A monocular omnidirectional vision-based method for structured light strip extraction and robot target localization under complex interference

A method for structured light extraction and depth reconstruction, tailored to monocular omnidirectional vision systems, is proposed in this study to improve localization accuracy under wide field-of-view conditions with multiple objects and complex interferences such as reflections and occlusions. This method incorporates a multi-threshold fusion adjustment mechanism and introduces new algorithms for structured light clustering and discontinuity repair, aiming to improve the accuracy of centerline extraction. By integrating a neural network algorithm, the position of an object in the robot coordinate system can be accurately estimated from a single monocular omnidirectional image snapshot. The experimental results demonstrate that, compared with conventional extraction methods, the proposed method reduces the depth reconstruction error by 69.18 % in interference environments. By integrating the algorithm into the robotic system, multi-object recognition and localization were successfully achieved using a monocular camera. This provides a reference for the application of monocular omnidirectional vision in robotic systems.