本篇論文提出一種可適用於嵌入式系統運算條件的打架行為事件偵測技術。本技術主要是運用移動能量分析方法來偵測是否發生打架行為。此外，本技術中採用自動調適的移動物偵測臨界值設定方法，使得行為偵測功能在不同環境照明條件下皆能正常運作。為驗證本論文方法的效能，實驗影片中除了打架事件外還包含握手、慢走、快跑、作操等行為事件以測試其偵測可靠度。實驗測試也包含室內正常照明和戶外夜間路燈照明兩種條件。本技術已實作於網路攝影機的內嵌式系統以確認其執行即時偵測的能力。

A video processing technique of fight detection based on a motion energy analysis method is proposed for implementation on embedded system. In order to make the fight detection work properly under various illumination conditions, a moving object detection with adaptive threshold is proposed in this paper. To verify the effectiveness and robustness of the fight detection, the experimental videos include fight events and other noise events, such as hand-shaking, loitering, running, and doing body exercises. In addition, the illumination in these videos include an indoor normal lighting and an outdoor night view under the lighting of street lamp. The technique has been implemented on an IP camera to verify its capability of performing real-time detection on embedded system.