To test the profile measuring accuracy of a hail dent recorder, the HDR 4200, we built a ½-scale automotive panel (“mini-hood”) using metals and automotive finishes comparable to a typical automobile. We dented the panel to simulate hail damage. Next, we contracted with a third-party company to measure the mini-hood using a highly accurate Zeiss Comet (“Comet”) structured light system. Using the Comet, ground truth dent profiles were measured to accuracies of 10 microns. In a lab at Quidient’s Technology Center, we used a Hail Damage Recorder (HDR) 4200 prototype to reconstruct the lab (including the light field) and the mini-hood from various angles. The HDR 4200 was comprised of a Quidient Locale Engine and a Teledyne Dalsa Genie Nano C2450 color/polarized camera. Averaged over 25 different dents, our resulting comparison shows agreement to ground truth to 28 microns for dent profile and 5.6 mm for dent size (SR), putting our 4200 on par with hail dent assessment experts whose repeatability when estimating dent size is about 5 mm.
Our first step was to purchase a ½ scale aluminum mini-hood. We added a stiff supporting structure to make the mini-hood stable for purposes of this study. To make realistic dents, we worked with an industry expert and created a procedure for dropping various sized metal balls through a PVC tube from different measured heights. Rather than create a regular pattern, we opted for a more realistic random mix of dent sizes and locations spread over the four painted quadrants of the mini-hood. Finally, we painted the mini-hood using four different colors (white, black, green and red) of high-gloss automotive paint.
We contracted with Direct Dimensions (DD) (www.dirdim.com) to create a digital scan of our mini-hood. First, DD applied a powder type coating to help reduce specular reflectance. After this, DD used a Zeiss Comet structured light system to capture a 3D ground-truth digital scan accurate to within 10 microns. After the DD scan and prior to our test scan using our HDR 4200, the powder coating was removed to return the mini-hood to its original conventional (shiny) state.
We then used our 4200, running a configuration of our Locale Scene Reconstruction Engine V1.4 interfaced with the Genie Nano C2450 polarimetric camera, to capture several images of our lab for pre-establishing the quasi steady-state light field. After this we took single-shot images of the mini-hood from different angles capturing dents of various sizes from each of the colored quadrants.
These images were then registered to the 3D ground-truth digital scan created by the Comet. In the figure (and associated video) below, we show on the left in red our 3D dent profile based upon a single image as compared to the 3D dent profile recorded by the Comet on the right in green.
Each of these 3D profiles can also be viewed as a 2D profile on a dent center plane as shown in the figure on the right. We then set out to compare both the size (S) and depth (D) of each dent as measured by our 4200 compared to the Zeiss Comet.
To make this comparison, we first determined a Normalized Surface located above the dent profile (see the dashed line in the figure above). This Normalized Surface can be derived for example from a 3D CAD model that is registered to each scan. As an alternative method, we calculated the Normalized Surface by measuring the undisturbed surface around each dent and interpolating using Class A (see wikipedia) surface (smoothness) considerations.
Next, we separately compared both the red and green 3D dent profiles to the Normalized Surface. Each red and green profile comparison to the Normalized Surface resulted in deviations along the (diametric) dent profile. We then fit these deviations to a 2D bell curve (one curve matching our red profile and the other curve matching the ground truth green profile). For each red and green 2D bell curve, we set the “Reference Depth” (DR) to be the maximum depth at the center of the bell curve and the “Reference Size” (SR) to be 6 sigma (σ), statistically representing 99.7% of the curve, or profile.
Results and Calculations
As shown in the table below, we compared the “(a) Ref Depth” (DR) and “(b) Ref Size” (SR) as determined using the ground truth’s (green) profile with the Ref Depth (DR) and Ref Size (SR) as determined using our HDR 4200’s (red) profile. Column “(c) Depth Error” calculates the difference between the ground truth’s Ref Depth and our 4200’s Ref Depth, shown to be 83 microns on average (i.e. less than 0.1 mm). Column “(d) Size Error” calculates the difference between the ground truth’s Ref Size and our 4200’s Ref Size, shown to be 5.6 mm on average, roughly on par with an expert. And finally, we provide our “(e) Profile Error” as compared to the ground-truth profile, where the Profile Error is measured as the RMSE (root mean square error) across several points along the profile curvature, where the average profile error was 28 microns.
We have demonstrated that it is possible for any conscientious person such as the vehicle owner to measure hail dent size to a vehicle comparable to an industry expert using our low-cost, hand-held HDR 4200 prototype. Furthermore, our 4200 can automatically locate each dent with respect to each panel of the vehicle and further create a highly accurate 3D model of each dent’s profile. By locating and measuring each dent, our data can be used by estimation software systems to estimate repair costs. This recordation can also be used for remote 3D inspection by repair shops to plan repairs. By storing these dent profile records in association with the vehicle, insurers will have a tool, for example, for distinguishing between hail events, identifying fraud, and determining non-related damage. We note that these results are based on single images. It is likely that better accuracies will be obtained by capturing multiple images (e.g. in a “video mode”