查詢結果分析
來源資料
頁籤選單縮合
題 名 | 以深度學習進行兩階段車輛事故偵測=Two-stage Vehicle Collision Detection by Deep Learning |
---|---|
作 者 | 史美華 ; 謝禎冏; 羅妤恩 ; 周雨璇; | 書刊名 | 大同學報 |
卷 期 | 38 2024.01[民113.01] |
頁 次 | 頁55-63+79 |
分類號 | 312.831 |
關鍵詞 | 道路路況; 車禍偵測; 深度學習; 兩階段辨識; Road situation; Vehicle accident detection; Deep learning; Two-stage classification; YOLO; ResNet; |
語 文 | 中文(Chinese) |
中文摘要 | 本研究蒐集臺灣汽機車各種不同程度車禍狀況畫面,透過人工智慧深度學 習影像辨識模型YOLOv4-Tiny與ResNet50,各自資料集經過不同數據增強後, 訓練出這兩種深度神經網路權重模型。接著使用現有道路監控畫面及行車紀錄 器影像,可以有效的用YOLOv4-Tiny和ResNet50模型進行兩階段影像即時辨識。 第一階段辨識出道路物件並同時透過Deepsort紀錄物件軌跡,第二階段辨識達 到偵測道路狀況與車禍嚴重程度。最後以Tkinter程式介面即時呈現出道路狀況判斷結果與所偵測物件影像、辨識信心度與物件追蹤軌跡等資訊。 |
英文摘要 | This study collected images of automobile and motorcycle accidents of varying degrees in Taiwan and used artificial intelligence deep neural network models YOLOv4-Tiny and ResNet50 to train on these datasets after augmentation to respective data. Then, existing road surveillance and driving recorder images could be effectively processed by the two-stage real-time recognition models. In the first stage, road objects are identified by YOLOv4-Tiny, and the object trajectories are tracked using Deepsort. In the second stage, car accident severity is classified by ResNet50. Finally, the Tkinter library is used to create the user interface to instantly display the road condition judgment results, including object images, recognition confidence, object tracking trajectory, and other related information. |
本系統中英文摘要資訊取自各篇刊載內容。