Nerve implicit three-dimensional reconstruction SLAM method based on loopback detection

The invention relates to the technical field of image three-dimensional reconstruction, and particularly discloses a neural implicit three-dimensional reconstruction SLAM method based on loopback detection, and the method comprises the steps: S01, carrying out the implicit three-dimensional reconstruction of an object in a scene, carrying out the decoding through an MLP multilayer perceptron, obtaining a feature layer, and calculating the occupancy probability value of the feature layer; s02, calculating a color value and a termination probability according to the occupancy probability value of the feature layering, and rendering the color value and the depth value according to a calculation result to obtain a rendered image; s03, creating a common-view optical flow matrix according to the rendered image, selecting a key frame from the common-view optical flow matrix, and calculating a relative optical flow of the key frame; step S04, performing loopback detection on the relative optical flow of the key frame