In order to enlarge surveillance range, constant speed dome cameras are applied to surveillance camera systems, but this application also makes image recognition of alarm more difficult. Template matching is used to calculate camera's movement distance, renew reference image and segment motion objects by accumulative difference image in the paper. For workload reduction of template matching and interference elimination of camera oscillating, recursive least square method is also applied to estimate the movement of camera in the paper. The test results indicate that motion objects can be segmented from image sequence of constant speed dome camera effectively for alarm pattern recognition.