Performance Evaluation of Feature Detection Algorithms and Their Impact on the Accuracy and Efficiency of Visual Odometry

Feature detection is a critical component of visual odometry, directly influencing position estimation accuracy. This process forms the basis for identifying key points in images, playing a pivotal role in subsequent operations such as feature matching and motion tracking. This study examines the impact of various feature detection algorithms on position estimation accuracy in visual odometry, focusing on a comparative analysis of the Harris, FAST, SIFT, CenSurE, and ORB algorithms. Performance evaluation was conducted based on accuracy and computational efficiency in position estimation. Each algorithm's average errors and processing times were calculated and systematically compared using an image dataset. Results indicate that the CenSurE algorithm is optimal for real-time applications and scenarios demanding rapid processing due to its lower computational cost. Its high-speed feature extraction capability makes it particularly suitable for such use cases. Conversely, despite its higher processing time, the Harris algorithm offers superior accuracy in position estimation and angular measurement, making it a preferred choice when precision is prioritized over speed. The FAST and SIFT algorithms balance accuracy and computational efficiency; the FAST algorithm, with its lower processing time, performs effectively in horizontal orientations, whereas the Harris algorithm excels in precision. The ORB algorithm exhibits moderate speed and acceptable performance but demonstrates reduced accuracy in certain positional features. This study enhances the understanding of the trade-offs between accuracy and efficiency in feature detection for visual odometry, providing a foundation for further research in optimizing algorithm selection for specific applications.