Big Data for Automated Driving Technology, Transportation Planning and Engineering 2018 10th SF Bay Area ITE/ITS CA Joint Transportation Workshop
numbers below refer to SLIDES, published separately in this BLOG
Good morning, I am Gregory Brodski, CEO and Co-Founder of Ticon. Thank you all for coming to this event. First and foremost I would like to thank Dr. Wu, all organizers and the audience for the opportunity to talk on my favorite subject: Ticon platform for traffic data collection and automated analysis.
We are not preaching to the choir: in this audience, there is no need to explain the relevance of timely, accurate and available traffic data. As to our team, 12 years of the experience in traffic optimization and traffic management developed in us a passion for relevant and ample information.
One can imagine the traffic data space in three axes: data resolution, road network coverage and time coverage. Each approach to traffic data collection has its own specifics and, therefore, boundaries, which also limit these data application for particular ITS tasks.
The detection-based approach was the only option for many years. There are a lot of state-of- the art detectors on the market, they are really precise, and have the highest resolution. But due to costly hardware, the road coverage is very limited.
Portable detectors are better in spatial coverage. But their time coverage is extremely low. Usually they are employed only for a few days, both for statistical or for improvement evaluation purposes. This approach is very popular due to budget reason, and its outcome – for example, Average Annual Daily Traffic values, are widely used. Sometimes we can see in the literature that the comparison with AADT is used as a proof of accuracy. Probably, we often underestimate the inaccuracy of portable detection method.
It is worth to know that, if you utilize AADTs originated from 2-day counts you can easily expect up to 50% mistake. We made this conclusion by studying thousands of data points, generated by DOTs permanent stations, which are publically available in many states. Low time coverage is usual cause of inaccurate and bias results in before-after analysis as well. The same applies to the other methods, inaccuracy of which is caused by using low resolution approach.
Frequently, this is a case for GPS methods, which use navigation data. The resolution of such data is often limited to long segments or TAZ polygons. Currently, low resolution, as well as dynamically change penetration rates, complicates the use of trips, or origin-destination data for accurate traffic flow and traffic demand estimations.
We performed a number of case studies to compare the outcomes of different technologies as applied to ITS tasks. The conclusion is simple: quality and quantity of traffic data needs to be drastically expanded.
Our experience forced us to approach the data collection task structurally, and to expand in all directions: road network coverage, time coverage and high resolution, in order to achieve desired quality of the information. We see Ticon’s task in delivering available, ample, accurate, affordable data, arranged for decision making.
Our effort to deliver ample information is based on Big Data. One can depict Ticon platform as a multi-layer stack, or, if you prefer environmentally-friendly term, as a tree. The roots of Ticon tree are in the collection of the information of different types from various sources. The trunk is formed by Artificial Intelligence modules, which process the data and pump it into a comprehensive model of transportation network. That model consolidates information and studies each road segment with analytical tools, including but not limited the depicted five. The resulting information is presented via Ticon reporting system. The reports address the special needs of Traffic engineers, ITS developers, Smart cities authorities, Transportation planners, First time responders, and other Concerned professionals.
Ticon platform uses 3 kinds of data: (1) direct - directly related to vehicles, traffic flows and road network, (2) indirect - somehow connected with traffic situation but not necessarily directly relevant to traffic flow parameters, and (3) additional - delivered from traditional sources, including randomly placed detectors of various types. All these data may come from different sources; we are always open to incorporate more data contributors, or substitute new ones, if they are better suited for the task or the area of interest. Of course, any such change or incorporation shall be preceded by careful consideration of the source. We understand that any information from any source may be outdated, unhealthy or non-existent in some areas. Therefore, all data is ranked by relevance, checked, verified and filtered. What makes Big Data fundamentally different is the number of data sources, which is thousands of times greater than in traditional methods. Indeed, each connected vehicle, each geographical, or, if you will, network-graphical input is an independent source. On the one hand, it greatly reduces the "weight" of each error, but on the other hand, it increases the importance of mathematical instruments, data filtering systems, ways of data ranking, as well as the role of appropriate interpretation of information supplied by each contributor and entered to Ticon model.
