Assessing the Accuracy of Vision-Based Accelerometry

Video-based motion measurement enables convenient tracking of multiple points in a plane with a single sensor: a CCD video camera. This paper presents a method to easily assess the accuracy of accelerations computed from the tracking of positions measured via video and quantifies these errors for a consumer-grade camera. A signal with time-dependent frequency and amplitude is applied to a shake table, and the motion of LED targets are tracked. The tracked target positions are filtered and twice-differentiated to compute accelerations. These vision-based acceleration measurements are time-synchronized with conventional accelerometer measurements. The vision-based and conventional acceleration measurements are directly compared, and through a series of experiments, an extensive investigation of the acceleration measurement sensitivity to signal frequency and amplitude is carried out.