As the measurement equipment of specific force and angular velocity, the inertial sensors reveal the essential properties of the rigid motion. Because of its advantages of all-weather, passive and autonomy, inertial sensors have been widely used in the fields of military, aerospace, scientific research since its appearance, and have influenced the world heavily. The computer vision which is a newly-developed discipline in the artificial intelligence area has arosed a worldwide research trend from the 1970s in the 20th century. As the sensing device in the computer vision, camera plays an important role in the applications of pattern recognition, smart surveillance, robot guidance and so on.The integrated strategy of inertial sensors and vision sensor is parallel to the human sensor system, for the inertial sensors are similar to the vestibular system of the human body and the vision sensor resemble human's eyes. This thesis focuses on the rigid motion estimation with inertial and vision sensors, whose main content covers software and hardware design of the sensor module, the calibration of the sensor module, the rotation estimation methods and system implementation based on the sensor module. The works and results we have done are as follows: (1) We have designed a prototype of the embeded smart sensor module which is consisted of a Blackfin DSP chip, a 6-DOF Inertial Measurement Unit(IMU), a 3-axes magnetic compass and a CCD digital camera, finished the hardware design and debug work, written the firmware related to the sensor control, data acquisition and communications. Now, this sensor module is successfully used for the research work of our lab. (2) The calibration of the deterministic item and the stochastic modeling of the nondeterministic item have been done. Methods which are suitable for the sensor calibration and modeling in low cost applications have been summarized. The parameters of bias, scale factors and the nonorthogonal angles of the sensitivity axes are calibrated by the norm constraint of the physical vectors which are applied to the sensor. The stochastic modeling of the bias drift employed the method of time series analysis. In order to resolve the rotation between the accelerometer frame and the camera frame, we have proposed a method which is based on the gravity and the plumb line. (3) A low cost real time attitude determination system which is based on quaternion and AEKF has been realized. The attitude algorithm has been ...
修改评论