无需微调，用双阶段方法精准定位监控视频中的罕见交通事故。
Two-Pass Zero-Shot Temporal-Spatial Grounding of Rare Traffic Events in Surveillance Video
- 分两步粗到精定位：先全片1帧/秒粗筛，再局部5帧/秒精修时间与位置。
- 联合定位准确率0.539，比基线提升超25%，显著优于单模型方案。
- 适合处理无标注数据的安防场景，尤其适合依赖API调用的零样本应用。
在真实监控视频中定位交通意外是一项罕见事件问题，因标注事故视频常被禁止，但仍需精确地联合定位时间、空间和碰撞类型。我们提出一种无需微调的流程，通过两个核心思想从冻结的视觉语言模型中获取联合输出：首先采用粗到精的双阶段分解策略——以1帧/秒对整段视频扫描，生成粗粒度的时间（t）、空间（x,y）与碰撞类型（c）元组；随后在±3秒窗口内以5帧/秒进行第二阶段精修，同时设置两个确定性置信度门控机制，在边界模糊或坐标逼近边缘时回退至粗略估计。其次引入专业角色分工：Qwen3-VL-Plus负责定位，Gemini 3.1 Flash-Lite负责中心片段的碰撞类型识别。在ACCIDENT@CVPR 2026基准（包含2,027段真实监控视频）上，达到时空定位准确率（ACC^S）0.539（95%置信区间[0.525, 0.553]），相比基准论文最佳基线（0.412）提升0.127，比最强单视觉语言模型基线（Molmo-7B，0.396）高0.143，比朴素基线（0.289）高出0.250。该方案每视频最多调用三次API（17%视频因失败转为物理推理），整体运行成本约20美元。
原文摘要 · Abstract (English)
Grounding traffic accidents in real CCTV footage is a rare-event problem where training on labeled accident video is often prohibited, yet accurate joint localization in time, space, and collision type is required. We present a no-fine-tuning pipeline that elicits this joint output from frozen vision-language models through two ideas. First, a coarse-to-fine two-pass decomposition: a full-video pass at 1 fps produces a coarse (t, x, y, c) tuple, then a second pass at 5 fps within a +/- 3 s window refines time and location, with two deterministic confidence gates that revert to the coarse estimate on boundary hedges or edge-clamped coordinates. Second, a specialist role assignment: Qwen3-VL-Plus handles grounding, Gemini 3.1 Flash-Lite handles typing on a centered video clip. On the ACCIDENT@CVPR 2026 benchmark (2,027 real CCTV videos) we reach ACC^S = 0.539 (95% CI [0.525, 0.553]): +0.127 over the benchmark paper's best-of-baselines oracle (0.412), +0.143 over the strongest single-VLM baseline (Molmo-7B, 0.396), and +0.250 over the naive baseline (0.289). The VLM path uses up to three API calls per video (17% fall back to physics on API failures); the full run costs ~$20.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。