基于压力传感阵列的方向盘握持状态识别

Steering Wheel Grip State Recognition Based on Pressure Sensing Array

自动驾驶是汽车智能化的重要组成部分,完全的自动驾驶是汽车智能化发展的制高点,但当前自动驾驶技术处于人机共驾的阶段,车辆的控制权需要在驾驶人与系统间来回切换,驾驶人仍是“人-车-路-环”的核心。对驾驶人手部握持状态的实时准确识别是实现车辆控制权安全切换的基础。为此,采用柔性仿生微纳米阵列压力传感器与力反馈方向盘相融合的方法,设计开发出用于驾驶人离手检测(Hands Off Detection,HOD)的智能方向盘系统,利用此系统采集了20名驾驶人在模拟驾驶状态下的握持方向盘压力数据集,并基于此数据集分别建立了方向盘握持位置的逻辑判断识别模型和握持指数识别的L-BP(Logic-Back Propagation)神经网络模型,进一步利用消融试验对L-BP模型的逻辑判断单元与BP(Back Propagation)单元对整体模型表现的贡献度进行量化,将此模型与纯逻辑判断模型、BP神经网络和支持向量机模型进行了对比测试。结果表明:L-BP模型的握持指数识别率为98.90%,比纯逻辑判断模型、BP模型和支持向量机模型的识别率分别提升了34.99%、4.00%与13.60%;握持位置逻辑判断的识别率为99.60%。所设计开发的HOD智能方向盘系统能准确实时地采集握力数据;所提出的识别模型能够准确实现驾驶人握持方向盘的手部位置与握持指数识别,为驾驶人危险驾驶检测、驾驶能力评估研究以及方向盘智能交互设计提供参考。

Automatic driving is an important part of automobile intelligence.Complete automatic driving is the commanding point of the development of automobile intelligence.However,the current automatic driving technology is in the stage of human-machine codriving.The control of the vehicle needs to be switched back and forth between the driver and system.The driver is still the core of the human-vehicle-road-environment.The real-time and accurate recognition of the driver's hand holding state is the basis for realization of the safe switching of vehicle control rights.To this end,a Hands-Off Detection(HOD)intelligent steering wheel system for real-time detection of the driver's grip strength was designed and developed by combining a flexible bionic micro-nanoarray pressure sensor with a force feedback steering wheel.This system was used to collect the steering wheel pressure data of 20 drivers in the simulated driving state.Based on these data,a logic judgment recognition model of the steering wheel holding position and Logic Backpropagation(L-BP)neural network model of the number of holding fingers were developed.Further,an ablation experiment was carried out to quantify the contribution of the logical judgment unit and BP unit of the L-BP model to the overall model performance.This model was compared to the pure logical judgment model,BP neural network,and support vector machine model.The recognition rate of the holding finger number for the L-BP model is 98.90%,which is 34.99%,4.00%,and 13.60%higher than those of the pure logic judgment model,BP model,and support vector machine model,respectively.The recognition rate of holding position logic judgment is 99.60%.The designed and developed HOD intelligent steering wheel system can accurately collect holding pressure data in real time.The proposed recognition model can accurately identify the hand position and holding finger number of the driver holding the steering wheel,and provide a reference for the driver's dangerous driving detection,driving ability evaluation research,and steering wheel intelligent interaction design.