPS receivers, local search, or whatever. It’s much more a replacement for Navteq and Tele Atlas than it is for Bing or Yahoo.

    However, though this is fine in theory, osm.org does need to act as a showcase and that’s why there are several (good) renderings from OSM data there.

    Should it also include routing? It’s an unanswered question. No-one has yet come up with an open-source router that could be easily maintained on OSM’s servers, so it’s easy to answer “no”. But even if a wondrous piece of code were in existence I think there’d still be some opposition: people would say “that’s not what we’re about”. The financial imperative doesn’t work for a volunteer, limited-outlay project where most contributors are taking part for the love of it.

    On the specific point of “What percent of their database is attributed for routing?”; as with everything in OSM, it’s iterative. If the base connectivity is right, you can do fairly simple routing over any OSM data. This applies to almost all OSM data in Europe, but not in Hicksville, Tennessee where the dodgy TIGER data hasn’t been fixed up yet. Pretty much all OSM vectors are within one of several core classes (“primary road”, “cycleway”, “residential road” etc. etc. etc.) which also enables this basic routing.

    But whether any additional attributes (speed limits, restricted turns, etc.) are present depends largely on the enthusiasm of the local contributors. Though some places like London and the German cities will no doubt be tagged up to the hilt with this kind of data, I suspect adding this will remain a minority pursuit until routing becomes more common with OSM data, at which point people will want to get it right.

    “How reliable is routing using the OSM data?” Depends entirely what you want to do. Would I use it for a fleet of delivery vans? No chance. But for cycle routing, for example, OSM is still the only game in town: http://www.cyclestreets.net is a great example of how OSM can work where other datasets don’t.

  5. Peter Miller

    You ask some very good questions about the OSM licensing and liability.

    Here are my thoughts on the subject:-

    OSM uses a strict share-alike licensing model requiring people who improve the data to make their improvements available to the whole community. Is that good? Well if only one commercial company were to spend a huge amount of money improving the whole OSM dataset and then all other commercial companies could ‘free-loaded’ on their work then this would probably be a very poor business decision and is unlikely to occur; however, what if a company only wants a routable model for a limited area and the cost of doing that is justified by the commercial benefits to them when compared to the alternatives then it may well make sense for them to do ‘their bit’ even if others get access to their work as well.

    Let’s assume for a moment that there is no routing information in the current OSM model and that a number of companies want to offer routing across significant areas of the world. We have shown that it isn’t attractive for a single company to do all the work as indicated above, but what if three or more large companies all chose to improve different parts of the DB and share their work with each other and everyone else? The cost is now 33% of the cost of going it alone. If one also factors in the savings because much of the geometry is already collected the cost to each company is now possibly only 15% of going it along. The cost will fall as further companies participate and it is of course likely that the community will also be collecting data cutting the costs further. The cost of maintenance of data is likely to considerably lower as more people have an interest in improving the data.

    If I am feeling bold I would say that everyone other than Google who wants pedestrian routing globally may need to consider this approach. I say that because I just don’t see who has the cash to compete with Google on their own (and I included Nokia and TomTom in that statement although they could prove me wrong). So… are we not in a Microsoft Windows/Linux situation again where the only way to compete with the dominant player is together? And if so then are the OSM licensing rules not exactly what are required to ensure that people work together.

    So how much support might OSM get. Firstly lets consider how much commercial support it has already received. I would estimate that OSM has already attracted well in excess of $1million of commercial support from a number of players including AND (data for the Netherlands and elsewhere), Cloudmade (staff and event support), many government agencies (free data), Google/Yahoo/Multimap and others (sponsorship of conferences), Digital Globe (satellite photography for Haiti), Geofabrik and ITO (tools and data processing) and many others. Even if one ignores the free data the value of this commerical work is already considerable. I am waiting for the first year in which $10 million is spend by commerical companies to improve OSM. After all IBM and Google put huge sums on money into support open source software to challenge the dominate software players and allow them to continue in business. This cost is after all a lot less than was paid to purchase Navteq and TeleAtlas (£10billion).

