使用深度學習方法檢測分心駕駛之研究

Research on Detecting Driving Distraction Using Deep Learning Methods

董乙慧(Yi-Hui Tung)

指導教授 ： 范敏玄

國立臺中科技大學/資訊與流通學院/資訊管理系碩士班/碩士(2021年)

摘要

本研究使用深度學習方法檢測分心駕駛的十種類別，包括在駕駛過程中使用手機、飲食、操作收音機、化妝、與乘客交談等等。本研究採用StateFarm及AUC兩個分心駕駛資料集，透過VGG16、Resnet50、InceptionV3三種不同的卷積神經網路，並結合注意力機制網路來判斷駕駛員是否有分心以及其行為所屬的類別。採用的資料集圖片均是在車內中拍攝收集的。本研究使用的方法運用了預訓練(pretraining)權重的方式來減少訓練過程的時間，並提高其準確度。本研究將此三種模型進行比較，結果顯示在兩個分心駕駛資料集上，VGG16結合CBAM模型準確率有最佳的性能，在StateFarm分心駕駛資料集準確率達99.2%，AUC分心駕駛資料集準確率為98.4%。

關鍵字

分心駕駛 ； 深度學習 ； 卷積神經網路 ； VGG16 ； 注意力機制

並列摘要

This research uses deep learning methods to detect ten categories of distracted driving, including using mobile phones, eating and drinking, operating the radio, putting on makeup, talking with passengers, etc. during driving. This research uses two distracted driving data sets, StateFarm and AUC, through three different convolutional neural networks, VGG16, Resnet50, and InceptionV3, combined with the attention mechanism network to determine whether the driver is distracted and the category of his behavior. The pictures used in the data set were all shot and collected in the car. The method used in this study uses pretraining weights to reduce the time of the training process and improve its accuracy. This study compares these three models, and the results show that on two distracted driving data sets, VGG16 combined with the CBAM model has the best accuracy. The accuracy of the StateFarm distracted driving data set is 99.2%, and the AUC distracted driving The accuracy rate of the data set is 98.4%.

並列關鍵字

Attention mechanism ； Convolutional Neural Network ； Deep Learning ； Driving distraction ； VGG16

參考文獻

[1] 中華民國內政部. (2020). 109年第52週內政統計通報(1-11月道路交通事故死傷人數較上年同期減少6,818人（-1. 6％）). 取自： https://www.moi.gov.tw/News_Content.aspx?n=2905 s=212063

[2] NHTSA. (2019). Distracted Driving. Retrieved from https://www.nhtsa.gov/risky-driving/distracted-driving?page=1

[3] Peden, M., Scurfield, R., Sleet, D., Hyder, A. A., Mathers, C., Mohan, D., Jarawan, E. (2004). World report on road traffic injury prevention: World Health Organization.

[4] Le, A. S., Suzuki, T., Aoki, H. (2019). Evaluating driver cognitive distraction by eye tracking: From simulator to driving. Transportation research interdisciplinary perspectives, 4, 100087.

[5] Kaber, D. B., Liang, Y., Zhang, Y., Rogers, M. L., Gangakhedkar, S. (2012). Driver performance effects of simultaneous visual and cognitive distraction and adaptation behavior. Transportation research part F: traffic psychology and behaviour, 15(5), 491-501.