Research On Real-time Image Rotation-elimination Based On A Retina-like Sensor

In certain imaging platforms, such as those included on airborne missile systems, robot vision systems, and spacecraft orbit docking systems, rolling of the carrier platform causes rotation of the imaging system and results in image rotation. In order to facilitate observation with the human eye, reduce the difficulty of target tracking and detection, and increase the efficiency and accuracy of target detection, image rotation-elimination processing for rotated images is needed to resume the stability of image. Usually the principal methods of real-time image rotation-elimination are based on image rotation algorithms and image interpolation algorithms for the output image of traditional rectangular CCD and CMOS sensors. These algorithms are complex, require a large amount of calculation, and bad real-time capabilities. However, the pixel layout of a retina-like sensor is different from that of a traditional rectangular sensor. The pixels of the retina-like sensor are arranged in concentric circles with rotational symmetry. In addition, the readout method for generating the output image is performed from the inside to the outside circles, which is very conducive to rotation-elimination processing. Therefore, in this paper our research focuses on real-time image rotation-elimination based on a retina-like sensor.This paper mainly focuses on the first retina-like CMOS sensor in China which has been developed successfully, analyzing of the pixel layout of the retina-like sensor and the output image from the retina-like camera. A model concerning the image rotation of a retina-like sensor is proposed. The correctness of the model is verified by software simulation. STM32 is used to drive a microelectromechanical systems digital accelerometer and gyroscope, and the real-time data of the sensor is filtered and fused to calculate the rotation angle of the camera. According to the rotation angle of the camera, a real-time image rotation-elimination solution based on the variable address readout sequence of the camera is proposed. And the real-time image rotation-elimination camera is realized by changing the camera driver. Several experiments of the static and dynamic image rotation-elimination have been performed, which showed that the proposed solution is simple, accurate, and rapid, providing theoretical basis and technical access for rapid image rotation-elimination.