The Attitude determination is an important element in guidance and control of moving bodies. In this thesis, we consider an attitude estimation algorithm for the moving body mounted with a camera, gyros and accelerometers. We propose an attitude estimation method using vector pairs, which are measured in the reference coordinate system and the body coordinate system about two reference directions by a imaging sensor and accelerometers, through the extended Kalman filter. In the processing getting the reference direction from an imaging sensor, we should track the interested region in consecutive input images. For tracking, the block matching method is a good candidate due to its simplicity; however, ordinary block matching may fail to track for some cases. To solve this problem, we propose the modified block matching algorithm which applies the weight mask with two-dimensional Gaussian density and considers rotational component. To verify performance of the proposed algorithm, we show some simulation results.