公开(公告)号：US06177885B1

公开(公告)日：2001-01-23

申请号：US09185321

申请日：1998-11-03

Applicant: Roark D. Weil ,  Asdrubal Garcia-Ortiz ,  John R. Wootton

Inventor： Roark D. Weil ,  Asdrubal Garcia-Ortiz ,  John R. Wootton

IPC: G08G101

CPC classification number: G08G1/0104

Abstract: A traffic incident detection system (10) includes both the collection and analysis of traffic data and employs a time-indexed traffic anomaly detection algorithm which partitions time into categories of “type of day,” and “time of day”. Using this partition, a fuzzy neuromorphic, unsupervised learning algorithm calibrates fuzzy sets as “normal” and “abnormal” for a plurality of traffic descriptors. Fuzzy composition techniques are used, on a per traffic lane basis, to combine multiple traffic descriptors in order to determine membership in a “normal” or “abnormal” lane status. Each lane status is then combined to determine the overall status of a road segment. Initial training of the algorithm occurs during the first few weeks after a sensor (12) is installed. On-line background training continues thereafter to continually tune and track seasonal changes affecting system performance.

Abstract translation: 交通事故检测系统（10）既包括交通数据的收集和分析，又采用时间索引的交通异常检测算法，将时间划分为“一天的类型”和“时间”。 使用该分区，模糊神经形态，无监督学习算法将模糊集合校准为多个业务描述符的“正常”和“异常”。 使用模糊组合技术，在每个车道基础上组合多个交通描述符，以便确定“正常”或“异常”车道状态的会员资格。 然后将每个车道状态组合以确定道路段的总体状态。 在安装传感器（12）后的最初几周内，算法的初始训练发生。 在线后台培训继续不断调整和跟踪影响系统性能的季节性变化。