Real-time 3D scene reconstruction method for UAV based on real-time simultaneous positioning and map construction

Abstract

The present invention proposes a real-time 3D scene reconstruction method for UAVs based on EG-SLAM. The method utilizes the visual information obtained in real time by the onboard camera of the UAV to reconstruct a large-scale 3D scene with texture details. Compared with many existing methods, the present invention directly runs on the CPU after collecting images, realizes positioning and reconstructs a three-dimensional map quickly and in real time; and does not adopt the traditional PNP method to solve the pose of the UAV, but uses the invention The EG‑SLAM method to solve the pose of the UAV, that is, to use the feature point matching relationship between two frames to directly solve the pose, reducing the requirement for the repetition rate of the captured image; in addition, the large amount of environmental information obtained makes the UAV With a more precise and detailed perception of the environmental structure, texture rendering is performed on the large-scale 3D point cloud map generated in real time, which realizes the reconstruction of the large-scale 3D map and obtains a more intuitive and realistic 3D scene.

Description

technical field [0001] The present invention relates to the fields of unmanned aerial vehicle autonomous flight, computer vision, and map construction. Specifically, it uses real-time synchronous positioning and map construction (EG-SLAM) to perform dense diffusion and texture rendering on the obtained 3D point cloud to realize the realization of unmanned aerial vehicles. 3D scene reconstruction of ground objects. Background technique [0002] Real-time simultaneous positioning and reconstruction of 3D scenes by UAVs has always been a hotspot and difficulty in the field of UAV autonomous flight and computer vision. UAVs have many advantages, such as: small size, low cost, good stealth, flexibility and convenience, and strong combat capability. In recent years, they have been widely used in many fields of military and civilian. In terms of military affairs, drones can be used for military tasks such as reconnaissance, attack, and electronic countermeasures. They can also be ...

AI Technical Summary

Problems solved by technology

[0006] 2) The redundancy of the initial image information obtained by the camera is extremely high, resulting in a waste of transmission bandwidth and computing resources

[0007] 3) At present, most of the large-scale maps generated by drones in real time using computer vision technology are 3D point clouds that only have geometric structures and do not contain texture information.

Images