基于GMM-CHMM 的城市道路换道行为识别

高级驾驶辅助系统(ADAS)是提高车内乘员安全性的主动安全系统之一，将车载参数和车辆位置参数相结合，提出一种能够应用到ADAS的城市道路换道行为识别模型. 在西安城市道路环境中进行实验，采集18 位驾驶员的9 个车载实时参数数据，以及前后车辆间的相对速度、相对距离、相对角度，提取412 个换道行为单元和824 个车道保持行为单元，共 88 992 条数据. 运用数理统计方法分析表明，方向盘转角、转向角速度、相对安全距离比在换道行为和车道保持行为之间有显著性差异，在这3 个特征参数的基础上，建立混合了高斯混合模型(GMM)和连续型隐马尔可夫模型(CHMM)的识别模型，用部分样本对模型效能评价. 结果表明，混合模型对换道行为的识别精度为93.6%，具有良好的识别效果，可以很好地应用到 ADAS.

Lane Changing Behavior Identification of Urban Road Based on GMM-CHMM

Advanced driver assistance system (ADAS) is one of the active safety systems to improve the safety of occupants in vehicles. In this paper, an urban road lane changing behavior identification model that can be applied to ADAS was proposed by combining vehicle parameters and vehicle position parameters. The experiment was carried out in the urban road environment of Xi'an, where 9 on-board real-time parameters data of 18 drivers as well as the relative speed, relative distance and relative angle between the front and rear vehicles were collected. 412 lane changing behavior units and 824 lane keeping behavior units were extracted, with a total of 88 992 data. The analysis of mathematical statistics shows that there is significant difference between steering wheel angle, steering angle velocity and relative safe distance ratio between lane changing behavior and lane keeping behavior. An identification model is established on the basis of these 3 features parameters and identification model is a mixture of Gaussian mixed model (GMM) and Continuous Hidden Markov Model (CHMM). Some samples are used to evaluate the effectiveness of identification model. The results show that the identification accuracy of the hybrid model is 93.6%, which has a good effect and can be applied to ADAS well.