Interior defect detection through vehicle glazing using multi-view motion analysis and asymmetric illumination
A method and system for detecting interior defects in vehicles through glazing without requiring door opening or interior sensor installation. The system captures multi-view images of vehicle window panes from spatially separated viewpoints during relative motion between vehicle and imaging device. Motion field data representing pixel displacement is computed and used to classify pixels into interior-origin pixels exhibiting apparent depth behind the window pane plane and reflection-origin pixels exhibiting apparent depth at the glass surface. An interior-enhanced image is generated through motion refocusing that sharpens interior textures while blurring glass-plane reflections. Optional asymmetric illumination and polarization provide additional reflection suppression. Transmittance compensation accounts for glazing tint and spectral characteristics. Defect detection algorithms identify interior defects including seat damage, missing components, and structural defects, with privacy-aware redaction of occupants integrated into the processing pipeline.
1 . A method for detecting interior defects in a vehicle through glazing, comprising: capturing, with an imaging device positioned external to the vehicle, a plurality of images of a vehicle window pane from at least two spatially separated viewpoints as the vehicle changes a position relative to the imaging device;
computing, for pixels within a region of interest corresponding to the window pane, motion field data representing pixel displacement;
classifying the pixels based on the motion field data into interior-origin pixels and reflection-origin pixels by comparing apparent depth of the pixels relative to a plane of the window pane, wherein pixels with apparent depth behind the window pane plane are classified as interior-origin pixels and pixels with apparent depth at or on the window pane plane are classified as reflection-origin pixels;
generating an interior-enhanced image using the interior-origin pixels and the reflection-origin pixels; and
applying a defect detection algorithm to the interior-enhanced image to identify one or more interior defects and outputting defect metadata identifying a detected interior defect and its location within the vehicle cabin.
2 . The method of claim 1 , wherein the plurality of images are captured during a temporal burst that spans 50-200 milliseconds.
3 . The method of claim 2 , wherein the temporal burst comprises 3-10 frames per viewpoint.
4 . The method of claim 1 , wherein the at least two spatially separated viewpoints comprise at least two cameras of a multi-camera rig.
5 . The method of claim 1 , wherein classifying the pixels comprises calculating, from the motion field data, a disparity signature for each pixel that indicates the apparent depth.
6 . The method of claim 5 , wherein calculating the disparity signature comprises calculating from the motion field data and known camera geometry.
7 . The method of claim 1 , wherein classifying pixels as reflection-origin pixels comprises determining that the apparent depth matches a specular flow pattern corresponding to reflections on the window pane plane.
8 . The method of claim 1 , wherein generating the interior-enhanced image comprises:
warping frames from the plurality of images according to an interior depth hypothesis using vehicle pose change and camera rig geometry to refocus on interior surfaces; and
aggregating the warped frames with weighting that suppresses the reflection-origin pixels, thereby sharpening interior textures while blurring glass-plane reflections.
9 . The method of claim 1 , further comprising:
estimating a per-pane transmittance map for the window pane; and
applying transmittance compensation to the interior-enhanced image based on the per-pane transmittance map.
10 . The method of claim 1 , further comprising:
detecting occupants or faces in the interior-enhanced image; and
redacting the occupants or faces before outputting the defect metadata.
11 . The method of claim 1 , wherein capturing the plurality of images comprises:
firing a left-bank illumination strobe at a first time;
capturing a first frame with the left-bank illumination strobe active;
firing a right-bank illumination strobe at a second time within 20 milliseconds of the first time; and
capturing a second frame with the right-bank illumination strobe active.
12 . The method of claim 11 , further comprising:
computing a reflection-response metric based on a difference between the first frame and the second frame; and
suppressing pixels with high reflection-response metric values during generation of the interior-enhanced image.
13 . The method of claim 1 , wherein capturing the plurality of images comprises capturing with one or more linear polarizing analyzers positioned in front of camera lenses of the imaging device.
14 . A system for detecting interior defects in a vehicle through glazing, comprising:
an imaging device configured to be positioned external to the vehicle and to capture a plurality of images of a vehicle window pane from at least two spatially separated viewpoints as the vehicle changes a position relative to the imaging device;
one or more processors; and
memory storing instructions that, when executed by the one or more processors, cause the system to:
compute, for pixels within a region of interest corresponding to the window pane, motion field data representing pixel displacement;
classify the pixels based on the motion field data into interior-origin pixels and reflection-origin pixels by comparing apparent depth of the pixels relative to a plane of the window pane, wherein pixels with apparent depth behind the window pane plane are classified as interior-origin pixels and pixels with apparent depth at or on the window plane classified pane are as reflection-origin pixels;
detect one or more interior defects and its location within the vehicle cabin based on an analysis of the interior-origin pixels and the reflection-origin pixels and
update an interface to present the one or more interior defects and its location within the vehicle cabin.
15 . The system of claim 14 , wherein the imaging device comprises a multi-camera rig with at least two cameras providing the at least two spatially separated viewpoints.
16 . The system of claim 14 , wherein the imaging device is configured to capture the plurality of images during a temporal burst that spans 50-200 milliseconds and comprises 3-10 frames per viewpoint.
17 . The system of claim 14 , further comprising illumination sources configured to provide asymmetric illumination to the vehicle window pane.
18 . The system of claim 17 , wherein the illumination sources comprise a left-bank strobe and a right-bank strobe controllable to fire within 20 milliseconds of each other.
19 . The system of claim 14 , wherein the imaging device includes one or more linear polarizing analyzers positioned in front of camera lenses.
20 . The system of claim 14 , wherein the memory stores further instructions that, when executed, cause the system to:
estimate a per-pane transmittance map for the window pane; and
apply transmittance compensation to the interior-enhanced image based on the per-pane transmittance map.