本研究主要目的為結合智慧型手機與攝影機，以物件偵測進行自動化人孔蓋檢測技術可行性與操作處理程序分析，並從手機陀螺儀的震幅強度數值透過小波分析過濾後，與拍攝影像採用深度學習之物件偵測的數值相互篩選，在振幅方面發現小波分解優於FFT，而深度學習的訓練模型YOLOv7 Basic相較於其他3個模型準確。
最後比較騎車時行經人孔蓋的準確率。研究初期時嘗試將手機的攝影機與加速度陀螺儀進行結合，但因為性能問題無法同步進行檢測，實驗後期則採用行車紀錄器以及手機兩部儀器同步檢測。
本研究採用即時物件偵測深度學習，比較四個YOLO v7的模型對於人孔蓋的準確度，發現一般的YOLO v7 Basic就有很高的準確率，研究對此模型辨識兩組18與60張包含有人孔蓋與無人孔蓋照片，預測準確度分別為96.6%與90%。
本研究提出，手機與攝影機的結合，在完全依賴遙測的方式來判斷道路上人孔蓋時，無法判斷是否有騎行經過，需要額外借助振幅的輔助來斷定是否有騎車行經過人孔蓋。將手機檢測的振幅數值，經由哈爾小波分解在高頻第3層且≥閥值5的判定下，針對9個實際偵測且有騎行經過之人手孔判斷有7個超過閥值，2個在誤差窗格外。額外3個為偵測誤為預測錯誤且騎行經過之路面，有1個判斷在誤差窗格內，經影像判斷為瀝青路鋪面塊。
比較手機加速度強度振幅時的快速傅立葉變換以及小波分析的結果，發現透過FFT較難過濾掉雜訊號，而透過哈爾小波分析過濾較具觀察性。單純只依靠遙測的方式判斷道路上的人孔蓋準確率為76.67%，結合了振幅強度以及小波分解後的輔助判定，準確率提高到88.89%，提升了12.22%

The purpose of the study is to combine a smartphone with a camera in analysing the feasibility and operating procedure of automatic manhole cover detection via object detection. Certain amplitude values from the gyroscope inside the smartphone were selected through wavelet analysis and subsequently compared with the deep learning-based object detection values in the photographed images. The results suggested that wavelet analysis surpasses the fast Fourier transform (FFT) in amplitude and that the deep learning training model YOLOv7 Basic yields higher accuracy performance than the other three models.
Finally, the study compared the accuracy rates of detecting moving scooter wheels that rolled over the manhole covers. At the initial stage, the study intended to combine the smartphone’s camera with an acceleration gyroscope; however, given that the two items could not be simultaneously used for detection due to functional problems, the study later used a dashcam and a smartphone for simultaneous detection.
The study conducted instant deep learning-based object detection and analysed the model’s identification of two groups of photographs (18 and 60) including or not including the images of manhole covers, with the predicted accuracy rates being 96.6% and 90%, respectively. By examining the FFT of the smartphone during an increase in the acceleration amplitude and the wavelet analysis results, the study found that using the FFT leads to difficulty in reducing noise and that applying Haar wavelet analysis in filtering can facilitate observation. Also, the study discovered that when relying solely on remote sensing, the accuracy rate of identifying manhole covers on the road is 76.67%, and after combined with amplitude analysis and wavelet analysis to facilitate identification, the accuracy rate increases by 12.22% to 88.89%.

1. Andrew Campbell, Alan Both, Qian (Chayn) Sun(2019), Detecting and mapping traffic signs from Google Street View images using deep learning and GIS, Computers, Environment and Urban Systems, P. 8-10.
2. ASUS ZenFone 6(2019), link：https://www.asus.com/tw/mobile/phones/zenfone/zenfone-6/techspec/, December 12, 2022
3. Chien-Yao Wang, Alexey Bochkovskiy, Hong-Yuan Mark Liao(2022), Source code for the GitHub. Link: https://github.com/WongKinYiu/yolov7,
4. Chien-Yao Wang, Alexey Bochkovskiy, Hong-Yuan Mark Liao(2022), YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors, Computer Vision and Pattern Recognition,V.1, P. 1.