    You mention liability… well that is also an issue for commercially collected data. This will certainly require suitable insurance, suitable terms of use and some approporiate due-diligence on the data prior to use in sat-nav systems however none of that will preclude the use of OSM data for commercial routing purposes. Thinking about it, I am sure that many of us already us trust open source software with our pensions, morgages and savings whenever we use a Firefox browser, a linux based router or when our banks use Linux servers somewhere within their operations.

  6. MDob

    Submitted on 2010/02/17 at 9:52am
    Great comments, Richard. Thanks for taking the time to help us out.

    You raise a question that I have been noodling on for some time. In a commerical map company you can direct your research staff to solve the problems most important to you (based on your market strategies and user feedback). OSM cannot “control” its users (nor can any concern using UGC), so the question of fixing things that are broken or incomplete would seem to rest on the preference of the users. The question then is, “Can you get to complete attribution for routing in some reasonable period of time and can you keep it updated?” I have wavered back and forth on this one. Thoughts, anyone?


  7. MDob

    Thanks for you comprehensive comments, Pete.

    I understand your argument on market dynamics, but am not sure that it will work out that way for OSM. Of course, if OSM and its supporters want to produce data that is routable, they will do so over time, but questions of consistency will be an issue during this effort. In this case, the interesting question is how long would this action take – which is why I asked the question about what proportion of the OSM map base was attributed for routing.

    I think the question about routing and liablility is very complicated. While users are forced to accept (click on) an exception statement to use your PND for routing and the Terms and Condiitions of all online routing services indicate the the routing instructions are advisory only and may not reflect real world conditions, it is less clear that any of these “limitations” would be upheld in a court of law. While online may have more flexibility in avoiding the issue, it will be hard to get out of the liablity if you are route guiding someone into a tree or off a cliff. In the case of data from NAVTEQ or TeleAtlas, or Google, we know who will be sued. However, I think OSM would be a challenge for litigators.

    I agree that many of us have great trust in Firefox and that much of our financial assets are represented on Linuz servers. If any particular systems fails, however, we will be suing the company that employed the technology, not the developers of the server software, as the ones who made unwise decision that harmed our financial well being.


  8. Peter Miller

    My main message is that ‘free is not free’. If companies put value on consistency and routeability of OSM data then they are going to have to put in some work to get it to a good enough standard so they can use it. There is this stange idea that just because the data is free and just because many people work on it for fun that it is not an appropriate place to expend commercial effort. The case to do this will only get stronger as the difference between what commerical users need and what OSM offers decreases. OSM has over 2K people making edits to the data each data at present, and my experience is that the vast majority of these edits are beneficial.

    With regard to liability, surely the Linux story tells us that most companies won’t buy OSM from ‘the crowd’ directly, commercial entities are likely to buy OSM data from a company that warrants it and does checks on their behalf and will be the recipient of any law suits. These companies will be close to the community and the data will be able to protect their users from potential problems just as Redhat help protect the banks from dodgy edits to Linux and from law suits. With OSM the name CloudMade comes to mind; ITO is also in that space and others are also planning such things.

    – Peter

    Thanks for your comments Peter.

    In respect to your thought of commericialization, I would add that commericial entities are often looking for a competitive edge and owning something that your competitor does not, such as custom data, may be a significant advantage in the market. I suspect that the desire to have an improved ability to target business address for advertising purposes than their competitors is what drives Google’s mapping efforts. Obviously, they are willing to let a wide variety of entities use their data, but they remain the keepers of the content. In short, there may be many valid business-related reason for not using OSM data, but I doubt that business strategists avoid these data because they are free or because people have fun while collecting the data.

