19th ITS World Congress, Vienna, Austria, 22/26 October 2012 EU-00068
Traffic monitoring sensor with vehicle trajectory measurement for acceleration detection
Martin Litzenberger1*
1. AIT Austrian Institute of Technology G.m.b.H., Department Safety and Security, Business Unit New Sensor Technologies, Donau City Strasse 1, 1220 Vienna, Austria, +4350550411,
Abstract
There is no commercially available traffic monitoring device that is able to collect information on the composition of the traffic in terms of vehicle acceleration states. This report presents a traffic monitoring device with the potential for vehicle acceleration detection, which will be able to give the required real-time input for online emission calculation with traffic models. The device is based on an optical sensor comprising dynamic vision sensor technology. The sensor is able to extract vehicle trajectories with high temporal precision, thus has the potential to classify individual vehicles according to their acceleration state (accelerating/constant speed/decelerating). The vehicle acceleration detection will be implemented and tested in the framework of the CARBOTRAF FP7 project and will be used to infer online traffic emission.
Keywords:
Traffic monitoring, acceleration detection, online traffic emission calculation
Introduction
Traffic congestion with frequent “stop & go” situations causes substantial increase in emissions (CO2, black carbon as well as air pollutants) in contrast to free flowing traffic. The traffic control measures available today, when driven by information collected in real-time from the road network, are able to support the reduction of emissions. To derive the necessary control actions from the information collected requires modelling and computing the traffic state depended emission of vehicles. The output of vehicle emission models strongly depend on the acceleration state of the vehicles. Today’s traffic monitoring devices typically collect information on traffic volume, fleet composition and average speed. However, there is currently no commercially available traffic monitoring device that is able to collect information in real-time on the composition of the traffic in terms of vehicle acceleration states (accelerating/constant speed/decelerating). This report presents a traffic monitoring device with the potential for vehicle acceleration detection, which will be able to give the required real-time input for online emission calculation with traffic models. The vehicle acceleration detection will be implemented and tested in the framework of the CARBOTRAF FP7 project (ICT for Transport) funded by the European Union [1]. The project started in Sep. 2011 and will run for three years.
Traffic Monitor based on Dynamic Vision Sensor Technology
The smart eye TDS traffic data sensor is a fully embedded traffic monitor based on dynamic vision sensor (DVS) technology. The DVS is a CMOS imager chip with very high temporal resolution (better than 1 millisecond) as well as on-chip motion detection and background suppression [2]. The pixels of the imager encode motion as time stamped x,y-coordinates in an asynchronous data stream. Thus the DVS efficiently encode the trajectories of the vehicles motion. The current smart eye TDS system monitors traffic on up to four lanes simultaneously and records speed and class individually for each detected vehicle [3,4].
Vehicle Acceleration Detection
Using the smart eye TDS traffic data sensor we have recorded vehicle trajectories for cars entering a roundabout. Figure 1a shows the still image of the roundabout and the two detection zones defined to monitor its entries. Each detection zone extends about 10m (equivalent to about 40 pixels of the sensor) in the direction of vehicle movement. Figure 1b shows example track data (yellow) in the x,y-plane for detection zone 2 with the principal axis of vehicle motion rotated parallel to the x-axis. Figure 2 shows examples of about 15 seconds of data of four vehicles trajectories in space-time x,t representation. The high time resolution of the sensor allows extracting the vehicles trajectories with high precision. The trajectories for zone 1 show two vehicles with constant speed (marked “C”), and one vehicle decelerating (“D”) and stopping at the entry.
(a) (b)
Figure 1 – Sensor detection zones (a) and extracted vehicle trajectories from zone 2 (b).
Figure 2 – Example vehicle trajectories from zones (1) and (2) for four vehicles.
The example trajectory from zone 2 shows one vehicle decelerating (“D”), almost stopping and accelerating again (“A”). The examples show that acceleration/deceleration phases of vehicles can be robustly distinguished using the sensors output.
Exploiting this information the individual vehicle data sets provided by the traffic data sensor in real-time are additionally tagged with acceleration information (accelerating/constant speed/decelerating states) to allow inferring the quantity of accelerating vehicles in the vehicle collective. This information will be used in the CARBOTRAF project to support the online calculation of traffic emissions for a test area.
