Investigators use digital methods to determine exactly what happened at a car accident
By Brett Duesing
Like a detective, Tom Bundorf of Packer Engineering is after just the facts. Insurance companies hire the engineering consultancy to reconstruct what happened in auto accidents using 3D scientific methodology. So, the Packer analysis often serves as the only objective evidence presented in court to reconcile conflicting testimonies.
"Sometimes witnesses believe they recall the event as crystal clear," Bundorf says. "They will try to estimate the speed [of a vehicle], but given the facts of the scene, we might know that particular speed was not physically possible."
In one of Bundorf’s recent cases, a Chevy Lumina veered into oncoming traffic and struck a Ford F-150 pickup head-on. One witness reported the Lumina speeding around the curve, though the driver claimed he was within the speed limit when he lost control. In question is the matter of driver negligence, which will significantly affect the amount of an insurance award.
Unlike a police detective, Bundorf, the director of laboratory sciences at Packer, is never present at the aftermath of an accident. Instead, the case file lands on his desk in his Ann Arbor, Michigan, office months later. Inside the envelope, Bundorf would like to find, as any scientist would, accurate measurements of the crash scene. However, in most insurance investigations, the only clues are photographs.
Due to the limitations in police resources, says Bundorf, "what you’ll find is that documentation tends to be proportional to the severity of the crash." What that means is fatalities generally receive a full mapping, but for a fender bender, there might not be any documentation. "The majority of the cases we get fall somewhere in the middle," he says. In the accident described above, Packer Engineering did not receive any numeric measurements. All that was left of the Lumina/F-150 collision were six color copies of JPEG images.
But from those rough images of the aftermath, Bundorf and his team can tell what happened in the critical seconds before impact, including the initial velocities of the vehicles, and in more spectacular crashes, the unique choreography of swerves and spins. This is done through PhotoModeler, desktop software from Eos Systems, that reverse engineers real-world objects into accurate three-dimensional measurements.
PhotoModeler generates CAD plans for the refurbishment of ancient buildings, creates mechanical drawings of complex machinery where no original documents exist, and, in the case of accident reconstruction, finds the hard facts—from just a few photographs.
PhotoModeler automates the scientific method of photogrammetry, which transforms a regular camera into an accurate measurement tool. The measurement units come from the interior geometry of the camera to form the basis of photogrammetric analysis. Essentially, these minute distances, collectively referred to as the "simplified interior parameters"—the focal length, lens distortion, format aspect ratio, principal point, and so forth—proportionally complement the exterior scene, enabling the real-world distance calculation of objects shown in a photograph.
PhotoModeler automates photogrammetric analysis on photographs from film, digital, or video cameras. The software can quickly construct a 3D CAD model of an object on the basis of just two scanned photographs taken from different vantage points. To find the simplified interior parameter data, PhotoModeler performs an automatic camera calibration.
Bundorf, however, had one small problem: no access to the original camera. The insurance investigation must do without not only the original camera for calibration purposes, but also knowing the type of camera used to take the accident scene photos. Consequently, the ordinarily easy photogrammetric process must undergo another step to find the basic camera calibration.
Inverse Camera Analysis
Although the crashed car and pickup have long since been towed from the intersection shown in Bundorf’s color copies, many of the background objects remain unchanged. A physical survey of the original location fills in the missing distance information in the equation.
PhotoModeler software determines distances within a photographic scene. In the Inverse Camera (IC) analysis, measured points from a survey of roadside objects provide the basic distance information for a 3D scene.
"We make exhaustive surveys when we go to the site," Bundorf explains. "I measure every feature I can find: any lines, points, edges. In this case, we surveyed guide-rail posts, road centerlines, concrete structures, and adjacent buildings." Many accurate readings must be taken to guard against error, as some seemingly stationary objects such as posts and signs glacially shift in a few months’ time. A ground survey can be achieved through surveying tools or photo-documented with a camera whose calibration is known by PhotoModeler.
The inverse camera (IC) analysis uses the known objects in the photographs as control points to calculate the interior camera information. Conceptually, the control points of the survey determine the unknown camera information, which, in turn, can then dimension the final objects—the vehicles and tire markings in the scene.
PhotoModeler has an IC function, which performs this three-step process at once. As Bundorf adds more and more three-dimensional data on the on-screen photographs, PhotoModeler adjusts the six shots into a probable x, y, z coordinate system. With enough control point adjustments, the 3D version of the scene emerges, complete with distance units.
Dynamic Analysis, Crush Profile
With the IC analysis completed, Bundorf exports a map of the accident scene into Autodesk’s AutoCAD, and passes it to Mike Bracki, an engineer at Packer’s Naperville, Illinois, office, who will examine the dynamics of the collision.
PhotoModeler reveals a true shape of the tire markings: the burnt rubber on the road indicating the extent of a vehicle’s swerve and its braking distance. The program also recasts the photo pixels orthogonally—with the distortion and foreshortening effects of perspective removed, so that the scene appears like an aerial image. This bird’s-eye view of the mark may be included as a layer on the CAD map.
In the Lumina/F-150 scenario, Bracki observed an 85-foot skid mark, starting from the inside lane and ending in the outside lane where the point of impact of vehicles occurred. "If you just assume the Lumina runs into the truck and just stops right there as it hits, and the truck was stopped at impact," explains Bracki, "that gives the Lumina a velocity of about 40 to 41 mph, just from the skid marks alone."
This velocity figure does not take into account the full kinetic behavior of the two vehicles (the PhotoModeler analysis determined that the Lumina was pushed back a few feet after impact), nor does it give a full answer to the speed the Lumina was traveling before the driver braked. The crush profile leads to these final details.
Bracki maps out the differences in the lengths between the crumpled metal and its original shape. The map of vectors, showing the force and angle required to cause those displacements, also gives the engineer clues about the collision. In some more complicated cases, an oblique force might spin the vehicles into other objects. Or, when a car back-ends a truck hauling a John Deere backhoe, for instance, multiple points of damage on the car indicate when and where the loose construction equipment got flung into the mix. The direct head-to-head clash of the Lumina and the pickup is much more straightforward.
"Since [Tom Bundorf] knew where the vehicle was at the point of rest, he could draw these two vehicles from a bird’s-eye view for me and show me what the mutual crush was between the two vehicles," says Bracki. "From that mutual crush, I could calculate the energy of the impact itself." Bracki works out the physics through the total energy method: From the energy required to make the crush, he calculates the real initial velocities of the two vehicles.
CAD views align the photographic evidence in respect to theoriginal camera angles.
"Essentially, the Lumina locked its wheels up from braking, and it was on the inside of the turn. And then the Lumina followed a tangential path into the outside (or oncoming) lane of traffic, where it struck the F-150 head-on," explains Bracki. "We have the car going—at the time when the driver stepped on the brakes—somewhere between 51 and 56 miles an hour. So he’s not speeding; it was a 55 mph zone. Something spooked him as he was coming around that turn, and he just locked up his tires and slid right into oncoming traffic. He was at fault because he crossed the centerline, but there was an accusation that he was coming around that turn at about 100 mph, and that’s not true. He was somewhere within the vicinity of the speed limit. He just made a mistake. He just shouldn’t have locked his brakes at that time because his velocity carried him into oncoming traffic."
Due to Packer Engineering’s detective work, its insurance client could proceed with the claim, confident that the evidence will stand up in court, if challenged. As Bundorf points out, the only alternative to the technological solutions presented here is to go out to the scene and "eyeball" the measurements in respect to the photographs—which, like some drivers’ testimony, would be less than reliable.
Brett Duesing is a technical writer.