An Intelligent Driver Training System Based on Real Cars

In driver training, the correct observation of the trainees' operation is the key to ensure the training quality. The operation of the vehicle can be expressed by the vehicle state changes. This paper proposes a driver training model based on a multiple-embedded-sensor net. Six vehicle state parameters are identified as the critical features of the reverse parking machine learning model and represented quantitatively. A multiple-embedded-sensor net-based system mounted on a real vehicle is developed to collect the actual data of the six critical features. The data collected at the same time are bound together and encapsulated into a vector and sequenced by time with a label given by the multiple-embedded-sensor net. All vectors are evaluated by subjective assessment conclusions from experienced driving instructors and the positive ones are used as the training data of the model. The trained model can remind the driver of the next correct operation during training, and can also analyze the improvements after the training. The model has achieved good results in practical application. The experiments prove the validity and reliability of the proposed driver training model.