Preliminary Results
First preliminary measurement results have been obtained with a sensor mounted overhead on a gantry approximately 5 m away from the stop line of a traffic light. The speed limit on this road section is 50 km/h. Figure 3 shows the test site with the coordinate system used for the investigations with the mount position of the TDS sensor. The sensor is mounted on a gantry (marked by a circle in the figure).
The image coordinates have been transformed to world coordinates by projecting the imager pixel on ground (road) level. For the vehicle preceding edge this transformation is satisfied as it is near to the ground plane. The transformation has been performed for the preliminary investigation using the known camera mount height, tilt and yaw angles but without further calibration efforts. For this preliminary investigation the velocities have been extracted manually from the data by marking and measuring the angles of vehicle tracks in the space-time diagram.
In figure 4 (top) vehicle trajectories in world coordinate space x and y versus time for a period of 120 seconds are shown. The x and y coordinate units are given in 0.1 m units. The traffic activity seen by vehicle tracks during the green (A) and red (B) traffic light phases can be clearly distinguished. In figure 4 (bottom) a close-up of 8 seconds of traffic data shows trajectories of four vehicles during free flow. The velocity of three of the vehicles has been estimated by measuring the angles of the vehicles tracks in the space-time diagram.
Figure 3 – Test site for acceleration detection at a junction.
The result is between 45 and 56 km/h for this coarsely calibrated data set. Figure 5 shows extraction of velocity from two accelerating vehicle’s trajectories on the centre lane. Note that the data of the centre lane is highlighted to simplify the extraction of the respective vehicle tracks. Velocity has been measured at begin of the acceleration phase and at the time the vehicle is passing under the sensor position. Table 1 summarizes these data extracted from the trajectories. For the vehicle samples number 4 and 5 an acceleration of approx. 10 km/h has been observed within 2 and 4 seconds time periods, respectively. The vehicles started to accelerate at a distance of approx. 20 m and 30 m from the sensor position on the traffic light gantry.
Figure 4 – top: Vehicle trajectories in world coordinate space. bottom: Zoomed trajectories, 8 seconds of data.
Figure 5 – Zoomed trajectories of 8 seconds of data, with three measured velocities.
Table 1 – velocities extracted from vehicle trajectories.
Sample (ref. to figures) / Time (s) / Velocity (km/h) / Approx. distance from sensor (m) / Time (s) / Velocity (km/h) / Approx. distance from sensor (m)1 / 1660.5 / 56.5 / 0
2 / 1663.5 / 45.7 / 0
3 / 1665.0 / 50.4 / 0
4 / 1630 / 22.7 / 20 / 1632 / 32.5 / 0
5 / 1630 / 19.1 / 30 / 1634 / 30.7 / 0
Conclusion and Outlook
The smart eye TDS traffic data sensor has shown to provide the principle ability to measure vehicle acceleration from recorded vehicle trajectories. This preliminary investigation used a simple calibration for the image to world coordinate transformation and manual extraction of the velocity change from a small sample of vehicle tracks for a proof of concept. The next steps are the development of an automatic acceleration detection algorithm and an implementation of the algorithm in the embedded software of the TDS device. The calibration of the image-to-world coordinate transformation will be improved to improve the absolute measurement accuracy. The performance of the algorithm will be tested in real-life operation during the pilot installation of the CARBOTRAF project.
Acknowledgement
The research leading to these results has received funding from the European Union
Seventh Framework Programme FP7/2007-2013 under grant agreement n° 287867.
References
1. http://carbotraf.eu/
2. Lichtsteiner, P.; Posch, C.; Delbruck, T., "A 128x128 120 dB 15 µs Latency Asynchronous Temporal Contrast Vision Sensor", Solid-State Circuits, IEEE Journal of, vol.43, no.2, pp.566-576, Feb. 2008
3. Bauer D., Belbachir AN, Donath N, Gritsch G, Kohn B, Litzenberger M, Posch C, Schön P and Schraml S, “Embedded Vehicle Speed Estimation System Using an Asynchronous Temporal Contrast Vision Sensor,” EURASIP Journal on Embedded Systems, vol. 2007, Article ID 82174, 12 pages, 2007. doi:10.1155/2007/82174
- Gritsch G.; Litzenberger M.; Donath N.; Kohn B., “Real-Time Vehicle Classification using a Smart Embedded Device with a `Silicon Retina' Optical Sensor”; IEEE International Conference on Intel-ligent Transportation Systems, ITSC08; ISBN 978-1-4244-2112-1; pp. 534 – 538, October, 12 – 15, 2008
1