The invention discloses an audio and video integration detection method orienting electric power machine room monitoring. The concrete detection steps of the detection method are divided into four steps: feature extraction of audio and video data is performed; an audio abnormal event is detected by adopting the mel-frequency cepstral coefficient as for audios of monitoring areas; feature selection is performed with 1.8 seconds acting as a unit; and the obtained video feature vectors and audio feature vectors are compared. The beneficial effects of the detection method are that safety event detection can be automatically performed in the electric power machine room key areas, and the whole process is automatically completed without too much human interaction; and occurrence of the abnormal event can be accurately detected as for the condition of noisy environment of the electronic power machine room.