© 2013 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other users, including reprinting/ republishing this material for advertising or promotional purposes, creating new collective works for resale or redistribution to servers or lists, or reuse of any copyrighted components of this work in other works. A Multi-Reflection-Point Target Model for Classification of Pedestrians by Automotive Radar Eugen Schubert 1 , Martin Kunert 1 , Andreas Frischen 2 , Wolfgang Menzel 3 1 Robert Bosch GmbH, Advanced Engineering Sensor Systems, P.O. Box 16 61, 71226 Leonberg, Germany 2 Robert Bosch GmbH, Corporate Sector Research and Advance Engineering, Gerlingen, Germany 3 Ulm University, Institute of Microwave Techniques, Ulm, Germany Email: eugen.schubert@de.bosch.com Abstract—Pedestrian Collision Mitigation Systems (PCMS) are in the market for a few years. Due to continuously evolving Euro NCAP regulations their presence will rapidly increase. Visual sensors, already capable of pedestrian classification, provide functional benefits because system responses can be better adapted to expected pedestrian’s behavior. Nevertheless their performance will suffer under adverse environmental conditions like darkness, fog, rain or backlight. Even in such situations the performance of radar sensors is not significantly deteriorated. Enabling classification capability for radar-based systems will increase road safety further and will lower PCMS’s overall costs. In this paper a multi-reflection-point pedestrian target model based on motion analysis is presented. Together with an appropriate sensor model, pedestrian radar signal responses can be provided for a wide range of relevant accident scenarios, without risk for the health of test persons. Besides determination of human classification features, the model provides identification of the limits in classical radar signal processing. Beyond these borderlines it offers the opportunity to evaluate parametric spectral analysis methods. Keywords— automotive radar; radar signal processing; spectral analysis I. INTRODUCTION According to the Euro NCAP Roadmap Pedestrian Collision Mitigation Systems (PCMS) will be evaluated as from 2016. Although modern PCMS will be able to handle these tests, improving the performance will still go on. Beyond the time frame of 2016 future PCMS will not just brake. Algorithms based on adaptive parameterized pedestrian tracking will be used to determine the most suitable emergency maneuver out of a diversity of mitigation strategies. Thereby, highest collision mitigation is achieved while simultaneously unjustified system interventions are prevented [1]. To adapt the tracking algorithms to several object classes, e.g. pedestrians, reliable classification algorithms are necessary. Because of its advantages compared to other sensing technologies, radar will be one of the essential sensor technologies in future PCMS. However one deficiency of the state of the art automotive radar sensors is still the lack of reliability in pedestrian classification. Addressing this functionality within future radar sensor generations will increase road safety especially under adverse weather conditions, like fog, rain or visual reflections of headlights on wet roads. The value of already existing safety functions will be increased by reliable classification of critical objects as pedestrians; because of a higher level of driver’s acceptance of automatic interventions in situations pedestrians are involved. Additionally unjustified system responses will be suppressed. For instance the activation of an active bonnet only in case of an expected classified pedestrian impact will minimize maintenance effort and associated costs. First approaches in automotive radar-based pedestrian classification mostly rely on the evaluation of object’s range and velocity spreads [3]. In this paper, a multipoint radar target model capable of simulating high resolution radar responses of pedestrian accident scenarios is presented. Detailed results of a well suited approach of motion analysis are discussed, followed by the description of development and integration of the target model in an automotive radar simulation environment. Finally simulation results of pedestrian accident scenarios and deduced features for pedestrian classification are presented. Additionally the impact of applying autoregressive spectral analysis to resolution and classification features is evaluated. II. HUMAN MOTION ANALYSIS Based on observations of pedestrians with a 24 GHz Pulse- Doppler radar, a six point model representing feet, knees and two points of the upper body is developed in [4]. Thereby the movements of the legs are approximated by sinusoidal oscillations in the direction of motion superimposed to the motion of the center of gravity (CoG). Swinging arms are neglected. A more detailed approach for pedestrian modeling to estimate human walking parameters is presented in [5]. Parameters, such as cycle frequency, cycle length or cycle phase are estimated and visualized with a human walking scene in virtual reality. In [6], high resolution 12.5 GHz inverse synthetic aperture radar (ISAR) is used for Doppler measurements of pedestrians. Human motion simulation and measurements are compared to define the constraints of image processing radar. Thereby 2D radar images were simulated from motion data partially obtained from [7] with respect of seven distinctive points (sternum, hands, knees, and toes). However, simulations of pedestrian accident scenarios are very rarely addressed in literature. Most existing target models