Accelerated Feature Extraction and Description Algorithm Based on Color Images

With the increasing demand for feature extraction and matching accuracy in computer vision neighborhoods, traditional algorithms based on grayscale images are gradually shifting to those based on color images to obtain more information. The improvement of hardware computing power also makes real-time extraction and calculation of color features possible. This article proposes an accelerated feature extraction and description algorithm based on color images (referred to as RGB-SURF algorithm), which extracts and describes color features from images to improve the accuracy of algorithm matching. To improve the real-time performance of the visual SLAM algorithm, GPU acceleration method is used. By improving the algorithm descriptor to enhance robustness against rotational changes. A comparative experiment was conducted using the TUM dataset and the Graffiti dataset to analyze the distribution of extracted feature points on the RGB three channels. At the same time, it was verified that the RGB-SURF algorithm improved the number of matching points while maintaining computational speed.