    I find your thought on liablity interesting but troubling. How will CloudMade or any other business that does not have the resources of a Google, NAVTEQ or a TeleAtlas certify data that they do not create? Certifying map data as fit for a specific use is a very difficult proposition and one that is difficult to verify even if your company collected the data. I do agree, however, that it is more likely that companies will attempt to certify data as fit for use over smaller geographic territitories and that companies like CloudMade could become aggregators of the offerings of these companies to make map coverages that could work over larger geographic regions. But, I guess we will have to let the future play itself out before we know. –Mike

  9. Russell Nelson

    I agree with you that the OSM licensing is a horrific mess. The only clear path forward is to put it all into the public domain, however a significant (and loud) portion of the OSM community is actively hostile to that idea. They think that a static dump of the data has value without them. They underestimate the value of continued participation in the OSM community. Several software vendors have tried to maintain Linux kernels separate from Linus’s. It was a nightmare. They gave it up. I’ve tried to maintain my own version of KA9Q’s NOS. It was a nightmare. OSM without the OSM community would be a nightmare.

    I’ve tried, face to face, to convince Steve Coast that OSM should go public domain, but he’s 100% convinced that ($GEODATA_VENDOR) would take a copy and never contribute back.

    Russel – thanks for the interesting comments. I am not sure I can see far enough into the future to understand how the licensing should play out, but I agree that what exists today is hindering growth and this is never desirable. I suspect that there are major differences of opinion within the OSM community and even within CloudMade. It will be interesting to see how the future rolls out for OSM and licensing. — Mike

  10. Marc Prioleau

    Good and necessary discussion. One thing to look forward too: In a few weeks Skobbler will launch an OSM based nav system for Germany, moving off NAVTEQ. Customer reaction (the ultimate determinate) will be interesting. Germany is a very sophisticated naviagtion market (having been at it longer than most other countries) and it is also the most complete OSM data set. So the real question will be how to measure customer reaction? Change in ranking in app store?

    Hey Mark, thanks for the comment and the information about Skobbler.

    Your measurement question is an intersting one that I had not thought about. I am not sure that App Store ranking would mean very much since it is so hard to evaluate applications on the basis of the information available when you buy them, although I suppose that the Skobbler app will get a lot of press and reviews that might help pre-sell buyers. A measure of how often the app was actually used might be more useful, but Skobbler may be the only one to know those stats. Certainly an interesting subject to ponder.

    Are App Store rankings available by geographic region? If not, Skobbler will have a hard time ever equalling Doodle Jump. Thanks again. –Mike

  11. Russell Nelson

    Oh, and for people laughing at “User Generated Cars”, there are a lot of people who would build their own cars. Trouble is that it’s very difficult to get permission to drive such a car on the roads. Typically, then, what people do is make a three-wheeler, which is legally a motorcycle. Since everyone knows that motorcyclists are insane, and at least halfway to death already, nobody really cares what kind of modifications you make to a motorcycle.

    Thanks Russel. I think the comment that was made to me and recounted in the blog was one in which the person meant that the process would be like listening to a group of eyewitnesses report an event witnesed by the entire group – they all would have different opinions on what they saw. While I am sure you or I could build a car that might be pleasing, if we asked 2,000 people to build the same car, we might never see the end result or want to drive it if it was ever produced. — Mike

  12. Matthias

    @Russell Nelson:

    German user can’t put there data under PD. putting OSM into PD would mean that you have to delete all contributions from german users.

    Hey, Russel, thanks for that clarification. –Mike

  13. Bobby Parikh

    First a quick intro and full disclosure: I manage the map data team at deCarta, a company that provides mapping/navigation/search platform and services for mobile and PND spaces. We have been experimenting with OpenStreetMap data for some time now. Some of you might remember our beta launch with OSM data last year. Our experience has been similar to a lot of other people who have commented here and elsewhere in creating a commercial offering using OpenStreetMap data.