5. Commandre, B., En-Nejjary, D., Pibre, L., Chaumont, M., Delenne, C., & Chahinian, N. (2017, June). Manhole cover localization in aerial images with a deep learning approach. In ISPRS Hannover Workshop: HRIGI 17–CMRT 17–ISA 17–EuroCOW 17 (Vol. 42, pp. 333-338).
6. Evolution of Object Detection(2020), Link: https://medium.com/analytics-vidhya/evolution-of-object-detection-582259d2aa9b, December 5, 2022
7. Gogoro, link: https://www.gogoro.com/tw/ √, December 3, 2022
8. Haar, Alfréd (1910), "Zur Theorie der orthogonalen Funktionensysteme", Mathematische Annalen, 69 (3): 331–371, doi:10.1007/BF01456326, hdl:2027/uc1.b2619563, S2CID 120024038
9. Hasan, Z., Shampa, S. N., Shahidi, T. R., & Siddique, S. (2020, June). Pothole and speed breaker detection using smartphone cameras and convolutional neural networks. In 2020 IEEE Region 10 Symposium (TENSYMP) (pp. 279-282). IEEE.
10. Heideman, M. T.; Johnson, D. H.; Burrus, C. S. Gauss and the history of the fast Fourier transform. IEEE ASSP Magazine. 1984, 1 (4): 14–21. doi:10.1109/MASSP.1984.1162257
11. Liu, W., Cheng, D., Yin, P., Yang, M., Li, E., Xie, M., & Zhang, L. (2019). Small manhole cover detection in remote sensing imagery with deep convolutional neural networks. ISPRS International Journal of Geo-Information, 8(1), 49.
12. NetConn Electronics, MS279WG Details, link: https://www.polaroidcarcam.com.tw/ms279wg.html November 24, 2022
October 9, 2022
13. Paul Viola(2001), Rapid Object Detection using a Boosted Cascade of Simple Features, Mitsubishi Electric Research Labs, P.2-4.
14. Salih, Y. M. M., Kattan, A., & Çevik, T. (2016). Detection of motorway disorders by processing and classification of smartphone signals using artificial neural networks. International Journal of Natural Sciences Research, 4(3), 56-67.
15. Wei, Z., Yang, M., Wang, L., Ma, H., Chen, X., & Zhong, R. (2019). Customized mobile LiDAR system for manhole cover detection and identification. Sensors, 19(10), 2422.
16. Yanqiu Zhang(2019), Research and Application of AdaBoost Algorithm Based on SVM, Institute of Electrical and Electronics Engineers.
17. Yi, C. W., Chuang, Y. T., & Nian, C. S. (2015). Toward crowdsourcing-based road pavement monitoring by mobile sensing technologies. IEEE Transactions on Intelligent Transportation Systems, 16(4), 1905-1917.
18. YOLO — You Only Look Once (2017), link：https://medium.com/@c.c.lo/yolo-%E4%BB%8B%E7%B4%B9-4307e79524fe , December 2, 2022
19. Yu, Y., Guan, H., & Ji, Z. (2015). Automated detection of urban road manhole covers using mobile laser scanning data. IEEE Transactions on Intelligent Transportation Systems, 16(6), 3258-3269.
20. Zhang, H., Dong, Z., He, A., Zhang, A. A., Wang, K. C., Liu, Y., ... & Ai, C. (2022). Efficient approach to automated pavement manhole cover detection with modified faster R-CNN. Intelligent Transportation Infrastructure, 1.
21. Zhou, B., Zhao, W., Guo, W., Li, L., Zhang, D., Mao, Q., & Li, Q. (2022). Smartphone-based road manhole cover detection and classification. Automation in Construction, 140, 104344.
