Design and experiment of a VSLAM system by calculating feature points with a wide-angle camera

Precise localization is important for the intelligent navigation of mobile robot systems. To address the issue of distortion in images from a wide-angle camera and enhance localization accuracy, an enhanced visual simultaneous localization and mapping (VSLAM) framework utilizing a correction model for wide-angle cameras is proposed. Firstly, images are divided into blocks, and the details of certain selected blocks are enhanced. Secondly, feature points from enhanced blocks are extracted, which improves the extraction of feature points across the entire set of images. Finally, points corresponding to incident light on a normalized plane are designed as template points. The template points and the images are matched using the wide-angle camera model, leading to the calculation of the pixel coordinates of the template points within the images, along with their descriptors. Furthermore, lots of experiments are conducted with public datasets. The results indicate a reduction in trajectory and translation errors when employing the proposed VSLAM framework, leading to the improvement of the average matching accuracy by about 3.5% compared to exist systems.