Markus Kerper

Learning Traffic Light Phase Schedules from Velocity Profiles in the Cloud

Traffic lights strongly impact vehicle movement and fuel consumption in cities. If drivers were aware of the traffic light phase schedule, they could predict the traffic light state at arrival time and could reduce fuel consumption. To acquire information like traffic light phase schedules, our vision is that drivers share their velocity profiles in a digital cloud, and in return benefit from smart algorithms evaluating the collected data. We present one such algorithm, Traffic Light State Estimation (TLSE), that operates on the velocity profiles to backward-estimate phase schedules of traffic light signal groups operating with fixed cycle length (representing about 80% of all traffic lights in the US). We present simulation results showing that phase schedule prediction on the base of TLSE is correct more than 90% of the time.

Driving More Efficiently - The Use of Inter-Vehicle Communication to Predict a Future Velocity Profile

Fuel-efficient driving is difficult in unknown or complex environments. To aid the driver with this task, we present a novel method of tactical route optimization by calculating a short-term fuel-reduced velocity profile. This profile is based on knowledge of location-dependent velocity profiles that are collected by the vehicles over time and shared with other vehicles. To determine a fuel-efficient velocity profile, we first split the planned route into segments. We cluster the historical velocity profiles within each segment using a Dynamic Time Warping algorithm, obtaining classes of velocity profiles and their probabilities. We construct a transition graph between velocity profile classes from adjacent segments and calculate the most probable path through the next segments ahead. This path represents the most likely future velocity profile under the assumption that the driver behaves like previous drivers on the same segment. Given this profile, we calculate the fuel-reduced velocity profile with help of a shortest-path algorithm in a vehicle-specific fuel-consumption graph. First results in an urban environment indicate possible fuel savings of about 8.3% compared to the most probable profile.

Compact Vehicular Trajectory Encoding

Many applications in vehicular communications require the collection of vehicular position traces. So far this has been done by recording and transmitting unencoded or merely linearly filtered position samples. Depending on the sample frequency and resolution, the resulting data load may be very large, consuming significant storage and transmission resources. In this paper, we propose a method based on two-dimensional cubic spline interpolation that is able to reduce the amount of the measurement data significantly. Our approach allows for a configurable accuracy threshold and performs in O(n^3). We evaluate our approach with real vehicular GPS movement traces and show that it is able to reduce the volume of the measurement set by up to 80% for an accuracy threshold of 20 centimeters.

Analyzing vehicle traces to find and exploit correlated traffic lights for efficient driving

Traffic lights strongly impact vehicle movement and fuel consumption in cities. If drivers were aware of the situation at arrival time, they could adapt their velocity and thus reduce the number of unnecessary stops and fuel consumption. To predict the influence of the traffic light ahead on the velocity of an approaching vehicle, our vision is that drivers share their vehicle traces in a digital cloud, and in return benefit from algorithms evaluating the collected data. With Traffic Light Coordination Analysis (TLCorA), we present one such algorithm analyzing vehicle traces. When a vehicle is approaching a traffic light, TLCorA finds traces of vehicles similar to that of the vehicle at the previous traffic light, and calculates from their approach to the upcoming traffic light whether there is a representative approaching trace. For this purpose, TLCorA classifies the approaching traces with help of a clustering algorithm based on dynamic time warping. We implement TLCorA in simulations of different traffic light signalization algorithms, and study the calculated approach probabilities depending on the respective traffic light correlation level in the scenarios.