A Modular Zero-Shot Pipeline for Accident Detection, Localization, and Classification in Traffic Surveillance Video
For researchers in traffic surveillance and video understanding, this work provides a modular zero-shot approach that avoids domain-specific fine-tuning, though it is an incremental application of existing techniques.
The paper presents a zero-shot pipeline for detecting, localizing, and classifying traffic accidents in surveillance video without labeled training data. The method achieves competitive performance on the ACCIDENT challenge benchmark, with temporal localization accuracy of 0.85 and classification accuracy of 0.72.
We describe a zero-shot pipeline developed for the ACCIDENT @ CVPR 2026 challenge. The challenge requires predicting when, where, and what type of traffic accident occurs in surveillance video, without labeled real-world training data. Our method separates the problem into three independent modules. The first module localizes the collision in time by running peak detection on z-score normalized frame-difference signals. The second module finds the impact location by computing the weighted centroid of cumulative dense optical flow magnitude maps using the Farneback algorithm. The third module classifies collision type by measuring cosine similarity between CLIP image embeddings of frames near the detected peak and text embeddings built from multi-prompt natural language descriptions of each collision category. No domain-specific fine-tuning is involved; the pipeline processes each video using only pre-trained model weights. Our implementation is publicly available as a Kaggle notebook.