    I’d like to comment on routability of OSM data (as opposed to whether OSM should provide a service for routing). When it comes to display, search, and pedestrian/bike navigation, the level of attribution in OSM is good. You could caveat that claim with coverage inconsistency, but that’s not the main topic here. Vehicle turn-by-turn routing is another story. Traditionally deCarta has supported commercial map vendors, such as NAVTEQ and Tele Atlas, and we still do. So naturally we were interested in determining the type of TbT routing applications one could build using the OSM data. I know it might make some folks mad, but IMO OSM data is not ready for TbT routing yet. Here are some numbers: Turn restriction count is perhaps one of the most important features for determining routability of any dataset. There are about 12K turn restrictions in the OSM database worldwide (Source: http://osmdoc.com/en/tag/restriction/#values). Slightly over 5K of those are in Germany alone. The US has about 1,300. By comparison, the traditional map databases have over 350K turn restrictions in the US alone. 1300 vs 350K…0.37%. To have really satisfying TbT guidance, that number has to be a lot higher. You may disagree so let me know. Believe me, I’d love to be wrong on this, but I don’t think I am. So get out there and include those turn restrictions!

    There are some challenges to capturing this data for OSM enthusiasts like us:
    – How do we get non-experts to move beyond capturing POIs and such simpler features to this sort of advanced content?
    – How do we get consistent modeling from users?

    [Side note: For those who didn’t know, User:Langläufer has built a cool visual analyzer for restriction relationships — http://osm.virtuelle-loipe.de/restrictions/

    I’ll posting our thoughts along these lines on our devBlog soon to coincide with our commercial product using OSM. We are trying to build a sort of readiness matrix. I’ll try to share that with the community.

    Bobby – Thanks for the great insights. Your comparative statistics will be very helpful for anyone hoping to benchmark the readiness of OSM for route guidance. Thanks also for the links. Finally, your question on how to get consistent modeling from users is a crucial item. Another issue revolves around how you might “direct” the efforts of volunteers to resolve complex spatial data quality issues that “break” routes. — Mike

  14. Valent Turkovic

    Great discussion!

    There are quite a few OpenStreetMap online routing services, you can check out this list:

    I know of a few iPhone apps that provide routing services with OSM data. I use OffMaps on iPhone and it works great in my hometown (Osijek, Croatia) in which I contributed most of the data. From what I have seen most iPhone developers use some online service to provide routing in their apps, they didn’t create their own routing engine. Most of them use OpenRouteService or Cloudmade as routing service.

    If you compare Offmaps vs iGo on iPhone it is 2$ vs 90$ price vise, developers of iPhone apps have much lower barrier of entry in this segment because OSM privides the data and maps. Of-course iGO has some better features (like turn by turn navigation) but I don’t see that it has better maps.

    The biggest issue for Turn By Turn (TbT) routing currently with OSM data is the lack of support for turn restrictions. None of online services supports turn restrictions! :(
    Only software I found that supports turn restrictions, and therefor gives correct and legal routing info, is Gosmore (http://wiki.openstreetmap.org/wiki/Gosmore).

    I started adding turn restrictions in my home town because to have accurate routing info in my home town but as soon as I realized that none of online services use turn restrictions data my enthusiasm was gone and I saw no point in adding new turn restrictions.

    Please, please, please if you are working on any of one online routing services please implement turn restrictions ASAP, pretty please 😉

    Could gosmore be used for online routing service because it already supports turn restrictions?

    As for liability – I got a Mio Map car navigation device that uses proprietary maps, after users turn it on for the first time they get an licence agreement that they must agree in order to use this device. Agreement clearly states than there is no liability, and that users can’t sue any company for anything and that users are to blame if anything happens while using this device.

    Hi, Valent, thanks for your comments.

    In regards to the liability issue, I am sure that all companies providing routing services are advised by their legal department to include a rejection of liablity notice. I suspect the “no-liability – It’s your fault” statement is about as strong as the agreement you sometimes must sign before you see a doctor, indicating that you will not sue if things do not work out.” When malpractice is involved, the agreement goes out the window, as it is considered non-enforceable by the plaintiff’s lawyer, and usually by the court.

