Wang Hongtao, Wen Hui, Yi Feng, Zhu Hongsong, Sun Limin
Beijing Key Laboratory of IOT Information Security, Institute of Information Engineering, CAS, Beijing 100093, China.
School of Cyber Security, University of Chinese Academy of Sciences, Beijing 100049, China.
Sensors (Basel). 2017 Mar 9;17(3):550. doi: 10.3390/s17030550.
Road traffic anomaly denotes a road segment that is anomalous in terms of traffic flow of vehicles. Detecting road traffic anomalies from GPS (Global Position System) snippets data is becoming critical in urban computing since they often suggest underlying events. However, the noisy ands parse nature of GPS snippets data have ushered multiple problems, which have prompted the detection of road traffic anomalies to be very challenging. To address these issues, we propose a two-stage solution which consists of two components: a Collaborative Path Inference (CPI) model and a Road Anomaly Test (RAT) model. CPI model performs path inference incorporating both static and dynamic features into a Conditional Random Field (CRF). Dynamic context features are learned collaboratively from large GPS snippets via a tensor decomposition technique. Then RAT calculates the anomalous degree for each road segment from the inferred fine-grained trajectories in given time intervals. We evaluated our method using a large scale real world dataset, which includes one-month GPS location data from more than eight thousand taxi cabs in Beijing. The evaluation results show the advantages of our method beyond other baseline techniques.
道路交通异常是指在车辆交通流量方面表现异常的路段。从全球定位系统（GPS）片段数据中检测道路交通异常在城市计算中变得至关重要，因为它们往往预示着潜在事件。然而，GPS片段数据的噪声大且稀疏的特性带来了多个问题，这使得道路交通异常检测极具挑战性。为解决这些问题，我们提出了一种两阶段解决方案，该方案由两个组件组成：协作路径推理（CPI）模型和道路异常测试（RAT）模型。CPI模型进行路径推理，将静态和动态特征纳入条件随机场（CRF）。通过张量分解技术从大量GPS片段中协同学习动态上下文特征。然后，RAT根据给定时间间隔内推断出的细粒度轨迹计算每个路段的异常程度。我们使用一个大规模的真实世界数据集对我们的方法进行了评估，该数据集包括来自北京八千多辆出租车的一个月GPS位置数据。评估结果显示了我们的方法相对于其他基线技术的优势。
