研究目的
To establish a driver intention prediction model for cooperative driving by identifying the driver's braking intention using functional near-infrared spectroscopy (fNIRS) to measure cerebral cortex activities.
研究成果
The study successfully established a driver's braking intention identification model with 80.39% accuracy using PCA and BPNN on fNIRS data. The model can predict braking intention before the actual operation, contributing to cooperative driving research and opening new avenues for brain-controlled applications. Future work should address complex driving conditions and individual differences to improve real-time accuracy.
研究不足
The experimental environment was ideal and did not consider surrounding traffic. Other driver actions were not accounted for in the model. The verification was based on data from only 52 drivers, and individual differences in brain activities may affect real-time identification accuracy.
1：Experimental Design and Method Selection:
The experiment was designed to study driver's braking intention in a virtual reality environment using a driving simulator and an fNIRS device (NIRx). The methodology involved measuring changes in hemoglobin concentration in the cerebral cortex and applying principal component analysis (PCA) for dimensionality reduction and back propagation neural network (BPNN) for classification.
2：Sample Selection and Data Sources:
52 participants (10 female, 42 male, aged 19-38 years, average
3：5 years) with valid Chinese driving licenses were selected. Data included driving data from the simulator and brain activity data from the fNIRS device. List of Experimental Equipment and Materials:
Equipment included a driving simulator (DS), NIRx fNIRS device (provided by NIRx Medical Technologies, LLC), and a virtual reality lab setup. Materials involved the experimental road design with speed limit signs.
4：Experimental Procedures and Operational Workflow:
Participants sat in a vehicle mock-up, drove in the simulator, and passed deceleration sections six times. The fNIRS device recorded data at
5：8152 Hz with wavelengths of 760 and 850 nm. Data preprocessing involved grouping into braking and non-braking groups based on time windows. Data Analysis Methods:
PCA was used to reduce dimensionality of the 41-channel data, and BPNN was trained and tested with 80% training and 20% testing data to establish the identification model.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
