掌握漁船捕獲重要種類的魚獲體長，是海洋資源保育與永續經營的重要工作。本研究延伸手動量測魚體長度方式，結合不同影像處理方式，進而達到魚體量測自動化的階段。影像處理的部分，包含魚眼、空間與魚體投影校正等方法，取得待處理魚體影像後，本研究提出以下三種方法比較其量測結果。方法一採用最小包覆圓法，找出以一個包覆魚體輪廓的最小外接圓，外接圓直徑即為魚體長度；量測誤差約在9%以下。方法二為單一角度量測法，利用通過魚體輪廓重心水平線與其交點視為魚體長度，量測的誤差約為20%。方法三多角度量測法，增加其參考線數量，進行長度的量測，結果顯示方法三能夠有效降低方法二的量測誤差。

While electric observer systems (EOS) have been popularly installed in fishing vessels to record fish catch videos and geo-graphic information, such systems still require human interventions to manually extract fish catch data from raw videos. Therefore, we propose an automatic video processing pipeline in this research to detect fish catches in real time from videos. For optimizing the system. This research tries to find out how to get fish data including catch time, Latitude, Longitude, cate-gory and fish length automatized. Then transport this data to server by satellite transmission to reduce error by human. We expect that the intelligence system can work effective in the future. In this research, the intelligent EOS can apply on simulation environment efficiently because of its higher camera resolution, reducing impact of fish men and fishing gear. However, this system can only get about 75% of correct rate in real environment.