Method of pose tracking by event camera using LED marker points

Using of vision sensors to achieve pose tracking has become a hot spot currently in the domain of computer vision. traditional vision sensors. However, the frame output of the conventional vision sensor and the redundant image data have a large delay and are highly dependent on the light intensity. There is a need for an attitude tracking technique that relies less on light and has a lower latency to improve system performance. Compared with traditional vision sensors, a new type of event camera is inspired by the neural network principle to develop event-based vision sensing technology, which is more robust in terms of delay and dependence on light intensity. We applied event processing algorithms to high-frequency flashing LED markers, including frequency filtering, clustering, and feature-point tracking. In the pose tracking of the event camera, the attitude is obtained from the PnP (perspective-n-point) algorithm, and a least-squares method is used to optimize the pose tracking. The experiment uses high-frequency LED as marking points, showing the comparison of event camera output in high and low light intensity environments, stability testing of frequency filtering clustering algorithm, and Z-axis data error of pose tracking within 5%. This method can be applied to attitude tracking in low light intensity environments.