基于GSOM神经网络模型的交通行为模式学习方法研究

Study on Traffic Behavior Pattern Learning Method Based on GSOM Neural Network Model

提出了一种用于基于视频的交通事件自动检测的交通行为模式学习方法。首先为了获取利用神经网络进行车辆行为模式学习所需的训练数据,一种基于运动估算的车辆跟踪算法被建立,将采集到的灰度视频图像序列转化为车辆标号场时空序列。其次,使用轨迹建模和编码的方法,将跟踪结果转化为轨迹数据用于网络训练。在此基础上,建立自组织神经网络,并针对自组织网络的不足使用改进的GSOM模型,选择欧氏范数作为测度,自主开发了试验软件,以U形转事件为对象开展试验,对轨迹数据进行学习。对比试验结果表明改进的GSOM算法能有效提取行为模式。GSOM相比SOM用于行为模式学习更为有效和准确。

A traffic behavior pattern learning method was presented for realizing automatic traffic event detection based on video. Firstly, in order to get training data by using neural one sort of vehicle tracking algorithm based on motion estimation was network for vehicle behavior constructed to translate gTay pattern learning, image sequences into spatial-temporal vehicle mark sequences. Secondly, after being modeled and coded, tracking turned into trajectories data. On this basis, SOM was constructed and then modified and developed software, u-turn events were tested to analyze and learn the trajectory as GSOM. With results were Euclid norm data. Experiment results show that improved GSOM algorithm is effective for extracting behavior pattern. Compared with SOM, GSOM is more effective and accurate for behavior pattern learning.