For the processing of multi-source data, Ticon platform uses several Artificial Intelligence Modules, which describe road network, means of traffic control and vehicles behavior. Data processing by Artificial Intelligence Modules allows accurate estimation of all vital parameters of the traffic flow. These includes true volume estimation, with the use of several complimenting and cross-verified methods, and among them is the utilization of deterministic speed-density relationship in dynamic form, which assure accurate flow volume computation in the whole range of speed domain – from free flow to oversaturated traffic regime.
These parameters form the basis of Ticon virtual transportation model. Graphic results are displayed as a logically multi-dimensional map, where every segment reflects volume or density or delay by its width, Federal Road Class or Level of Service or speed by its color. On this particular picture we dropped the color to emphasize traffic volume values, estimated for all roads, with no exclusion, which is unique feature of our product.
This map shows traffic volume value by width, and speed – by color. Mutual supplementing and cross-verification of multi-sourced data by AI modules assures 100% coverage with a resolution: in time – up to 15 minutes, and in space – up to exact street address. Ticon virtual transportation model is great for visualization of traffic patterns, but more importantly, it can generate numerical ranking of road sections by different factors, like traffic delay, saturation degree, total driver time loss, and depict the impact of each section on areal mobility. This feature may save a lot of resources for city planners and can be extremely helpful in choosing the most fruitful project for further activity. In Ticon virtual transportation model, a full set of parameters is available for each segment of the road, for each time interval, and for the entire period under consideration.
These settings are reflected in the right pane on the slide, and you can see there: period of averaging, day or days under consideration, AADT, as well as intraday speed and volume graphs. Good performance of Ticon algorithm is fueled by the use of high resolution data, both from GPS and GIS sources, both timewise and spatial wise. Its accuracy, from our prospective, is enough for most of the ITS tasks. Naturally, the cost is directly related to the complexity of the data, so customer should compromise between price, resolution and accuracy. Overall we came to a conclusion that Ticon virtual transportation model contains everything required to compute all, or almost all metrics needed in practice and recommended by Highway Capacity Manual. Of course, there are some parameters that we fundamentally cannot determine with the current data model, for example, the number of stops, or the length of queues. And, there are those for which we still lack precision, for example turning movement estimation. We are working on raising the accuracy to fit those tasks.
Analytical tools extract specific information from virtual transportation model to deliver metrics, which address the specific ITS problem. For before-after analysis, for example, such metric can be the traffic delay, for congestion analysis – travel time delay, saturation degree, etc. Based on Customers feedback we constantly increase the number of analytical tools, and, accordingly, the variety of customized reports.
Ticon product – customer oriented reports, which are generated based on simple formal requests, and provide the user with ready-made information for the solution of particular ITS- task. Reports are prepared and pre-arranged for decision making, and require no special training, additional data processing or modeling. Therefore, even sophisticated analysis turns out to be easily accessible and independent, which sometimes may be important not only for transportation professionals, but also for municipal authorities or the local communities. Let’s see how this concept works, using before-after analysis as an example.
For our example we used the Atlanta Smart Corridor, which was announced by city authorities last year. The corridor is a signalized artery with over 20 intersections, and Smart City package includes, among other things, Surtrac adaptive signal timing control system. More details about the project are available at the City of Atlanta website. For this analysis, we did not request any special information from the city authorities or any contractors connected with this project, and used only our own or publically available sources.
We call our before-after analysis tool “TrafficScope” since it aims to provide very detailed information for the eyes of interested observer. Here, with the help of TrafficScope, we check if Customer’s expectations for the mobility improvement had been met after implementation of adaptive control system technology, which took control over the intersections on October 11, 2017, which was widely announced in local newspapers. Surtrac sets the expectation of 25% travel time reduction and 40% traffic delay reduction, which is even published on this company’s web-site. Because we believe in weekly and seasonal traffic regularities, we run the Ticon report to compare January 2017 – well before this intelligent traffic signal control implementation and January 2018 – well after it.
First of all, TrafficScope generates Matrix of Benefits, where each cell shows travel time reduction and traffic delay benefit for each hour of the week. On this matrix for Traffic Delay, we can immediately see that the implemented improvement measures are overall effective: the majority of cells are blue, which means there is some gain. However, the gain is significantly less than anticipated: 25%, not 40%. For travel time the reduction is 6%, not 25%, which is 4 times less than expected. From this Matrix of Benefits we can clearly identify problematic time slots, where new system caused degradation rather than improvement. These slots are red, with the brightest - and the worst - on Friday afternoon.
