According to inspection problems of horizontal and vertical pendulum amplitude of ship models, a method of character recognition used to inspect pendulum amplitude of ship models based on sequence images in complex background is discussed. Firstly, Gaussian low-pass filtering algorithm is adopted to update image background in real time based on gray value with the characteristic of slow change, which is at the adjacent regions of background images. Secondly, an adaptive threshold segmentation algorithm based on statistical information of difference image is put forward to eliminate the influence of the change of light and environment on background. Finally, character recognition algorithm based on the improved scan line algorithm and template matching algorithm is researched. The experiment results show that the algorithm can identify the characters printed on ship model quickly and completely in the different illumination circumstances. And it can be used to inspect pendulum amplitude. So it has great practical value.