Single Monocular Image Building Object Scale Estimation

In recent years,three-dimensional reconstruction techniques have been widely used in construction recovery,smart urban construction and intelligent robot navigation,and related reconstruction techniques are increasingly mature.Regardless of the fine surface reconstruction and classification of indoor beam columns and furniture,there is a large number of related research in the reconstruction of architectural contours.Building object scale is estimated to play a crucial role in three-dimensional reconstruction.In order to realize the estimation of the real scale of the building in the outdoor 3D scene,this paper conducts research from two aspects.One is to propose a true scale estimation method based on reference object detection and depth estimation.First,estimate the depth information of the monocular scene image,add the depth information to the three-dimensional space coordinate information,and then convert the pixel coordinates into the camera space coordinates,and then obtain the estimated scale value of the reference object in the image;secondly,obtain the image in the image through target detection Know the true scale of the reference object,and then compare the true scale value of the reference object with the estimated scale value to obtain the correction coefficient of the building size estimation;finally,use the correction coefficient to re-estimate the true scale of other buildings in the image.The rationality and effectiveness of the proposed correction coefficients are demonstrated.The experimental results show that the estimation method of building body size with correction coefficients has smaller errors.Compared with calculating the object scale information directly from the depth information estimation algorithm,the scale estimation method after introducing the real scale information of the known object to modify the scale estimation method can effectively improve the estimation accuracy of the building body scale in the monocular image.The second is that it is difficult to find a parallel line group that meets the requirements with the traditional three-vanishing point calibration method from the monocular image of the outdoor real scene.Using the real-scale information of the reference object(for example:brick)that can be obtained from the image,generalization is proposed.A higher building size estimation algorithm based on two vanishing point calibrations and a reference line segment of known scale.First,obtain the two sets of parallel lines on the building to be estimated on the monocular image through user interaction,and then calculate the two vanishing point coordinates of the monocular image,and then use the two vanishing points to calibrate the monocular image to calculate The corresponding rotation matrix is then used to calculate the translation vector with the reference object of known real scale.Finally,the three-dimensional information of the building body are obtained through coordinate transformation to realize the monocular image building body scale estimation.The experimental results show that this method can effectively estimate the true scale information of the building in the monocular image.