基于交通轨迹数据挖掘的道路限速信息识别方法
Recognition method of road speed limit information based on data mining of traffic trajectory

分析了道路限速信息的时空变化性,提出一种基于轨迹数据挖掘技术的道路限速信息自动识别方法。为了实现海量交通轨迹数据的快速处理,研究了快速地图匹配与数据清洗等预处理算法,分析了交通轨迹数据的速度分布特性与最高车速限制指标。基于路段行车速度的统计特性,构建了道路特征向量模型,以快速提取海量轨迹数据的潜在特征信息。提出了多投票K近邻分类算法对数据特性进行训练与学习,以实现对道路限速信息的快速识别。以福州市交通路网及其浮动车轨迹数据构建试验样本集进行训练、学习与交叉验证试验。试验结果表明:在训练过程中,当样本数量达到1 200时,方法的识别准确率最高达到93%,在仅有150个小训练样本下,方法的识别准确率也达到75%; 方法具有近线性的处理性能,处理1.0×106条道路的限速信息仅用时46 ms。
The spatiotemporal variability of speed limit information was analyzed, and an automatic recognition method of road speed limit information was proposed based on the mining technique of trajectory data. To fast process the massive traffic trajectory data, the pretreatment algorithms such as rapid map matching and data cleaning were researched. The speed distribution features of traffic trajectory data and the maximum speed limit index were analyzed. Based on the speed features at road section, a road feature vector model was constructed to rapid extract the latent characteristics information from the massive trajectory data was achieved. In order to implement a rapid recognition of speed limit information, a classification algorithm based on multi-voting K-nearest neighbor(MV-KNN)algorithm was proposed for the training and learning process of data feature. The training, learning and cross-validation experiments were completed by using the sample sets constructed by actual floating car trajectory data and traffic network in Fuzhou City. Experimental result indicates that the highest system recognition accuracy of proposed method is up to 93% by using 1 200 samples in the training process, and the system recognition accuracy is 75% by using only 150 samples. The near-linear processing performance of proposed method is revealed, and the system operating time is only 46 ms in processing 1 000 000 samples of road speed limit information. 1 tab, 12 figs, 29 refs