Massive traffic surveillance data extracted from vehicle detectors such as cameras provide essential information for revealing urban traffic pattern. However, most existing tools only allow users to analyze the data in specific time periods and regions with particular requirements. In this paper, we work closely with traffic domain experts and investigate a novel way of reframing visual traffic analysis tasks into the combinations of various atom categorical/numerical features and visual presentation. The categorical features contain primitive attributes such as vehicle type, O/D status and driving direction, and the numerical features contains information such as vehicle frequency and speed. The combination of above features includes four basic operations, namely and, or, xor and not to support diversified user requirements. Basic and advanced visualization methods such as trajectory view and flow distribution view are provided to demonstrate the combination results. Through interactive assembling of various atom operations, analysts could derive different query conditions to meet existed and potential upcoming analysis requirements such as locating suspicious vehicles (e.g., fake plate vehicles). Furthermore, AtoMixer, a visual analytic system is developed to support spatio-temporal investigative tasks for traffic surveillance data. We evaluate the effectiveness and scalability of our approach with real world traffic surveillance data.

从摄像机等车辆检测器提取的大量交通监控数据为揭示城市交通模式提供了重要信息。但是，大多数现有工具仅允许用户分析具有特定要求的特定时间段和区域中的数据。在本文中，我们与交通领域专家紧密合作，并研究了一种将可视交通分析任务重新构架为各种原子分类/数字特征与可视表示相结合的新颖方法。类别特征包含原始属性，例如车辆类型，O / D状态和行驶方向，数字特征包含信息，例如车辆频率和速度。的上述特征的组合包括四种基本操作，即与，或，异或和不支持多样化的用户需求。提供了基本的和高级的可视化方法，例如轨迹视图和流量分布视图，以演示组合结果。通过各种原子操作的交互式组装，分析人员可以得出不同的查询条件，以满足已存在的和潜在的即将到来的分析要求，例如定位可疑车辆（例如假车）。此外，还开发了视觉分析系统AtoMixer以支持交通监控数据的时空调查任务。我们使用现实世界的交通监控数据评估我们方法的有效性和可扩展性。