Scholarlink

<p>This study presents a quantitative assessment of the impact of rainfall-induced visual degradation on camera-based object detection and Autonomous Emergency Braking (AEB) performance using a physics-based simulation framework in Simcenter Prescan<sup>®</sup>. A Physics-Based Camera (PBC) was employed, and key rain-related visual parameters, including intensity, blur characteristics, and droplet size, were controlled through the Data Model API (DMAPI). A deep learning regression model was trained to infer these parameters from real driving videos and used to reconstruct realistic rain effects within the simulation environment. Similarity analysis between real and simulated raindrop-area distributions demonstrated high fidelity under Heavy and Very Heavy rainfall conditions, with Kolmogorov–Smirnov p-values above 0.9 and Wasserstein distances based on optimal frame pairs ranging from approximately 83 to 105. Under Extreme Rain conditions, however, similarity decreased due to droplet merging and film-flow effects, indicating a fundamental limitation of image-based parameter estimation under severe rainfall conditions. AEB simulations using the inferred parameters revealed progressive delays in initial detection, reduced detection distances, lower detection continuity, and delayed TTC-based braking activation as rainfall intensity increased. In the Extreme Rain scenario, the Detection Continuity Rate dropped to 37.88%, and all braking stages were triggered simultaneously, failing to prevent a collision. The proposed real-to-simulation parameter estimation framework enables realistic and repeatable evaluation of camera-based ADAS performance under adverse weather conditions.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Real-to-Simulation Rain Parameter Estimation for Vision Sensor Performance Evaluation and AEB Assessment

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • Hyunseo Han,
  • Jiin Bae,
  • Yeonsub Lee,
  • Jungwoo Park,
  • Woojeong Jeon,
  • Jonghyuk Kim

摘要

This study presents a quantitative assessment of the impact of rainfall-induced visual degradation on camera-based object detection and Autonomous Emergency Braking (AEB) performance using a physics-based simulation framework in Simcenter Prescan®. A Physics-Based Camera (PBC) was employed, and key rain-related visual parameters, including intensity, blur characteristics, and droplet size, were controlled through the Data Model API (DMAPI). A deep learning regression model was trained to infer these parameters from real driving videos and used to reconstruct realistic rain effects within the simulation environment. Similarity analysis between real and simulated raindrop-area distributions demonstrated high fidelity under Heavy and Very Heavy rainfall conditions, with Kolmogorov–Smirnov p-values above 0.9 and Wasserstein distances based on optimal frame pairs ranging from approximately 83 to 105. Under Extreme Rain conditions, however, similarity decreased due to droplet merging and film-flow effects, indicating a fundamental limitation of image-based parameter estimation under severe rainfall conditions. AEB simulations using the inferred parameters revealed progressive delays in initial detection, reduced detection distances, lower detection continuity, and delayed TTC-based braking activation as rainfall intensity increased. In the Extreme Rain scenario, the Detection Continuity Rate dropped to 37.88%, and all braking stages were triggered simultaneously, failing to prevent a collision. The proposed real-to-simulation parameter estimation framework enables realistic and repeatable evaluation of camera-based ADAS performance under adverse weather conditions.