22. 王慧靜，蒲寶明，孫宏國2012(12):4,等人. 基於小波閥值的心電訊號去噪演算法[J]. 電腦系統應用。
23. 台南市各類管線數量統計(2022)，網址：https://data.tainan.gov.tw/dataset/pipeline/resource/5799457d-6825-40ba-a940-e726ab9c69d6 , December 2, 2022
24. 交通部高速公路局(2022)，道路交通標誌標線號誌設置規則，網址：https://www.freeway.gov.tw/Upload/DownloadFiles/03-%E7%AC%AC%E4%B8%89%E7%AF%87_%E6%A8%99%E7%B7%9A_0907.pdf, December 2, 2022
25. 各縣市道路事故熱點(2022)，網址：https://roadsafety.tw/Dashboard/Custom?type=%E7%B5%B1%E8%A8%88%E5%BF%AB%E8%A6%BD%E5%9C%96%E8%A1%A8 , December 3, 2022
26. 行車資訊補給站-高速公路標誌標線(2022)，網址：https://www.freeway.gov.tw/Publish.aspx?cnid=1688&p=2641, December 3, 2022
27. 宋宗勳(2004)，柔性鋪面狀況指標檢測之研究，碩士論文，國立中央大學土木工程學系。
28. 張禾(2021)載具對量測之平整度指標 AARI 之影響，碩士論文，國立台灣大學工學院土木工程學系。
29. 陳可中(2017) Google 街景影像交通標線偵測之研究，碩士論文，國立中興大學土木工程學系。
30. 陳林君(2014)，影像辨識技術應用於鋪面破壞調查之研究，碩士論文，國立中央大學土木工程學系。
31. 微軟ML.NET(2022)，網址：https://learn.microsoft.com/zh-tw/dotnet/machine-learning/tutorials/image-classification-api-transfer-learning, December 6, 2022
32. 葛孝萱(2009)，鋪面績效整合與孔蓋平坦度特徵之研究，碩士論文，國立臺灣大學土木工程學研究所。
33. 路上觀察：馬路上的祕密徽章──各式孔蓋(2020)，網址：https://www.thenewslens.com/article/131108, December 3, 2022
34. 道安資訊查詢網: 交通事故統計快照(2022)，網址：https://roadsafety.tw/,
35. 演算法YOLO演進 — YOLOv7 論文閱讀(2022)，網址https://medium.com/ching-i/yolo%E6%BC%94%E9%80%B2-yolov7-%E8%AB%96%E6%96%87%E9%96%B1%E8%AE%80-97b0e914bdbe,
36. 維基百科(2021)，哈爾小波轉換，網址：https://zh.m.wikipedia.org/wiki/%E5%93%88%E7%88%BE%E5%B0%8F%E6%B3%A2%E8%BD%89%E6%8F%9B, December 11, 2022
37. 維基百科(2021)，國際糙度指標，網址：https://zh.m.wikipedia.org/wiki/%E5%9C%8B%E9%9A%9B%E7%B3%99%E5%BA%A6%E6%8C%87%E6%A8%99, December 11, 2022
38. 維基百科(2021)，瀝青路面，網址：https://zh.m.wikipedia.org/wiki/%E7%80%9D%E9%9D%92%E8%B7%AF%E9%9D%A2, December 3, 2022
39. 維基百科(2022) ，AdaBoost，網址：https://zh.wikipedia.org/wiki/AdaBoost, December 11, 2022
40. 維基百科(2022) Java，網址：https://zh.wikipedia.org/wiki/Java
41. 維基百科(2022) 快速傅立葉變換，網址：https://zh.wikipedia.org/wiki/%E5%BF%AB%E9%80%9F%E5%82%85%E9%87%8C%E5%8F%B6%E5%8F%98%E6%8D%A2
42. 維基百科(2022)頻譜，網址：https://zh.m.wikipedia.org/wiki/%E9%A2%91%E8%B0%B1