A Localization Optimization Algorithm with Improved Feature Tracking Mechanism

Aiming at the traditional SLAM algorithm to extract the matching time consuming, which makes it difficult to run in real time, this paper proposes a fast SLAM algorithm with improved tracking method. The algorithm introduces the tracking decision of fusion of feature and optical flow, filters the input frames first, classifies different frames into feature frames and optical flow frames, extracts feature points only in feature frames, and tracks the points in feature frames in optical flow frames, which avoids time-consuming matching and improves the real-time performance of the algorithm; in the time-sharing asynchronous tracking session, feature tracking and bidirectional optical flow tracking are performed for the classified feature frames and optical flow frames respectively. The test results in the publicly available TUM dataset show that the algorithm of this paper has better advantages in real-time performance and adaptability to the environment.