Search Results

Forensic disciplines are called upon to locate evidence from a single camera or static video camera, and both the angle of incidence and resolution can limit the accuracy of single image photogrammetry. This research compares a baseline of known 3D data points representing evidence locations to evidence locations determined through single image photogrammetry and evaluates the effect that object resolution (measured in pixels), and angle of incidence has on accuracy. Solutions achieved using an automated process where a camera match alignment is calculated from common points in the 2D imagery and the 3D environment, were compared to solutions achieved in a more manual method by iteratively adjusting the camera’s position, orientation, and field-of-view until an alignment is achieved. This research independently utilizes both methods to achieve photogrammetry solutions and to locate objects within a 3D environment.

Photogrammetry and the accuracy of a photogrammetric solution is reliant on the quality of photographs and the accuracy of pixel location within the photographs. A photograph with lens distortion can create inaccuracies within a photogrammetric solution. Due to the curved nature of a camera’s lens(s), the light coming through the lens and onto the image sensor can have varying degrees of distortion. There are commercially available software titles that rely on a library of known cameras, lenses, and configurations for removing lens distortion. However, to use these software titles the camera manufacturer, model, lens and focal length must be known. This paper presents two methodologies for removing lens distortion when camera and lens specific information is not available. The first methodology uses linear objects within the photograph to determine the amount of lens distortion present. This method will be referred to as the straight-line method.

Small unmanned aerial systems have gained prominence in their use as tools for mapping the 3-dimensional characteristics of accident sites. Typically, the process of mapping an accident site involves taking a series of overlapping, high resolution photographs of the site, and using photogrammetric software to create a point cloud or mesh of the site. This process, known as image-based scanning, is explored and analyzed in this paper. A mock accident site was created that included a stopped vehicle, a bicycle, and a ladder. These objects represent items commonly found at accident sites. The accident site was then documented with several different unmanned aerial vehicles at differing altitudes, with differing flight patterns, and with different flight control software. The photographs taken with the unmanned aerial vehicles were then processed with photogrammetry software using different methods to scale and align the point clouds.

When a vehicle with protruding wheel studs makes contact with another vehicle or object in a sideswipe configuration, the tire sidewall, rim and wheel studs of that vehicle can deposit distinct geometrical damage patterns onto the surfaces it contacts. Prior research has demonstrated how relative speeds between the two vehicles or surfaces can be calculated through analysis of the distinct contact patterns. This paper presents a methodology for performing this analysis by visually modeling the interaction between wheel studs and various surfaces, and presents a method for automating the calculations of relative speed between vehicles. This methodology also augments prior research by demonstrating how the visual modeling and simulation of the wheel stud contact can extend to almost any surface interaction that may not have any previous prior published tests, or test methods that would be difficult to setup in real life.

This paper presents a methodology for accurately representing dawn and dusk lighting conditions (twilight) through photographs and video recordings. Attempting to generate calibrated photographs and video during twilight conditions can be difficult, since the time available to capture the light changes rapidly over time. In contrast, during nighttime conditions, when the sun is no longer contributing light directly or indirectly through the sky dome, matching a specific time of night is not as relevant, as man-made lights are the dominate source of illumination. Thus, the initial setup, calibration and collection of calibrated video, when it is dark, is not under a time constraint, but during twilight conditions the time frame may be narrow. This paper applies existing methods for capturing calibrated footage at night but develops a method for adjusting the footage in the event matching an exact time during twilight is necessary.

Accurately reconstructing the speed of a yawing and braking vehicle requires an estimate of the varying rates at which the vehicle decelerated. This paper explores the accuracy of several approaches to making this calculation. The first approach uses the Bakker-Nyborg-Pacejka (BNP) tire force model in conjunction with the Nicolas-Comstock-Brach (NCB) combined tire force equations to calculate a yawing and braking vehicle's deceleration rate. Application of this model in a crash reconstruction context will typically require the use of generic tire model parameters, and so, the research in this paper explored the accuracy of using such generic parameters. The paper then examines a simpler equation for calculating a yawing and braking vehicle's deceleration rate which was proposed by Martinez and Schlueter in a 1996 paper. It is demonstrated that this equation exhibits physically unrealistic behavior that precludes it from being used to accurately determine a vehicle's deceleration rate.

