System and method for object dimension estimation using 3D models

A non-transitory computer readable medium storing a computing device program product for estimation of dimensions of an object present on images obtained from a set of N cameras and representing the same scene from different viewpoints, based on a 3D object model, the computing device program product comprising:

• first programmatic instructions for determining a 3D Morphable Model (3DMM) model;
• second programmatic instructions for matching the 3DMM model to each of N images of the set in order to determine a set of parameters hi, ρi, λi, wherein hi describes the orientation of the matched model with respect to the i-th image;
• ρi describes the 3D shape of the model matched to the t-th image and λi describes the appearance/texture of the model matched to the i-th image;
• third programmatic instructions for calculating relative positions of the cameras, from which the images were collected, based on 3D transformations Hij of the matched 3DMM model from the i-th image to the matched 3DMM model from the j-th image;
• fourth programmatic instructions for calculating the total matching error Ec for the model as a sum of matching errors of a model described by parameters hc, ρc, λc to individual images:
• E
• c
• =
• ∑
• i
• =
• 1
• N
• ⁢
• E
• i
• fifth programmatic instructions for estimating the dimensions of the object based on locations of characteristic points from vector s, calculated as:
• s
• =
• s
• _
• +
• ∑
• k
• ⁢
• p
• k
• ·
• s
• ^
• k
• wherein s is the vector of the average 3DMM model shape, ρk are elements of vector ρc, and ŝk are eigenvalue vectors of the 3DMM model.

A system for estimation of dimensions of an object present on images obtained from a set of N cameras and representing the same scene from different viewpoints, based on a 3D object model, the system comprising:

• a 3DMM model generator for determining a 3D Morphable Model (3DMM) model;
• a model matching module for matching the 3DMM model to each of N images of the set in order to determine a set of parameters hi, ρi, λi, wherein hi describes the orientation of the matched model with respect to the i-th image;
• ρi describes the 3D shape of the model matched to the r-th image and λi describes the appearance/texture of the model matched to the i-th image, the matching being performed by:
• calculating relative positions of the cameras, from which the images were collected, based on 3D transformations Hij of the matched 3DMM model from the i-th image to the matched 3DMM model from the j-th image;
• calculating the total matching error Ec for the model as a sum of matching errors of a model described by parameters hc, ρc, λc to individual images:
• E
• c
• =
• ∑
• i
• =
• 1
• N
• ⁢
• E
• i
• a dimensions estimator for estimating the dimensions of the object based on locations of characteristic points from vector s, calculated as:
• s
• =
• s
• _
• +
• ∑
• k
• ⁢
• p
• k
• ·
• s
• ^
• k
• wherein s is the vector of the average 3DMM model shape, ρk are elements of vector ρc, and ŝk are eigenvalue vectors of the 3DMM model.