Fast Redundant Feature Point Elimination for Enhanced Real-Time Panoramic Image Matching in Driver Assistance Systems

The vehicle-mounted panoramic imaging system has gradually been implemented in vehicles with driver assistance functions. Through panoramic images, drivers can obtain richer road scene information, significantly enhancing driving safety and comfort. The future realization of autonomous driving will also depend on the development of this technology. A key component of this technology is image feature matching. Classical image feature matching algorithms generate a large number of redundant feature points during the feature extraction phase, leading not only to increased matching times but also to reduced matching accuracy. To address this issue, this paper proposes a new method for fast redundant feature point elimination (FRFEM). This algorithm assesses the redundancy of feature points using defined distance and description redundancy metrics, subsequently eliminating redundant feature points based on this assessment. Additionally, a new filtering strategy tailored to the characteristics of vehicle-mounted panoramic imaging is designed, which avoids unnecessary computations by excluding nonredundant feature points rather than searching for potential redundant ones. This strategy is implemented using a subspace clustering filtering method, employing low-cost distance estimates instead of precise high-dimensional calculations, thereby significantly reducing computational costs and improving real-time performance. To validate the performance of the FRFEM algorithm, multiple comparative experiments were designed for evaluation. The results indicate that the FRFEM algorithm enhances the matching accuracy, repeatability, and root mean square error (RMSE) of scale-invariant feature transform (SIFT). When applied to the SURF algorithm, it achieves up to a 16% improvement in matching accuracy and reduces matching time by as much as 44% compared to the original SURF algorithm. In summary, the FRFEM algorithm can significantly enhance the generation speed and accuracy of vehicle-mounted panoramic imaging.