This paper presents a methodology for determining the position and speed of objects such as vehicles, pedestrians, or cyclists that are visible in video footage captured with only one camera. Objects are tracked in the video footage based on the change in pixels that represent the object moving. Commercially available programs such as PFTracktm and Adobe After Effectstm contain automated pixel tracking features that record the position of the pixel, over time, two dimensionally using the video’s resolution as a Cartesian coordinate system. The coordinate data of the pixel over time can then be transformed to three dimensional data by ray tracing the pixel coordinates onto three dimensional geometry of the same scene that is visible in the video footage background.

Previous studies have reported and validated equations for calculating the lean angle required for a motorcycle and rider to traverse a curved path at a particular speed. In 2015, Carter, Rose, and Pentecost reported physical testing with motorcycles traversing curved paths on an oval track on a pre-marked range in a relatively level parking lot. Several trends emerged in this study. First, while theoretical lean angle equations prescribe a single lean angle for a given lateral acceleration, there was considerable scatter in the real-world lean angles employed by motorcyclists for any given lateral acceleration level. Second, the actual lean angle was nearly always greater than the theoretical lean angle. This prior study was limited in that it only examined the motorcycle lean angle at the apex of the curves. The research reported here extends the previous study by examining the accuracy of the lean angle formulas throughout the curves.

In the 2016 SAE publication “Data Acquisition using Smart Phone Applications,” Neale et al., evaluated the accuracy of basic fitness applications in tracking position and elevation using the GPS and accelerometer technology contained within the smart phone itself [1]. This paper further develops the research by evaluating mid-level applications. Mid-level applications are defined as ones that use a phone’s internal accelerometer and record data at 1 Hz or greater. The application can also utilize add-on devices, such as a Bluetooth enabled GPS antenna, which reports at a higher sample rate (10 Hz) than the phone by itself. These mid-level applications are still relatively easy to use, lightweight and affordable [2], [3], [4], but have the potential for higher data sample rates for the accelerometer (due to the software) and GPS signal (due to the hardware). In this paper, Harry’s Lap Timer™ was evaluated as a smart phone mid-level application.

This paper presents a methodology for generating photo realistic computer simulation environments of nighttime driving scenarios by combining nighttime photography and videography with video tracking [1] and projection mapping [2] technologies. Nighttime driving environments contain complex lighting conditions such as forward and signal lighting systems of vehicles, street lighting, and retro reflective markers and signage. The high dynamic range of nighttime lighting conditions make modeling of these systems difficult to render realistically through computer generated techniques alone. Photography and video, especially when using high dynamic range imaging, can produce realistic representations of the lighting environments. But because the video is only two dimensional, and lacks the flexibility of a three dimensional computer generated environment, the scenarios that can be represented are limited to the specific scenario recorded with video.

This paper presents analyses of 21real-world pedestrian versus vehicle collisions that were video recorded from vehicle dash mounted cameras or surveillance cameras. These pedestrian collisions have in common an impact configuration where the pedestrian was at the side of the vehicle, or with a minimal overlap at the front corner of the vehicle (less than one foot overlap). These impacts would not be considered frontal impacts [1], and as a result determining the speed of the vehicle by existing methods that incorporate the pedestrian travel distance post impact, or by assessing vehicle damage, would not be applicable. This research examined the specific interaction of non-frontal, side-impact, and minimal overlap pedestrian impact configurations to assess the relationship between the speed of the vehicle at impact, the motion of the pedestrian before and after impact, and the associated post impact travel distances.

All camera lenses contain optical aberrations as a result of the design and manufacturing processes. Lens aberrations cause distortion of the resulting image captured on film or a sensor. This distortion is inherent in all lenses because of the shape required to project the image onto film or a sensor, the materials that make up the lens, and the configuration of lenses to achieve varying focal lengths and other photographic effects. The distortion associated with lenses can cause errors to be introduced when photogrammetric techniques are used to analyze photographs of accidents scenes to determine position, scale, length and other characteristics of evidence in a photograph. This paper evaluates how lens distortion can affect images, and how photogrammetrically measuring a distorted image can result in measurement errors.

The accident reconstruction community has previously relied upon photographs and site visits to recreate a scene. This method is difficult in instances where the site has changed or is not accessible. In 2017 the United States Geological Survey (USGS) released historical 3D point clouds (LiDAR) allowing for access to digital 3D data without visiting the site. This offers many unique benefits to the reconstruction community including: safety, budget, time, and historical preservation. This paper presents a methodology for collecting this data and using it in conjunction with aerial imagery, and camera matching photogrammetry to create 3D computer models of the scene without a site visit.

