The driver's 'divine assist” on the road! BIT develops hybrid brain-computer interface driving assistance system to improve driving safety

​With the improvement of people’s living standards, cars have entered thousands of households. However, while vehicles provide travel convenience, traffic accidents have also become an important threat to the life safety of drivers and pedestrians.

According to incomplete statistics from the World Health Organization in 2018, road traffic accidents are one of the important factors causing casualties and economic losses. Traffic accidents cause nearly 1.35 million deaths and 20-50 million injuries every year. Nearly 3% of the GDP is consumed by traffic accidents every year.

Among them, fatigue driving is an important factor causing traffic accidents, second only to speeding. Therefore, driving safety is no small matter, even if you are an "experienced driver".

Based on driving safety issues, recently, Assistant Professor Luo Longxi and doctoral student Ju Jiawei from the intelligent human-machine system team of Professor Bi Luzheng from the School of Mechanical and Vehicle Engineering of Beijing Institute of Technology proposed a method Intelligent Driving Assistance Systems (IDAS), or synchronized sequential hybrid brain-computer interfaces (hBCIs), combine electroencephalography (EEG) and electromyography (EMG) signals to classify the driver's braking and normal driving intentions.

To put it simply and crudely, this intelligent assistance system can indirectly affect vehicle control by identifying emergencies that the driver may encounter, or it can directly control the vehicle after discovering the emergency. Effectively improve driving safety.

The research was published in the form of a paper in the English science and technology journal Cyborg and Bionic Systems.

Hybrid brain-computer interface——hBCI​

Currently, the input information of IDAS mainly includes vehicles Information related to environment, behavior, and biological signals. Vehicle and surrounding environment information mainly comes from vehicle parameters and traffic information. Some IDAS need to detect the driver's drowsiness state, while other systems rely on driving behavior detection and driving intention prediction.

So where does the driver’s relevant information come from? The answer is obtained by monitoring the activity of the driver's feet, limbs and nerves.

Sources of biological information include electroencephalography (EEG) signals and electromyography (EMG) signals. Due to the early emergence of EEG signals, brain-computer interfaces (BCIs) based on EEG signals have been used in driving behavior research. Although these EEG interfaces have made great progress in braking intention detection, their detection performance is not stable due to the characteristics of the EEG signal itself. As an effective solution, hybrid brain-computer interface (hBCI) can solve the shortcomings of EEG-based BCIs such as low stability, poor performance, and insufficient reliability.

Based on how signals are combined, hbci can be divided into two modes, using feature-level fusion strategy (hBCI-FL) and classifier-level fusion strategy (hbci-cl). The first mode combines two or more EEG signals, and the other mode combines EEG with other signals such as EMG signals and ECG signals.

The researchers invited 13 subjects aged between 24 and 30 to participate in the experiment. By collecting EEG signals, EMG signals and vehicle information during simulated driving, the detection of driver's hard braking intention in virtual driving scenarios was studied. Then, they used the hBCI model that combines EEG signals, EMG signals, and vehicle information to detect the upcoming emergency braking intention.

Three driving intention classifications

In the experiment, the R&D team compared and analyzed several simultaneous The sexual and temporal hBCI models use spectral features and temporal features respectively, as well as one VS rest or decision tree classification strategies to perform multiple classifications of the three driving intentions.

The "one VS rest" classification strategy decomposes the three categories into three parallel binary classifications, including normal driving vs. other, soft braking vs. other, and hard braking vs. other. For the one VS rest classification strategy, the final result is obtained based on the maximum value of all two classifiers.

Experimental results show that the R&D team’s hBCI system recognizes hard braking intentions 130 m/s faster than the model based on pedal deflection. The hBCI-SE1 classification algorithm and one-on-one classification strategy based on spectral features have the highest classification accuracy, and the average system accuracy is 96.37%. Finally, the team selected optimal order hBCI, optimal order hBCI and models based on single brain electrical signals or electromyographic signals for comparison.

The results show that optimal simultaneity and sequential hbci are significantly better than those based on single EEG or EMG signal method. In the test, the results obtained were in good agreement with the offline test results.

This research has certain reference value for human-centered intelligent assisted driving systems to improve driving safety and driving comfort. However, the project currently has certain limitations. For example, there are various stimulus factors that induce hard braking and soft braking, the impact of subject differences, the inconvenience of the collection device, etc. Next, the team will solve the above limitations and explore more effective feature and strategy fusion to improve performance.

This research was partially funded by the National Natural Science Foundation of China (51975052) and the Beijing Natural Science Foundation of China (3222021).

Paper address:

https://downloads.spj.sciencemag.org/cbsystems/aip/9847652.pdf​

The above is the detailed content of The driver's 'divine assist” on the road! BIT develops hybrid brain-computer interface driving assistance system to improve driving safety. For more information, please follow other related articles on the PHP Chinese website!