For these particular time slots, from 4 pm to 6 pm, Trafficscope will generate the graph for Traffic delay along the corridor, which supports spatial identification of the problem. From this graph we can notice the intersections where improvement measures caused the increase of traffic delay. Obviously, the controls of such intersections need a re-adjustment for that ‘problematic Friday afternoon’ period of time. Before demanding re-adjustment, we need to make sure that the approaches of the intersections are not all oversaturated during the critical time slot. If oversaturation is a case, it will be hard to achieve further improvement by signal timing optimization. Probably, we should check this on the example of Parkway Drive, the intersection, underlined on the graph, which mostly suffered from the implementation of ‘improving’ measures.
For the purpose of this examination, TrafficScope provides saturation heat map, where each cell represents saturation degree for each approach, for every 15-minute bin. We can see that traffic demand in conflict directions changes rapidly, so it is desirable to have individual timing tables for most of the time slots. It shouldn’t be a problem for the intelligent signal control system, like the one used in Atlanta, though. We see motley colors almost in every time slot, and overall saturation degree for the problematic period does not exceed 80%. So, we can conclude that there is still some room for further signal timing improvement.
The information for the re-adjustment contains in TrafficScope Speed-Volume graphs, which are generated for every day, and, additionally, for every problematic time slot. This one is for the same ‘problematic Friday afternoon’. This graph provides the values of traffic volumes and speed for every section of the corridor of study, averaged for the period of interest. By the way, it is visible from this graph that the traffic speed at several sections on the right had decreased after the adaptive control implementation. At the same time, there are no significant changes in volume at these sections. So, this picture confirms the necessity and possibility of signal timing improvement for previously noted intersections, including underlined Parkway Dr.
Speed/Volume graphs are very helpful for the detailed analysis of the corridor. In addition to the ones, averaged for the whole corridor, as you saw on previous slide, TrafficScope can provide speed and volume values for every intersectional approach, that also can be used for signal timing optimization. These graphs are built for each road section, for each day, with up to 15- minute bin resolution. Graphs of any type can – and, in our opinion, for correct understanding of corridor operation, should - be generated not only for the artery of study but also for the crossing streets. Trafficscope can do this task without major impact on the report price. Updated reports can be generated by TrafficScope even cheaper than the initial one. It should be noted that Ticon technology, as hardware free and 100% automated, is very decently priced. Evaluation reports may be repeated with updated traffic data as many times as technically needed, even weekly, which can be done with very limited budget.
Therefore, TrafficScope allows customers to form a closed mobility improvement cycle. Here certain improvement horizons, like 40% traffic delay decrease, are stipulated at the project stage, and the real achievements are checked after the initial implementation. Further analysis support tune-ups of implemented system, and outcomes are checked again in order to repeat the cycle until the challenge is resolved with the desired quality, or until you have exhausted the reasonable network capacity. Such a workflow supports the performance-based pricing model that the contractors are not very fond of, but which is adored by customers and especially by the public.
I’d like to mention yet another feature of Ticon technology which opens radically new opportunities for mobility improvement planning. It is based on the analysis of measures that had previously been implemented on the road network. For most areas, we have the ability to build reports for any period of time during the past 10 years. In other words, it is possible to conduct before-after analysis of previous activities, and see what the effect of each one was. This "analysis of mistakes" will be undoubtedly better suited to select the technologies for the further changes, as well as build certain ranking records.
We believe that the ability to obtain more detailed traffic data quickly, hardware free, and more economically, will serve our common cause, and it is my great pleasure to introduce to you today the result of three years of our work in this direction. Suggested conclusions summarize our current view on the benefits offered by Ticon technology: - Ticon opens new possibilities for mobility improvements due to its ample data model, comprehensive processing algorithm and Customer-oriented reporting system. Good performance of Ticon algorithm is fueled by the use of multi-sourced, high resolution data processed by comprehensive proprietary algorithm (patents pending). - Ticon platform is “affordably unlimited”, which allows fast scaled implementation of Smart city philosophy, as well as effective tune-ups of implemented ITS - Ticon areal study of road capacity along with forensic retrospective analysis allows for choosing most efficient projects and transportation planning aimed for fast mobility improvement - Ticon allows for performance based pricing model for ITS implementation
Once again, I want to thank the organizers and participants of this wonderful event.

Presented by: Dr. Gregory Brodski