Blackspot identification via floating car data

Student thesis: Master thesis (including HD thesis)

  • Robin Bjørnlund Jensen
4. term, Transport Engineering, Master (Master Programme)
The following report is the result of a master thesis of a 10th semester student on Aalborg University.
The main objective with this report has been to determine, whether it is possible to find risky locations using floating car data and thereby locate black spots.
The project is justified and relevant for several reasons. Among these is the fact that the more traditional method of detecting black spots is by using accident reports as the basis to finding risky locations. Although this method might make sense, because the risky locations are where accidents occur, there are some still problems with this method. One problem is that it is based on historical data, meaning that accidents in fact has to appear for the method to find them. Another and at least as important issue is that these accident reports are being more and more neglected. There is a great knowledge of traffic accidents fatalities and serious accidents. However the knowledge of minor accidents is becoming smaller and smaller due to a lower and lower degree of reporting. Statistics shows that we now only know about 10 percent of the occurring accidents. This is a great problem in using accident reports to find risky locations because it raises the question, whether we are finding the most serious conflicts and whether we are finding the locations that are indeed in need of improvements.
GPS and thereby the potential of tracking cars and using floating car data is becoming more common in all devices which means that there is a great potential in using these data to explore driving behavior. Among these acceleration and change in acceleration, jerk, can be measured and used to find risky locations. This can be done due to the assumptions that the greatest jerks are happening where there are problems.
A study of some of the existing literature has shown that the most common reaction at conflicts is braking, which results in a momentarily increase of jerk.
The following study uses data from a previous study, called ITS Platform to develop and calibrate a method with parameters to use jerks to locate risky locations.
This is done by selecting data that is happening on the roads, that is happening at minimum speed so that there is a potential risk in the situations. The data that is selected is also not occurring at bumps or other unstable surfaces that might result in great jerks. Left is therefore a set of data occurring places where there might be risks as a basis of the detections.
This dataset is afterwards further adjusted by selecting and using data in two different geographic scopes and with three different parameter variations. All these variations are six different tries of using the data to find black spots. The parameter variations that has been used are variations in the size of the jerks and the number of jerks that has to occur on each location to locate risky locations.
After the parameter variation each location has been investigated by looking at tools such as satellite images and Google Street View to find a reason for the presence of the risks at each location.
Finally after this investigation it is concluded that the parameter variations used in this study did not exclusively find risky locations and it did also probably not find all locations where there might be risks.
There are several reasons for this, one is of cause that using floating car data is not a good way of detecting these black spots. Another other and probably a better reason is the fact that only a small dataset has been used in this study. The data that has been used is produced by tracking around 390 cars for three months. That means that a great part of the road network has not been driven on in the data. In addition it is questionable whether the cars over this period have been in enough serious conflicts for making a sufficient dataset. One of the criteria for finding a risky location was that there had to be a certain number of detections in each location. If the dataset wasn’t sufficient large enough for enough occurrences at each location or the criteria to the size of jerks was too small due to data then the real black spots might not have been detected.
This is probably what has happened in this study. That the used dataset was not extensive enough because even though the locations that were found in this study makes sense they are to a large extend not black spots. The study has however shown that the most important criteria in finding risky locations is the size of the jerk, which has to be rather large. Also the study did not show any indications that finding black spots using floating car data is impossible or unpractical. Because of that this method is still on the verge of getting its breakthrough.
Publication date10 Jun 2015
Number of pages62
ID: 213893234