Pedestrian Anomaly Behavior (PAB)
收藏资源简介:
Pedestrian Anomaly Behavior (PAB) 数据集由西安交通大学、合肥工业大学和澳门大学联合创建,旨在支持基于文本的行人异常行为搜索任务。该数据集包含1,013,605个合成图像-文本对和1,978个真实世界图像-文本对,涵盖了广泛的正常和异常行为,如跑步、表演、踢足球以及相应的异常行为如躺下、被击中和跌倒。数据集的创建过程包括使用生成模型合成图像和通过多模态大语言模型生成文本描述,确保了数据集的多样性和真实性。PAB数据集主要应用于智能城市、安全监控和个性化服务等领域,旨在解决在实际场景中识别行人异常行为的关键问题。
Pedestrian Anomaly Behavior (PAB) Dataset was jointly developed by Xi'an Jiaotong University, Hefei University of Technology, and University of Macau, with the aim of supporting text-based pedestrian anomaly behavior retrieval tasks. This dataset consists of 1,013,605 synthetic image-text pairs and 1,978 real-world image-text pairs, covering a wide spectrum of normal and abnormal pedestrian behaviors, including normal behaviors such as running, performing, and playing football, as well as corresponding abnormal behaviors like lying down, being struck, and falling over. The construction of the PAB dataset involves synthesizing images using generative models and generating text descriptions via multimodal large language models, which ensures the diversity and authenticity of the dataset. The PAB dataset is primarily applied in fields such as smart cities, security surveillance, and personalized services, and it aims to address the critical challenge of pedestrian anomaly behavior recognition in real-world scenarios.

- 1Beyond Walking: A Large-Scale Image-Text Benchmark for Text-based Person Anomaly Search西安交通大学, 合肥工业大学, 澳门大学 · 2024年



