Healthy prospects for floating vehicle data systems

Elmar Brockfeld, Alexander Sohr and Peter Wagner from the German Aerospace Center's 2202 Institute of Transport Systems look at the prospects for floating vehicle data systems

Although Floating Vehicle Data (FVD) or probe vehicle fleets have been around for about a decade, the idea behind them is of course much older: from probe vehicles that flow with the traffic it should be possible to get a precise, fast and spatially near-complete picture of the prevailing traffic flow conditions in an area under surveillance. From initially small and specialised fleets (such as taxis) with small communication bandwidths, the technology has now grown to include fairly large fleets with some thousands of vehicles, a development which is supported by the advent of modern communication technology and its ever-decreasing costs for bandwidth. Estimation of traffic states from FVD is meanwhile being supported by several other sources of data. In the near term, probe phone data is perhaps the most prominent (but technologically, especially in terms of data quality, also the most challenging) example. However, we cannot forget the rising technologies which will provide us with Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) communication and which have the potential to deliver an enormous amount of additional data in the future. Furthermore, in addition to traffic service providers which receive data from their own and probably a couple of other large fleets, smaller services are coming up which collect data from increasing numbers of private users, for example over some iPhone applications.

The fundamental principle is to use speed estimates from the probe vehicles. This can either be via a direct read-out of the GPS signal, or (a bit more involved, especially in the case of large time intervals between subsequent GPS readings) by building a route in the network between two subsequent vehicle positions. The latter process is an iterative one, since it requires the prior existence of some form of speed map of the network under consideration, however it is much more general and works for all kinds of GPS data; it builds the relevant travel time information for the time sequences for all road sections on a route and not only those 'hit' by GPS data.

As the collected data are from single vehicles and thus only fractions of the hundreds or thousands of vehicles per road section which make up complete traffic streams are measured, the shares (or overall fleet penetration rates) are of major importance. The more data, the more precise and valid the generated information will be. Dependent on this, innovative applications are possible which are in parts already running today.

Obvious applications

The simplest application of an FVD system is certainly the provision of most recent travel times. For companies in the navigation market (for example, 1692 TomTom, 490 Garmin, 2204 Navman and others) and data providers (such as 745 ITIS Holdings), who mainly deal with travel times, the field is definitely mature as a result of their using their own fleets and supporting cellular phone data. The information gathered and processed is used for routing purposes to provide most recent traffic states to navigation devices. However, it seems that in particular the most advanced and most interesting part, which is the use of all available data to provide users with the most up-to-date traffic information (and but one example here would be the HD Traffic service from TomTom, however there are others), does not work as well as these companies had hoped. The traffic and route information which is generated has reached a point where it is timely and of good quality but still certain types of information (such as those concerning traffic jams and disturbances) seem to be missing. This may change as the penetration rates get higher, which will mean that more and more GPS-equipped fleets deliver their data for generation of traffic information and users of other GPS-capable devices subscribe to services.

More specific applications

Apart from the travel time applications and anything that may be derived from that, other rather more spectacular applications are slowly coming into focus. For instance, for local authorities real-time traffic management is the most important but also the most difficult application because it needs a huge number of probe vehicles. While a fleet penetration rate of less than 1 per cent can already be shown to be sufficient for reliably estimating travel times, such a low level of penetration can only support the simplest of traffic management applications. For example, it is possible to generate from daily FVD travel time information curves that show the same structures in traffic flow which are detected by local measurement. These traffic counts can be used to control traffic lights. However, because of the relatively low penetration rates encountered at present it is not possible to achieve 'online' management whereby the signal control may be directly based on most recent data. That said, the application may already be of value in an 'offline' mode whereby a signal control strategy for a particular day is programmed or adjusted by estimates for the day based on historical data (gained from the day(s) before, the same weekdays before, the same event day in the year(s) before, and so on. This procedure is also possible with data from loop detectors but as these are typically not installed for each signalised intersection, FVD opens up the possibility to adopt it in the whole street network.

The field of true online traffic management applications such as the operational control of traffic signals needs fleet penetration rates of more than 20 per cent. Beyond this, the FVD-based methods reach the same quality levels as the traditional methods based on loop detectors. However, these methods benefit strongly from fusion with other sources of data. It may be interesting, then, to implement them in the near future (at least to a limited extent) and learn from such prototype implementations. The 2206 German Aerospace Center's AIM (Application Intelligent Mobility) platform supports (amongst other things) prototypal implementation of such methods.

Users' concerns

While for online applications the field is not yet mature, for offline planning applications it most definitely is. But from local authorities there are still a lot of reservations about FVD. A first big problem seems to be that in contrast to traditional traffic state detection, which delivers counts, local speeds and local traffic flows, FVD delivers travel times and travel speeds, which are not 'usual' measurements. A second problem is of course that not all data from all vehicles in a traffic stream are measured but only some, from which the state of a complete traffic stream is then calculated. This drives the necessity to describe the quality of the results in detail related to the penetration rates (which may differ a lot spatially as well as during time). Furthermore, it is especially difficult for local authorities to assess the final costs: is it really going to be cheaper, for instance, to at some stage in the future buy the (now necessary) data from a company which effectively controls the market? And if this answer turns out to be affirmative, it is still not completely clear what can and cannot be done with these data. Especially when it comes to hard online control of traffic and traffic signals, there is still a technology gap between what is provided today by loop detectors and what is possible with FVD and 'friends'.

Besides the fact that for online applications the data as available today are often too sparse for online traffic control, there is big potential, especially for local authorities, to analyse historical data for evaluation and planning purposes. The data may be aggregated over periods of some weeks, months or even years and then compared. With this, negative as well as positive developments in traffic flows on the street network may be systematically assessed and the information can be used for infrastructure as well as for traffic management planning. Such comparisons can also be used for classical before/while/after analyses concerning any traffic management or infrastructure activities (change of speed limits, construction sites, changes of turning lanes at an intersection and so on). Especially for intersections analyses with speeds and travel times may be extended by the calculation of average spill back jam lengths which makes an evaluation and observation of the performance of traffic signal plans possible.

Conclusions and future prospects

On a panel discussion at the NeArctis workshop in June 2010 in Lausanne, Siebe Turksma, product manager, research at 101 Peek Traffic BV claimed that there is right now a real chance to replace loop detectors with new technologies such as video, FVD and V2V/V2I communication. But he also cautioned that the loop detector has been declared dead already many times... and yet still it lives.

When it comes to the use of travel time data for online purposes in routing, navigation and traffic management there may be two different routes for the future: FVD and its current complementing technologies deliver area-wide traffic data and finally replace any other device/ means of doing so; or FVD and its current complementing technologies deliver area-wide traffic data but a limited amount of additional techniques will still be needed, especially for highly reliable data such as that needed for online signal control.

Assuming the second route, an interesting question for the future is how private service providers and public local authorities will interact. Local authorities may buy in private information and replace their own infrastructure in order to save costs. If they do not, there may be gaps or even conflicts between private and public information. However, especially if penetration rates get higher and thus private information becomes more valid, for local authorities it will be necessary at least to exchange the advice strategies for traffic control and routing in order to coordinate the data sources and ensure most effective traffic management.