    If the PND manufacturer created a faulty routing engine (the functionality was broken and routed you the wrong way down one-way streets) or the company knowingly included data that did not support the functionality of the device (data that lacked turn restrictions), they (or the data provider) would be liable for negligence in most courts of law. Put another way, consumers should be able to reasonably assume that a device advertised as capable of routing them between destinations while their car is moving, should be able to expect that the database and functionality of the device are capable of performing the task by providing “legal” maneuvers when the car must change direction to navigate the path provided by the device. If the directions provided by the device are not “legal”, there is a liability question.


  15. Apmon

    Bobby – Those numbers are very interesting and roughly match to what I would have expected currently. I would agree, in my experience attribution for routing, such as maxspeed values and turn restrictions, is often still a bit of a problem. I think one of the reasons for this though is that these attributes are neither visualised on any of the much used maps (as it is not features to show on standard maps) nor were they supported by most routing engines that did exist for using OSM data. Therefore motivation of volunteers for including such attributes were fairly limited as they don’t get an “immediate reward”. So good visualisations (that are well known in the community) of routing attributes, like the restriction map you linked to (which unfortunately isn’t well known) or support in much used navigation apps like Skobbler or deCarta, can quickly change this. At least in areas where there is already a vibrant community, like Germany or the UK. The US, however is a bit of a different story. Due to large scale imports like TIGER, the data coverage appears much more complete than there is a community to support it yet. Unfortunately the TIGER quality is very poor for routing, no routing attribution at all and large connectivity problems. So it would be nice if you could do the comparison on routing attributes on Germany too, which is probably the “showcase piece” of OSM. Even there routing attribution is by no means “complete”, but from my own very limited experiments with comparing the routing of OSM data using GpsMid to that of a Navigon PNA around Karlsruhe, Germany, the routes provided were surprisingly similar. And that despite the fact that the routing algorithms of GpsMid (saying it as one of the programmers of GpsMid) are unlikely to be as good as those used in Navigon. So I think routing with OSM definitely has potential, if we can get the right tools out to expose those routing attributions to the community.

  16. ij

    “By comparison, the traditional map databases have over 350K turn restrictions in the US alone. 1300 vs 350K…0.37%”

    Does 350k include restrictions which are implicit in OSM data such as not being allowed to access psv only ways or oneway streets in the opposite direction?

    Thanks IJ – I don’t know the answer, perhaps Bobby Parikh will know. — Mike

  17. Bobby Parikh

    Apmon-I’ll do what I can (legally and workload-wise) to share some of the stats later on deCarta’s devBlog.

    ij and Mike-The 350K count includes only turn-restrictions (vehicle- and time-based included). What you’re referring to are link-based restrictions. For a model to be complete, it needs to have both. Imagine a four-way intersection between two roads–one being perfectly vertical and the other one perfectly horizontal. Let’s name the links with the four cardinal directions. If the South link has a PSV-only restriction, then the complete model would require the following:
    1. A PSV-only restriction on the south link, and
    2. No turn for vehicles other than PSV from the East and West links
    3. No continue for vehicles other than PSV from the North link

    The 350K count includes only turn-restrictions of types 2 and 3, not 1.

    Sorry for the long explanation. I just wanted the answer to be complete.

    Thanks for your comment Bobby. It was helpful. –Mike

  18. ij

    Thanks for the answer Bobby,

    Based on that it seems that we would be able to infer much more than those 1300 restrictions, assuming access=no and psv=yes markings would be correctly set on all “South streets” (that of course depends heavily of the state of the map in a particular area, I’m not claiming they would all be there :-)).

    …I was whole the time thinking of turn restrictions anyway. There are two types of them:

    a) those which are inferrable from link-based restrictions (all restrictions in your example were such). I don’t even see any point in adding these as relations in the first place.

    b) extra restrictions, ie., e.g. not allowed to turn left even though straight crossing of the street from right to left would be fine, etc.

    To allow a fair compare of the number, it seems that the implicit restrictions would have to be deduced first.