Assessing the influence of variables on vehicle speed determination through reverse projection analysis

Abstract

Reverse projection photogrammetry has long been established as a forensic tool for estimating height and, more recently, vehicle speed from video. Previous work has demonstrated its feasibility under controlled daytime conditions using digital video recorders (DVRs). This study extends that foundation by systematically evaluating how additional variables specifically headlight bloom as a marker, and frame rate variability in network video recorders (NVRs) and cloud-based consumer systems might affect the accuracy of speed determination. Vehicle speeds were independently documented using Light Detection and Ranging (LIDAR) and in-dash speedometer recordings, providing complementary ground truth: LIDAR offered precise point-in-time measures, while the speedometer delivered a continuous average across the test interval. Together, these methods established a baseline for comparison against photogrammetric calculations.

Structured experiments were conducted at a closed test track, with multiple DVR, NVR, and cloud devices recording daytime and nighttime runs at varying speeds. Speed was estimated by measuring vehicle position (or headlight bloom at night) via reverse projection photogrammetry and calculating elapsed time using frame counts or, when necessary, frame analysis tools. Results showed that higher and more stable frame rates reduced uncertainty, while consumer cloud systems and some NVRs introduced dropped or inconsistent frames, producing errors outside the expected margin. Conversely, headlight bloom proved to be a reliable nighttime surrogate for vehicle position.

These findings demonstrate that reverse projection photogrammetry provides a reliable means of estimating vehicle speed when recorder timing is properly accounted for and understood. By quantifying sources of uncertainty across diverse recording systems and lighting conditions, this research strengthens the evidentiary basis for using video-derived speed estimates in forensic analysis and supports their defensible application in casework.