Hardware Accelerator of Bundle Adjustment Algorithm

Sammanfattning: With the popularization and development of CV technology, the SLAM algorithm is widely used in scenarios such as self-driving cars and autonomous navigation robots. As a key step in the SLAM system, the BA algorithm is responsible for optimizing camera parameters and 3D point coordinates. BA obtains more accurate estimates by shrinking the re-projection error. So as to support the SLAM system in building a more accurate 3D model of the surrounding environment and a more reliable trajectory of the moving camera. However, due to the high computational complexity of the BA algorithm, its computational efficiency becomes a bottleneck limiting the real-time performance of SLAM. In order to improve the performance of the BA algorithm in practical applications, the goal of our thesis work is to build and implement an efficient hardware accelerator for BA. The main tasks are as follows: • Theoretical Understanding: To fully understand the theory and ideas of the BA algorithm, do a comprehensive review of the related literature. For hardware implementation, this understanding provides a strong basis. • High-Level Architecture: Create a high-level architecture with an emphasis on the CC and JU components. This work offers a well-organized framework for the BA algorithm and points out the parts that affect performance. • Hardware Implementation: Create specialized hardware accelerators by translating the high-level architecture into particular hardware designs. The hardware accelerator needs to effectively process the JU and CC components of the BA algorithm. We aim to focus on the performance of the BA accelerator.