Star sensor is a device that relies on measuring the pointing of the navigation star in the spacecraft coordinate system to determine the attitude of the spacecraft, and is the highest precision attitude sensor, whose accuracy can reach the angular second level, with the advantages of no drift, long working life, etc. It is a fundamental and critical device for the survival and performance improvement of spacecraft. It is of strategic importance in space applications such as remote sensing to the earth, deep space exploration and high precision detection. Most of the current star sensors use traditional optical sensors. Conventional star sensors use frame-based imaging modes, such as global exposure and roll-up exposure, which generate a frame of star image within a fixed exposure time (usually 100 ms). However, under the high dynamic conditions of large spacecraft maneuvers, this imaging mode produces motion blur and "star point trailing" effect, which, together with the noise effect, significantly reduces the signal-to-noise ratio of the star map, resulting in lower accuracy or even failure of star point extraction, and severely limits the dynamic performance of the star sensor. Star point center of mass extraction is a crucial part of the star sensor, which is directly related to the final attitude measurement accuracy of the star sensor. The sensitive navigation stars are long-range targets, and their orientations in the celestial coordinate system are high-precision astronomical observations with milliarcsecond accuracy and almost constant, so the final attitude accuracy of the star sensor depends on the extraction accuracy of the star point masses in the digital images. Therefore, the accuracy of the final solution of the star sensor depends mainly on the extraction accuracy of the star point center in the digital image. In the traditional star sensor, the average gray value of the whole image is used as the background noise gray value, and then the background noise is subtracted from the star point region gray value to obtain the corrected star point, and finally the exact center of the star point is calculated by the gray scale center method. In this paper, we propose an event-based star point extraction method for high-dynamics star point sensors to address the problem that the accuracy of star point center extraction decreases or even fails under high dynamics (≥3°/s) due to the imaging mode of traditional star sensors. The method takes advantage of the low latency and high temporal resolution of the event camera to avoid the motion blur under high dynamic conditions. The event camera has no fixed exposure time and no frame concept; it outputs an event stream, and each event in the event stream characterizes the change of pixel brightness, and the event contains four main elements: pixel row coordinates x, pixel column coordinates y, the moment t when the event occurs, and the polarity p that characterizes the change of brightness. Compared with the traditional frame-based imaging mode, the event camera has two main advantages: first, high temporal resolution, event cameras have extremely fast response times, in the microsecond or even nanosecond range, allowing for higher dynamic star tracking; second, lower power consumption, which generates a smaller number of events relative to the number of pixel positions in a black star background, consumes less energy. Event-based star point extraction for high dynamic star-sensitive instruments uses the imaging mode of the event camera, but there are two main challenges in processing event stream information: first, there is a certain amount of noise in the event stream due to the internal circuit structure of the event camera coupled with external environmental factors; second, existing star map processing methods can not be used directly on asynchronous event streams. Therefore, the method proposed in this paper is mainly divided into two steps: 1) based on the spatiotemporal correlation and spatial density characteristics between events, a method based on spatiotemporal density is proposed to denoise the star point event stream, remove the noisy events and retain the star point events as much as possible; 2) based on the asynchronous output characteristics of the events, we propose a mean drift-based star point localization method to calculate the center of the star point event cluster as the extracted star point center of mass. Through simulation experiments under different conditions, it is verified that our method can not only effectively remove most of the noise in the event stream but also extract the star point center of mass under the dynamics of 3°/s~10°/s with an average error of 0.04 pixels. when reaching 15°/s, the average error is less than 0.1 pixel. Under the dynamics of 15°/s~20°/s, the average error is still in the sub-pixel level. The average error is still in the sub-pixel level, and under this dynamic condition, the accuracy of the conventional star-sensitive mass extraction is greatly reduced, and even no star point is extracted, therefore, our method has obvious advantages under high dynamics.