Abnormal driving behavior recognition algorithm combining attention mechanism and timing features
Due to the high data dimension,difficulty in detection,low accuracy and weak stability of driving behavior recognition,an abnormal driving behavior recognition algorithm that integrates attention mechanism and timing features was proposed.The process data of drivers performing specific segments of driving behavior through sensors was collected and cleaned.The cleaned data was processed by template and then normalized into N×N×C and M×C to adapt to the input of network model.The ATFN model was constructed by integrating attention mechanism and timing features to complete the classification and recognition of seven driving behaviors,including sharp acceleration/deceleration,sharp left/right turn,sharp left/right lane change and normal driving.Compared with LSTM,ALSTM and CLSTM algorithms on the open data set.The results show that the average accuracy of LSTM,ALSTM,CLSTM and ATFN algorithm in this paper is 92.94％、94.28％、90.98％、95.47％,respectively,and the accuracy of ATFN model is the highest.Compared with the other three models,the increase is 2.53％、1.19％、4.49％respectively.Combined with loss value,recall rate and F1,ATFN model has the best overall performance.The algorithm in this paper has good effect and high stability,meets the requirement of actual detection accuracy,and can provide technical support for abnormal behavior warning and driving risk assessment.4 tabs,7 figs,29 refs.
NETL
NSTL
万方数据
