Online recognition methods of abnormal driving based on Encoder Decoder attention network

DOI:	10.11809/bqzbgcxb2023.08.010
Keywords:	abnormal driving; deep learning; Encoder Decoder; LSTM; Attention mechanism
Abstract:	Abnormal driving is a major threat to the safe operation of vehicles, which seriously endangers the safe and efficient delivery of personnel and materials. Based on low cost non contact mobile phone multi sensor data, this paper proposes an online recognition method of abnormal driving by combining Encoder Decoder deep network and Attention mechanism. This method consists of three modules: Encoder Decoder based on Long Short Term Memory (LSTM), Attention mechanism and a classifier based on Support Vector Machine (SVM). The system recognition method includes six steps: input coding, attention learning, feature decoding, sequence reconstruction, residual calculation and driving behavior classification. This technical method uses the mobile phone sensor data collected under natural driving conditions to carry out the experiment. The experimental results show that: 1) The mobile phone multi sensor data fusion method is effective for driving behavior recognition; 2) Abnormal driving will inevitably cause abnormal fluctuations in data; 3) The Attention mechanism helps to improve the learning effect of the model. The recognition accuracy rate of the proposed model F1 score is 0.717. Compared with that in the similar classical models, the accuracy rate has been significantly improved; 4) For abnormal driving behaviors, SVM has better recognition effect than Logistic and random forest algorithm.
Published:	2023-08-28
PDF HTML