| The scale of China’s dangerous chemical transport tankers is expanding,and dangerous chemical transport tankers are often over speeding,stuck at height and width limits,overturned in overtaking,overturned on accident prone roads and other safety accidents,as well as tanker safety incidents caused by driver distracted driving and fatigue driving,which seriously threaten the safety of national life and property.Therefore,it is important to identify the traffic signs in front of the tanker truck and the driver’s distracted driving behaviour and fatigue driving status to assist the driver to obtain traffic information in order to take avoidance measures in advance,as well as to monitor the driver’s distracted driving and fatigue driving to avoid traffic accidents and protect the safety of the driver and passengers.This thesis takes the safety of hazardous chemical transport as the research object,and develops the YOLOv5 algorithm-based sign detection,distracted driving behaviour detection and fatigue state judgement,the main research content and conclusions are as follows:(1)Establishing the dataset required for the experiment with respect to the research content.Some data samples were selected from the public dataset,then the dataset was expanded and the number of images of each category was balanced through actual scene shooting and various data enhancement techniques,and finally the images were labelled according to the target detection objects and the dataset was divided proportionally.The experiments showed that the dataset could improve the model performance after data enhancement,and the data-enhanced algorithm model was used as the benchmark model in the experiments,and its m AP@ was 93.6%.(2)Aiming at the danger signs encountered during the transportation of hazardous chemical tankers,attention mechanism was introduced to improve the baseline model and improve the recognition accuracy of the model.Three attention mechanisms,CBAM,ECA,and CA,were introduced into the Neck,Backone,and C3 modules of the baseline model YOLOv5 s to improve the performance of the model.Comparative experiments were conducted on a dangerous traffic sign dataset.The experimental results showed that the YOLO5s+CA+C3CBAM model,which introduced CA and CBAM attention mechanisms into the Backone and C3 modules respectively,was beneficial to improving the performance of the model m AP@0.5 and FPS increased by1.3 and 5 percentage points compared to the benchmark model.(3)For distracted driving and fatigue detection of drivers of hazardous chemical transport tankers,a lightweight network is introduced to improve and reduce the number of model parameters and computational effort.The model’s parametric count and computational effort were analyzed on the driver distracted fatigue dataset,and it was shown that the YOLOv5s+Shuffle Net V2 model was the best lightweight model,with its parametric count and The number of parameters and the amount of floating point operations are reduced by 85.6% and 105% compared to the benchmark model.The YOLOv5s+Shuffle Net V2 model was used for target scene detection experiments,which showed that it could meet the detection requirements.(4)Implementing a multi-channel video detection system for hazardous chemical transport tankers.By establishing judgment criteria for the identification of fatigue status,the YOLO5s+CA+C3CBAM model and YOLOv5s+Shuffle Net V2 model are transplanted into the multi-channel video detection system,which can simultaneously detect and remind drivers of dangerous traffic signs ahead,distracted driving behaviour and fatigue status,facilitating drivers to take measures in advance to ensure the safety of dangerous chemical tanker transport.The thesis has 66 figures,16 tables and 93 references. |
PDF Full Text Request