Vehicle inspection system and method with three-dimensional defect clustering on CAD models

A method for consolidating vehicle defect detections from multiple viewpoints, comprising:

• capturing a plurality of images of a vehicle from different viewpoints using a multi-camera array;
• detecting, in the plurality of images, multiple defect detections that potentially correspond to a same physical defect observed from different viewpoints;
• obtaining a three-dimensional reference model representing geometry of the vehicle;
• for each defect detection, determining a three-dimensional location on the reference model by projecting the defect detection from its two-dimensional image location onto a surface of the reference model;
• clustering the three-dimensional locations on the reference model based on spatial proximity to identify groups of defect detections corresponding to the same physical defect;
• and
• generating consolidated defect data representing each identified physical defect with a single spatial location derived from the clustered three-dimensional locations, wherein clustering the three-dimensional locations comprises:
• computing geodesic distances along the reference model surfaces using mesh connectivity information;
• applying density-based clustering algorithms with adaptive distance thresholds based on local surface curvature;
• and
• consolidating defect clusters with cluster validity indices ensuring each cluster represents a single physical defect with confidence exceeding a predetermined threshold.

A method for consolidating vehicle defect detections from multiple viewpoints, comprising:

• capturing a plurality of images of a vehicle from different viewpoints using a multi-camera array;
• detecting, in the plurality of images, multiple defect detections that potentially correspond to a same physical defect observed from different viewpoints;
• obtaining a three-dimensional reference model representing geometry of the vehicle;
• for each defect detection, determining a three-dimensional location on the reference model by projecting the defect detection from its two-dimensional image location onto a surface of the reference model;
• clustering the three-dimensional locations on the reference model based on spatial proximity to identify groups of defect detections corresponding to the same physical defect;
• and
• generating consolidated defect data representing each identified physical defect with a single spatial location derived from the clustered three-dimensional locations;
• wherein the three-dimensional reference model comprises a computer-aided design (CAD) model of the vehicle retrieved from a database based on a vehicle identifier,
• wherein projecting the defect detection comprises:
• applying inverse camera calibration matrices to convert two-dimensional pixel coordinates to normalized camera coordinates;
• constructing a three-dimensional ray vector extending from a camera optical center through the normalized coordinates into world space using camera extrinsic rotation and translation parameters;
• and
• computing ray-surface intersection between the ray vector and triangular mesh faces of the CAD model to yield three-dimensional coordinates where the defect observation ray intersects the vehicle surface.

A vehicle inspection system comprising:

• a multi-camera array positioned to capture images of a vehicle from multiple viewpoints;
• a memory which stores program instructions and a template library;
• one or more processors configured to access the memory for:
• retrieving a reference template from the template library and a three-dimensional CAD model corresponding to the vehicle;
• detecting defects by comparing the captured images to the reference template;
• projecting detected defects from two-dimensional image locations onto the three-dimensional CAD model;
• clustering the projected defects on the CAD model to consolidate redundant detections from different viewpoints;
• and
• generating an inspection report with consolidated defect locations;
• wherein the one or more processors are configured to project detected defects by:
• computing, for each defect detection at two-dimensional pixel coordinates in a camera image, a ray extending from a camera optical center through the pixel coordinates into three-dimensional space;
• and
• determining an intersection point between the ray and a surface of the CAD model using ray-casting algorithms and calibrated camera parameters.

A vehicle inspection system comprising:

• a multi-camera array positioned to capture images of a vehicle from multiple viewpoints;
• a memory which stores program instructions and a template library;
• one or more processors configured to access the memory for:
• retrieving a reference template from the template library and a three-dimensional CAD model corresponding to the vehicle;
• detecting defects by comparing the captured images to the reference template;
• projecting detected defects from two-dimensional image locations onto the three-dimensional CAD model;
• clustering the projected defects on the CAD model to consolidate redundant detections from different viewpoints;
• and
• generating an inspection report with consolidated defect locations;
• wherein the one or more processors are configured to cluster the projected defects by:
• computing geodesic distances between three-dimensional defect locations along surfaces of the CAD model;
• and
• grouping defect locations having geodesic distances below a threshold distance into clusters representing unique physical defects.

A vehicle inspection system comprising:

• a multi-camera array positioned to capture images of a vehicle from multiple viewpoints;
• a memory which stores program instructions and a template library;
• one or more processors configured to access the memory for:
• retrieving a reference template from the template library and a three-dimensional CAD model corresponding to the vehicle;
• detecting defects by comparing the captured images to the reference template;
• projecting detected defects from two-dimensional image locations onto the three-dimensional CAD model;
• clustering the projected defects on the CAD model to consolidate redundant detections from different viewpoints;
• and
• generating an inspection report with consolidated defect locations;
• wherein the one or more processors are further configured to:
• determine camera poses for cameras in the multi-camera array relative to the CAD model by optimizing reprojection error between projected CAD features and observed vehicle features in the captured images.