發展YOLO物件偵測技術實現之駕駛者辨識系統

Development of driver identification system using YOLO object detection

黃韋融(Wei-Jung Huang)

指導教授 ： 吳建達

國立彰化師範大學/技術及職業教育學院/車輛科技研究所/碩士(2020年)

摘要

大型客車能一次載送大量民眾，有助於舒緩交通堵塞的問題，是目前廣泛運用於大眾交通運輸服務的車種。由於載送的乘客數量較多，所以發生事故時往往造成嚴重的傷亡，而駕駛者超時工作或是疲勞駕駛經常是肇事的重要原因之一。本研究企圖發展一套利用影像處理為基礎的駕駛者辨識系統。在架構中，辨識系統是將臉部辨識技術應用在辨識駕駛者身份辨識上，藉由駕駛者身份的確認來避免駕駛工時過長的問題。在實驗中，本研究利用安裝在駕駛者前方之鏡頭取得駕駛者的臉部資訊，再將駕駛者的臉部資訊輸入到NVIDIA的嵌入式晶片Jetson TX2中，並配合已經在Jetson TX2上建立的YOLO物件偵測技術進行駕駛者身份之辨識。本研究利用四種不同的模型架構和五種不同的參數，訓練出二十種不同的模型，針對六位不同身份之人員進行辨識，並在車上環境中進行身分辨識，最後再對這二十種模型進行比較。實驗室的結果顯示本系統可以確實的辨識駕駛者的身份，並有不錯的辨識率，未來期望能安裝於實際的大型客車上驗證以達到實際應用的結果。

關鍵字

影像處理 ； 臉部辨識 ； 駕駛者辨識系統 ； YOLO物件偵測

並列摘要

Large passenger cars can carry a lot of people in batches, which can help alleviate the problem of traffic jams. Large passenger cars are widely implied in public transportation services. Due to the large number of passengers carried, accidents often cause serious casualties, with drivers working overtime or fatigued do to extensive driving often one of the important reasons for the accident. This study developed a driver identification system based on image processing. In the framework a face recognition system recognizes the driver's identity, to reduce the problem of excessive driving hours by confirming the driver's identity. In these experiments a lens was installed in front of the driver to obtain the driver's face information, the driver's facial information was input into NVIDIA's embedded chip Jetson TX2, which cooperates with the already established Jetson TX2 YOLO object detection technology. This study used four different model architectures and five different parameters to train twenty different models for six different identities for body recognition in the car environment. These twenty models were compared. The laboratory results show that the system can reliably identify the driver's identity at a good recognition rate. The proposed system is expected to be installed in an actual large passenger car for verification to achieve practical results.

並列關鍵字

Image processing ； Face recognition ； Driver identification system ； YOLO object detection

參考文獻

REFERENCES

[1] Police Agency of the Ministry of the Interior, Table of important statistical results, number of motor vehicle registrations

https://www.npa.gov.tw/NPAGip/wSite/lp?ctNode=12902&nowPage=2&pagesize=15

[2] Redmon, J. & Farhadi, A. (2018). Yolov3: An incremental improvement. arXiv preprint arXiv:1804.02767.

[3] He, K., Zhang, X., Ren, S. & Sun, J. (2015). Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition. arXiv preprint arXiv:1406.4729.