A Real-time Monocular Tracking Method for Low-Cost Mobile Robot

摘要

Moving object locating and tracking are always the challenges in computer vision, especially in autonomous mobile robot vision. Due to the relative movement between the object and the robot as well as the image overexposure, the images are easily got blur, or worse, the object is loss and unable to be tracked. High-performance hardware and complex intelligent algorithms can make up these abuses in some degree, however, these solutions are of high cost and unable to be realized in consumer robots. Addressing on this problem, an improved Apriltag-based tracking algorithm running on a low cost embedded mobile robot with Raspberry Pi3 is proposed. Dynamic downsampling algorithm is first proposed to improve the real-time performance of the traditional Apriltag-based tracking algorithm. The dynamic downsampling coefficient is then studied through experiments and fitting technology. Then the relative velocity is also taken into account to enhance the downsampling accuracy. Image enhancement based on histogram equalization is used to reduce the image blur caused by downsampling and the relative motion. Finally, a secondary detection algothm is designed based on Apriltag’s recognition principle to improve the locating and tracking rate.