In this thesis, methods for radar-based environment perception from the vehicle safety point of view are presented. The proposed methods comprise advanced topics of radar-based target detection and tracking in dynamic pre-crash scenarios, as well as ghost object identification. The problem of a wandering dominant scatter point on the target surface and corresponding challenge for accurate target tracking in low-range configurations is considered. The proposed method presents a procedure to estimate target wheel positions and corresponding bulk velocities to serve as fixed scatter points on the target surface. Input to this method are raw radar data. The technique spatially resolves the micro-Doppler signals, generated by the rotating wheels of the target vehicle, to determine characteristic scatter points on the target surface. A micro-Doppler parameter is defined to quantify detections that are with high probability generated by the rotating target wheels. This group of detections is processed to estimate the wheel position and corresponding bulk velocities of the target, referred to as wheel hypotheses. The proposed method is evaluated in dynamic driving scenarios, where the driver performs an emergency evading action to avoid a collision. Subsequently, the detected wheel hypotheses serve as input to a developed tracking framework, which is used to estimate the target object static and dynamic states. Since the number of detected wheel hypotheses varies, a random-finite-set-based measurement model is used to incorporate multiple wheel hypotheses detected for one extended target object. The tracking performance is evaluated in critical evading scenarios using real vehicles as the target object. In addition, the thesis emphasized the problem of ghost object generation due to multipath propagation in pre-crash scenarios. Radar sensors, perceiving the immediate vehicle environment, show an elevated ghost object presence due to a higher probability illuminating potential reflection surfaces, e.g., road boundaries or buildings. At times, these ghost objects appear to be on a collision trajectory with the ego vehicle, whereas the vehicles are in uncritical driving scenarios, e.g., an urban intersection. In real-world driving scenarios, one target object may generate multiple false-positive targets. Based on the propagation and reflection behavior of electromagnetic waves, a geometric multipath model is derived, illustrating the occurring multipath reflections on real-world surfaces, e.g., buildings or road-bounding barriers. The proposed geometric propagation model describes the relative positions of the false-positive reflections and is validated with extensive real radar data. A custom reflector target mounted on a platform, creating deterministic point targets as dominant backscatter centers of a vehicle body, validated the different multipath reflections and the overall accuracy of the model. Moreover, radar measurements of a vehicle during an intersection scenario proved relevance to multipath reflection behavior and confirmed the model assumptions. Third, the relevance of skid scenarios with high magnitudes of side slip angles in pre-crash phases is highlighted. A novel test methodology, to non-destructively transfer vehicles with mounted surround sensors in skid situations, is developed and a survey analyzing a state-of-the-art radar sensor revealed the potential to improve object tracking performance. A test vehicle, equipped with a state-of-the-art automotive radar sensor and a reference sensor, was tested in real skid situations using a kick plate and a standardized radar target. The test method utilizes the side slip angle as a criticality criterion, which may be adjusted by the kick plate. Subsequently, a novel, modified motion model is derived, estimating side slip angles in these skid driving situations. The contribution emphasizes the estimation of horizontal vehicle motion using the proposed model considering an additional lateral force applied to the vehicle rear axle. Based on these results, an Extended-Kalman filter is designed to estimate the target object relative position and velocity in skid scenarios. The evaluation includes both the tracking and side slip angle estimations in real car tests using the above-mentioned test method.