Risk assessment for Collision Avoidance Systems

Houenou, Adam; Bonnifait, Philippe; Cherfaoui, Veronique · 2014 · OpenAlex-citations

DOI: 10.1109/itsc.2014.6957721

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of risk assessment for Collision Avoidance Systems (CAS), specifically focusing on estimating the probability of collision between an ego vehicle and surrounding obstacles. The motivation stems from the need for CAS to predict scene evolution and assess threat levels accurately to trigger appropriate warnings or autonomous actions. Existing methods often rely on deterministic Time-to-Collision (TTC) metrics or assume constant uncertainty over time, which fails to account for the increasing uncertainty of trajectory predictions. The authors propose a method that propagates the known error covariance of the current vehicle pose along predicted trajectories, allowing for a more accurate, time-varying estimation of collision risk. The methodology combines trajectory prediction with uncertainty propagation and Monte Carlo simulation. Trajectory prediction utilizes a hybrid approach: a deterministic kinematic model (Constant Yaw Rate and Acceleration) for short-term accuracy and a Maneuver Recognition Module (MRM) for long-term prediction, particularly for lane changes. Since the combined trajectory lacks a closed-form equation, the authors propagate the initial error covariance matrix along the sampled trajectory using local linear approximations and sequential estimation. For risk assessment, the system treats each future time sample as a potential collision time. It employs a Monte Carlo simulation to approximate the probability of collision by generating random poses for both the ego vehicle and target objects based on their predicted poses and associated covariance matrices at each time step. The approach was evaluated using simulated data in two scenarios involving an ego vehicle and two target vehicles on a two-lane road. In the first scenario, the ego vehicle attempted to overtake a slower vehicle, which simultaneously initiated a lane change, resulting in a collision. The results demonstrated that the proposed method correctly identified the peak probability of collision at the actual time of impact (7.4 seconds into the simulation). The probability curve showed a gradual increase as the collision time approached, reaching approximately 50% two seconds before impact and 100% less than one second before. This behavior allows the CAS to react proportionally to the threat level. The second scenario, involving a non-collision overtaking maneuver, confirmed that the system correctly assessed low risk when no collision occurred. The use of 100 Monte Carlo samples per pose provided realistic results suitable for real-time execution. The significance of this work lies in its ability to provide a pseudo-continuous evolution of collision risk over the prediction horizon, rather than a single deterministic TTC value. By accounting for the propagation of uncertainty, the method offers a more robust assessment of threat levels, enabling CAS to distinguish between varying degrees of criticality. This allows for more nuanced decision-making, such as issuing warnings earlier or taking autonomous actions only when the probability of collision is sufficiently high. The consistency of the results in simulated scenarios validates the effectiveness of combining hybrid trajectory prediction with covariance propagation for improved collision avoidance.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction