Active Bionic Visual SLAM and Semantic Map Generation in Building Environments

Simultaneous localization and mapping (SLAM) have extensive applications in the field of robotics. However, current visual SLAM mainly constructs semantic maps using passive image information and lacks the ability to autonomously explore the environment. To address this issue, this paper proposes an active semantic capture SLAM system based on a bionic stereo camera, designed for structured environments such as corridors. The system first uses a pose estimation network for initial target localization, then uses the six-degree-of-freedom bionic stereo camera to dynamically adjust the viewpoint, autonomously control the camera to fixate on the target, and optimize the target's pose. Additionally, it analyzes surrounding feature points to determine the position of textual information, guiding the camera to capture it. Experimental results show that, compared to the fixed ORB-SLAM2, the proposed system improves target localization accuracy by an average of 36.1%, effectively integrates object and textual semantic information, and enriches the semantic map.