Objective To explore the main movement characteristics of unplanned extubation (UEX) in ICU patients, and to provide reference for abnormal activity recognition and early warning from artificial intelligence technologies. Methods The camera was mounted at the caudal side of the patient and UEX movements captured through the video cameras were analyzed. Results In total, 28 patients had UEX. The UEX process was divided into pre-intention stage, intention stage and extubation completion stage, showing significant difference in duration among the three stages (P<0.05). In the pre-intention stage, the upper limb appeared as the most involved body part (82.1%), followed by the head and neck (28.6%). The most frequent movement was upper-limb-groping (53.6%), followed by hand-lifting (25.0%), and touching the tube (25.0%), etc. The most involved fingers for extubation were the thumb and index finger (100%), followed by the middle finger (71.4%). Conclusion The duration of pre-intention stage was the longest in UEX, and upper limb movements such as groping, lifting and touching the tube were the most frequent. Thumb and index finger were most frequently involved. Artificial intelligence should be introduced to identify abnormal motion of UEX and achieve early warning, thus to reduce UEX and guarantee patient safety.