This paper introduces responsibility distribution estimation for ego-view traffic accident videos, a new task in which a model predicts the percentage of responsibility assigned to each involved agent, and constructs an LLM-assisted responsibility annotation pipeline and fine-tune multimodal large language models under multiple input settings.
Abstract
Recent studies on multimodal traffic accident understanding have mainly relied on infrastructure-camera footage, satellite imagery, or structured crash records. However, such data sources are costly to deploy and maintain at large scale, and they cannot objectively capture what the driver was actually able to observe before the accident. In contrast, ego-view accident videos directly represent the driver's visual perspective, making them suitable for reasoning about avoidability and driver responsibility. In this paper, we introduce responsibility distribution estimation for ego-view traffic accident videos, a new task in which a model predicts the percentage of responsibility assigned to each involved agent. We construct an LLM-assisted responsibility annotation pipeline and fine-tune multimodal large language models under multiple input settings, including raw frames, segmentation-enhanced input, and textual descriptions. Experimental results establish a strong initial benchmark, demonstrating that multimodal LLMs can effectively perform this nuanced, constraint-based reasoning task. Our findings suggest that ego-centric accident videos provide a promising foundation for socially and legally meaningful multimodal reasoning beyond conventional accident classification and explanation tasks.
CAViAR exposes a practical Perception--Reasoning Gap: current VLMs may recognize salient context, but do not reliably map visible agent actions to annotated rule-relevant responsibility categories in safety-critical driving scenarios, and all models degrade sharply on accident type and responsibility reasoning.
Sparsh Garg, Yi-Wen Chen, Vijay Kumar et al.· 0 citations
As road traffic accidents are high-frequency incidents. Although there are sufficient and clear traffic control cameras at some perception-dense sites, in other places there is often a lack of enough cameras. In order to make full use of limited information to efficiently obtain an overview of the accident scene in a short time, assist on-site handling and investigation, and help reduce secondary casualties and restore traffic. This paper makes full use of existing technologies and builds a framework that can fully utilize the image and text information of the accident scene. First, the images uploaded by vehicles are used for VLM to generate objective descriptions, and the descriptions from multiple sensors within a single vehicle are fused. Then the network architecture is designed to carry out credibility calculation and severity classification. Finally, the summary is selected to form the final accident consensus report. Through the above work, semantic description and severity assessment of the event can be quickly achieved, while false reports are minimized to the greatest extent.
Traffic video understanding has become an important problem in intelligent transportation, as road videos provide direct evidence for accidents, violations, and interactions between vehicles and vulnerable road users. A useful system should explain how a traffic event develops, why it happens, and when the relevant interaction occurs, yet this remains difficult for multimodal large language models (MLLMs) because traffic videos contain sparse events and varied viewpoints. We introduce UniTraffic-Agent, the MR-CAS solution for Track~3 of the 10th AI City Challenge, which includes Traffic Anomaly Reasoning (TAR) and two out-of-domain evaluations: FETV for fisheye traffic events and PSI-VQA for pedestrian intention reasoning. UniTraffic-Agent follows an observe--reason--act--verify workflow that samples timestamped visual evidence, reasons over all questions from the same clip in one request, and converts responses through task-specific action adapters. On the official Public leaderboards, MR-CAS ranks 16th on TAR with a score of 0.5780, 2nd on FETV with 0.4884, and 4th on PSI-VQA with 64.4161. The code is available at https://github.com/Roclp/UniTraffic-Agent.
Peng Li, Qianqian Xu, Shilong Bao et al.· 1 citation
EgoSafe-Bench is introduced, a benchmark specifically designed to probe forensic reasoning in egocentric safety scenarios, generated by pairing each of the 3,000 video clips with a QA chain governed by the proposed Hierarchical Reasoning Evaluation (HRE) protocol.
Understanding the real-time behaviour of pedestrians and their surrounding context is essential for road safety. Unfortunately, current perception systems struggle to capture how multiple cues jointly influence behavior in diverse traffic scenarios. Most existing models in Intelligent Transportation Systems (ITS) and Autonomous driving system (ADS) predict only a single attribute, such as intent or trajectory, while overlooking the interaction between behavioral and environmental factors. We propose a unified multi-attribute framework based on a Vision Transformer, which is enhanced with lightweight, parameter-efficient adapters. This innovative modular design enables independent learning of eight complementary attributes, including pedestrian behavior (action, attention, crossing, occlusion) and scene context (weather, time of day, road presence, density). Our proposed framework updates fewer than 2% of the model parameters. We ran our proposed framework using the JAAD dataset, and the framework achieves over 90% accuracy on most attributes and exhibits consistent behavior–context relationships and demonstrates robustness under varied environmental conditions. To evaluate real-time suitability, we implemented an adaptive strategy that activates only the most relevant adapters based on scene conditions. This approach reduces computational cost by 28.5% and increases throughput by 40.8% without loss of accuracy. This research distinctly highlights the practical potential of modular, PEFT-enhanced transformers for scalable multi-attribute perception as foundational vision models continue to advance. The best-performing adapters and inference code are available on Github Repo.
Traffic surveillance cameras capture accidents continuously, yet converting raw CCTV footage into structured event records that pinpoint when, where, and what type of collision occurred remains unsolved at scale. The ACCIDENT @ CVPR benchmark evaluates exactly this joint prediction under a strict constraint: no labeled real-world training data is available. We introduce a training-free, two-pass coarse-to-fine pipeline that pairs a frozen Qwen3-VL-32B-Instruct vision-language model with YOLO11x object detection and BoT-SORT tracking. A first pass sparsely samples the full clip to anchor the collision moment in time; a second pass re-examines a tight window around that estimate using frames annotated with stable vehicle identities and normalized bounding-box coordinates, which gives the model both a visual overlay and an explicit numeric description of the same scene. On the official 2,027-clip real-CCTV test set, our system achieves a three-way harmonic mean score of 0.504, surpassing all organizer-published baselines including the best multi-model ensemble (0.412) by a 22% relative margin.
Dipit Saha, Shahruz Mannan, Mohammad Raihan Rashid et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
