Automatic assessment of damage and repair costs in vehicles

A non-transitory computer-readable storage medium storing instruction that, when executed by a processor, cause a computer system to perform the steps of:

• performing computerized image processing based on two or more images to generate two or more damage detection images, wherein each damage detection image is a two dimensional (2D) image that includes indications of one or more areas of damage to a vehicle in the damage detection image;
• training a Convolutional Neural Network (CNN) with the two or more damage detection images to detect the indications of the one or more areas of the damage to the vehicle in the damage detection image;
• executing the CNN to detect the indications of the one or more areas of the damage to the vehicle in the damage detection image;
• mapping the two or more damage detection images to a three-dimensional (3D) model of the vehicle to generate a damaged 3D model that indicates one or more areas of the vehicle that are damaged, wherein the damaged 3D model includes a plurality of vertices,
• wherein, for each vertex in the plurality of vertices, an angle for the vertex is formed between (i) a line from a camera center to the vertex, and (ii) a normal direction associated with the vertex,
• wherein, for each vertex in the plurality of vertices, the vertex is visible to a camera in response to the angle for the vertex being smaller than 90 degrees and the vertex is invisible to the camera in response to the angle for the vertex being larger than 90 degrees, and
• wherein each vertex in the plurality of vertices that is visible to the camera is associated with a damage intensity value from a corresponding region of a damage detection image from the two or more damage detection images, wherein multiple damage detection images include regions that correspond to a first vertex on the damaged 3D model, wherein the first vertex is visible to the camera, and wherein a maximum damage intensity value at the first vertex from the multiple damage detection images is selected as the damage intensity value for first vertex;
• and
• calculating an estimated repair cost for the vehicle based on the maximum damage intensity value at the first vertex of the damaged 3D model.

A server device, comprising:

• a memory storing instructions;
• and a processor configured to execute the instructions to cause the server device to:
• perform computerized image processing based on two or more images to generate two or more damage detection images, wherein each damage detection image is a two dimensional (2D) image that includes indications of one or more areas of damage to a vehicle in the damage detection image;
• train a Convolutional Neural Network (CNN) with the two or more damage detection images to detect the indications of the one or more areas of the damage to the vehicle in the damage detection image;
• execute the CNN to detect the indications of the one or more areas of the damage to the vehicle in the damage detection image;
• map the two or more damage detection images to a three-dimensional (3D) model of the vehicle to generate a damaged 3D model that indicates one or more areas of the vehicle that are damaged, wherein the damaged 3D model includes a plurality of vertices,
• wherein, for each vertex in the plurality of vertices, an angle for the vertex is formed between (i) a line from a camera center to the vertex, and (ii) a normal direction associated with the vertex,
• wherein, for each vertex in the plurality of vertices, the vertex is visible to a camera in response to the angle for the vertex being smaller than 90 degrees and the vertex is invisible to the camera in response to the angle for the vertex being larger than 90 degrees, and
• wherein each vertex in the plurality of vertices that is visible to the camera is associated with a damage intensity value from a corresponding region of a damage detection image from the two or more damage detection images, wherein multiple damage detection images include regions that correspond to a first vertex on the damaged 3D model, wherein the first vertex is visible to the camera, and wherein a maximum damage intensity value at the first vertex from the multiple damage detection images is selected as the damage intensity value for first vertex;
• and
• calculate an estimated repair cost for the vehicle based on the maximum damage intensity value at the first vertex of the damaged 3D model.