摘要：AbstractCollision avoidance of vehicles is an essential safety feature of modern-day vehicles. The widely used Time to Collision (TTC) approach for collision risk assessment provides a false alarm in many situations like road turning, traffic intersection, and near-miss. Therefore, this risk assessment approach cannot be applied to many realistic scenarios where knowledge of future trajectory plays an important role in collision risk assessment. After evaluating conventionally used techniques, this paper proposes a novel probabilistic approach of collision risk assessment utilizing Line of Sight (LOS) concept for front-to-front end forward-collision situation. This approach does not require high computational power during online execution and is expected to reduce false alarm rates to a significant level. For the implementation of this approach, a large number of forward-collision scenarios are generated, and various motion parameters are characterized. Further, Bayesian learning is used to update the risk at every sampling instant for each scenario followed by a risk threshold generation based on Receiver Operating Characteristic (ROC) plot. Finally, a decision is made by predicting the collision risk at certain distances and then comparing them with the threshold of risk. Simulations using relevant industry-standard software and realistic assumptions have been performed, which produces results ensuring the effectiveness of the proposed methodology.
关键词：KeywordsLine of SightCollision AvoidanceHead-on CollisionBayesian LearningROC
