Offline Optimization Of Curve Speed Warning Applications, Vassilis Kaffes, ICCS
- 1. OFFLINE OPTIMIZATION OF CURVE SPEED WARNING APPLICATIONS Vassilis Kaffes Institute of Communications and Computer Systems (ICCS) Iroon Polytechniou St. 9, 15773 Athens, Greece +30 210 772 3865, 30 210 772 2291, vkaff@iccs.gr Dr. Manolis Tsogas Institute of Communications and Computer Systems (ICCS) Iroon Polytechniou St. 9, 15773 Athens, Greece +30 210 772 3865, 30 210 772 2291, mtsog@iccs.gr Panagiotis Lytrivis Institute of Communications and Computer Systems (ICCS) Iroon Polytechniou St. 9, 15773 Athens, Greece +30 210 772 3865, 30 210 772 2291, panagiotis.lytrivis@iccs.gr Dr. Angelos Amditis Institute of Communications and Computer Systems (ICCS) Iroon Polytechniou St. 9, 15773 Athens, Greece +30 210 772 2398, 30 210 772 2291, a.amditis@iccs.gr Dr.-Ing. habil. Christoph Arndt Ford Research & Advanced Engineering Europe Süsterfeldstr. 200, D-52072 Aachen, Germany + 49 (0) 241-9421-381, + 49 (0) 241-9421-301carndt1@ford.com ABSTRACT In automotive safety applications the system must be capable of early recognizing the maneuvers performed by the driver and the intention associated with them in order to take preventive measures or trigger warning alarms. In this paper the problem of optimizing the core algorithm of a curve speed warning application is investigated. The offline tool will determine the accuracy of specified map data compared to real data using vehicle dynamics. The work described here is part of the research activities in the European research project euroFOT. 1
- 2. INTRODUCTION Road safety applications such as lane departure warning and curve speed warning systems are based on the knowledge of the road lane geometry. Nowadays systems care for more active solutions which cannot be provided by current systems based only on vision technology (7). Digital maps and on-board sensors estimating dynamic state of the ego vehicle such as radars and laser scanners (4) extend the accuracy of road geometry specification (2), (3). Thus overall safety in the road is about to increase in such systems dealing with situations where lane markings are missing or are ambiguous or the visibility is restricted due to weather conditions (1). The goal of this paper is to evaluate the map data accuracy and reliability compared to real world reference data. CURVE SPEED WARNING APPLICATIONS Curve-speed warning (CSW) technology has been developed (5) to help drivers identify potentially dangerous situations e.g. the vehicle is driving in a curve too fast, and warn the driver in advance allowing him enough time to react properly. A curve is represented as a list of shape points which are retrieved from the map database. Figure 1. Map Data in the vehicle research platform 2
- 3. The CSW application is based on Map Data, well known from commercial navigation systems. Nevertheless, the usage of Map Data (6) can be extended to active safety system, when the route of the vehicle in known in advance. Therefore a prediction of the path (the Most Likely Path) is used to extract relevant data (i.e. shape points, geometry information like radius, speed limits, etc.) from the oncoming route. These data are the Electronic Horizon (EH) and they provide the basis for the CSW application. By processing the coordinates of these shape points it is possible to extract the road geometry. Thus analysis of the geometric characteristics of the bend combined with external factors such as weather conditions, the maximum recommended velocity for the bend is estimated. If the velocity exceeds the maximum recommended velocity the driver is warned. CSW may also assist the driver during night driving and also helps reduce the blinding effect from the headlights of oncoming vehicles. Using digital maps (see figure No. 1) we create segments of shape points according to a set of rules and create the road geometry utilizing the clothoid model. Also using the vehicle dynamics we can estimate the curvature of the road cR and create warnings when the factor FC = c/ cR is greater than a selected threshold. OPTIMIZATION ALGORITHM The offline tool will carry out statistical processing on map data, such as distance to shape points from map-data and can-data for different road classes or distance along the vehicle path in meters from minimum radius and min radius from map data. The parameters in algorithm calculating road geometry so as to provide accurate curve speed warnings to driver will be adapted according to the derived results from the processing. THE SETUP OF THE OPTIMIZATION TOOL The optimization of the algorithm requires an additional tool that handles all aspects of the CSW parameter variation. The first step of this tool is the reconstruction of the Electronic Horizon (i.e. the path in front of the vehicle). All relevant data for CSW are extracted from the Map-Data. This includes the shape-points and the radius of the road at the shape-points. This data are visualized in figure 2. 3
- 4. Figure 2. The Path in front of the Vehicle Figure 2 shows the position of the vehicle at the upper right corner of the image. Each oncoming geometry point has a coordinate and a radius. The radius is shown as a red line, perpendicular to the route. From these data the algorithm computes the limit velocities for the oncoming curves. Comparison of the current velocity and limit velocity leads to a warning, if the driver approaches an oncoming curve too fast. The given warning is shown in figure 3. Figure 3. Speed warning displayed to the driver 4
- 5. Figure 3 is displayed to the driver and it indicates a right turn in 650m. The maximum velocity should be 46 kph and thus the driver should reduce the velocity to drive safe through the next curve. Going back to the previous image (figure 2), we see that this warning is coming from the sharp right turn marked with an arrow in the geometry plot. Even though the driver does not even see this curve, the warning already indicates that the current velocity exceeds the limit. VERIFICATION OF THE ALGORITHM AGAINST MAP-DATA The sensor information of CSW comes from the map-data and from vehicle data. Therefore the maps have to be evaluated regarding reliability of their geometric data. Figure 4. Curvature from vehicle data and curvature from map-data (*) Assuming constant circle conditions, we are able to compare the curvature from vehicle data and map-data. In this case of normal driving conditions, the curvature data show a good correlation in figure 4. There is some noise from the vehicle sensors and there are some outliers in the map data. For example: At time 4600 sec or 4950 sec, there is obviously a wrong sign in the map data. 5
- 6. RESULTS The evaluation of the CAN-data, vehicle data provide real world reference, and SQL-data, map data from the database, enable us to extract all relevant CSW-information. The dedicated SW tool that developed will help us to provide reliable results. The difference between the vehicle data curvature and the curvature from map data is a measure for the reliability of the map data and it provides hints for the tuning of the algorithm. Figure 5 shows the vehicle curvature data and the interpolated map curvature data. The histogram of the difference, also shown in figure 5 (the lower plot) shows a good correlation of both data. Figure 5. Upper plot: Curvature from vehicle data (blue) and from map data (red) Lower plot: Histogram of the difference of both curvatures. 6
- 7. The difference in the curvature seems to be not large. However, the effect of these differences on the application has to be evaluated with our automatic evaluation tool, to tune the parameters correctly. Furthermore, this is only an example of an overall view on the evaluation procedure. There are several attributes of the map data (Functional Road Classes, ADAS-flag, etc.) that have to be evaluated separately. Only the complete evaluation of the data and the parameters will lead to an optimized tuning and finally to an optimized application. CONCLUSION The CSW algorithms improvement contribute to safe driving and also to euroFOT project. The results will be a starting point for evaluating the performance and necessity of curve-speed warning systems in real time conditions for driver safety applications. For the future the results that derived from the evaluation tool on correlating curvature from vehicle data and from map data will be used to tune certain parameters of the application. Many of the attributes that are provided by the map data will be evaluated too and all together will be used to optimize and provide an accurate and reliable application. AKNOWLEDGMENTS This work was supported by the European Commission in the euroFOT project. The authors would like to thank all partners within euroFOT for their cooperation and their valuable contribution. REFERENCES (1) M. Tsogas, A. Polychronopoulos, A. Amditis, Using Digital Maps to Enhance Lane Keeping Support Systems, in Proceedings of the IEEE Intelligent Vehicles Symposium, June 13-15, 2007, Istanbul, Turkey, pp. 148-153. (2) Klotz, J. Sparbert, D. Hötzer, Lane data fusion for driver assistance systems, in Proceedings of the 7th International Conference on Information Fusion, June 28- July 1, 2004, Stockholm, Sweden, pp. 657-663. (3) H. Weigel, H. Cramer, G. Wanielik, A. Polychronopoulos, A. Saroldi, Accurate Road Geometry Estimation for a Safe Speed Application, in Proceedings of Intelligent Vehicles Symposium, June 13-15, 2006, Tokyo, Japan, pp. 516- 521. (4) Kirchner, T. Heinrich, Model based detection of road boundaries with a laser scanner, in 7
- 8. Proceedings of IEEE International Conference on Intelligent Vehicles, Stuttgart, Germany, 1998, pp. 93-98. (5) Manolis Tsogas, Panagiotis Lytrivis, Angelos Amditis, Enhanced Curve Speed Warning Application Using Multiple Sources Of Information For Extracting Road Geometry, 10th International Conference on Application of Advanced Technologies in Transportation, Athens, Greece, May 27-31, 2008. (6) V. Blervaque, ERTICO, K. Mezger, Daimler Chrysler Research, L. Beuk, Siemens VDO, Eindhoven, J. Loewenau, BMW Group Research and Technology, ADAS Horizon – How Digital Maps can contribute to Road Safety, Advanced Microsystems for Automotive Applications 2006, Monday, July 31, 2006 (7) Chenhao Wang, Zhencheng Hu, Tomoki Maeda, Naoko Hamada, and Keiichi Uchimura , Predictive Lane Detection for Simultaneous Road Geometry Estimation and Vehicle Localization, 2009 IEEE International Conference on Robotics and Automation, May 12th 2009, Kobe, Japan 8