This paper introduces a method for calculating vehicle speed and uncertainty range in speed from video footage. The method considers uncertainty in two areas; the uncertainty in locating the vehicle’s position and the uncertainty in time interval between them. An abacus style timing light was built to determine the frame time and uncertainty of time between frames of three different cameras. The first camera had a constant frame rate, the second camera had minor frame rate variability and the third had more significant frame rate variability. Video of an instrumented vehicle traveling at different, but known, speeds was recorded by all three cameras. Photogrammetry was conducted to determine a best fit for the vehicle positions. Deviation from that best fit position that still produced an acceptable range was also explored. Video metadata reported by iNPUT-ACE and Mediainfo was incorporated into the study.

This paper presents yaw testing of vehicles with tread removed from tires at various locations. A 2004 Chevrolet Malibu and a 2003 Ford Expedition were included in the test series. The vehicles were accelerated up to speed and a large steering input was made to induce yaw. Speed at the beginning of the tire mark evidence varied between 33 mph and 73 mph. Both vehicles were instrumented to record over the ground speed, steering angle, yaw angle and in some tests, wheel speeds. The tire marks on the roadway were surveyed and photographed. The Critical Speed Formula has long been used by accident reconstructionists for estimating a vehicle’s speed at the beginning of yaw tire marks. The method has been validated by previous researchers to calculate the speed of a vehicle with four intact tires. This research extends the Critical Speed Formula to include yawing vehicles following a tread detachment event.

In 1980, research by Thebert introduced the use of photography equipment and transparencies for onsite reverse camera projection photogrammetry [1]. This method involved taking a film photograph through the development process and creating a reduced size transparency to insert into the cameras viewfinder. The photographer was then able to see both the image contained on the transparency, as well as the actual scene directly through the cameras viewfinder. By properly matching the physical orientation and positioning of the camera it was possible to visually align the image on the image on the transparency to the physical world as viewed through the camera. The result was a solution for where the original camera would have been located when the photograph was taken. With the original camera reverse-located, any evidence in the transparency that is no longer present at the site could then be replaced to match the evidences location in the transparency.

The accident reconstruction community relies on photogrammetry for taking measurements from photographs. Camera matching, a close-range photogrammetry method, is a particularly useful tool for locating accident scene evidence after time has passed and the evidence is no longer physically visible. In this method, objects within the accident scene that have remained unchanged are used as a reference for locating evidence that is no longer physically available at the scene such as tire marks, gouge marks, and vehicle points of rest. Roadway lines, edges of pavement, sidewalks, signs, posts, buildings, and other structures are recognizable scene features that if unchanged between the time of accident and time of analysis are beneficial to the photogrammetric process. In instances where these scene features are limited or do not exist, achieving accurate photogrammetric solutions can be challenging.

This paper investigates the dynamics of four motorcycle crashes that occurred on or near a curve (Edwards Corner) on a section of the Mulholland Highway called “The Snake.” This section of highway is located in the Santa Monica Mountains of California. All four accidents were captured on video and they each involved a high-side fall of the motorcycle and rider. This article reports a technical description and analysis of these videos in which the motion of the motorcycles and riders is quantified. To aid in the analysis, the authors mapped Edwards Corner using both a Sokkia total station and a Faro laser scanner. This mapping data enabled analysis of the videos to determine the initial speed of the motorcycles, to identify where in the curve particular rider actions occurred, to quantify the motion of the motorcycles and riders, and to characterize the roadway radius and superelevation throughout the curve.

This paper presents a methodology for generating video realistic computer simulated rain, and the effect rain has on driver visibility. Rain was considered under three different rain rates, light, moderate and heavy, and in nighttime and daytime conditions. The techniques and methodologies presented in this publication rely on techniques of video tracking and projection mapping that have been previous published. Neale et al. [2004, 2016], showed how processes of video tracking can convert two-dimensional image data from video images into three-dimensional scaled computer-generated environments. Further, Neale et al. [2013,2016] demonstrated that video projection mapping, when combined with video tracking, enables the production of video realistic simulated environments, where videographic and photographic baseline footage is combined with three-dimensional computer geometry.

This paper examines a method for generating a scaled three-dimensional computer model of an accident scene from video footage. This method, which combines the previously published methods of video tracking and camera projection, includes automated mapping of physical evidence through rectification of each frame. Video Tracking is a photogrammetric technique for obtaining three-dimensional data from a scene using video and was described in a 2004 publication titled, “A Video Tracking Photogrammetry Technique to Survey Roadways for Accident Reconstruction” (SAE 2004-01-1221).