A general motion model and spatio-temporal filters for 3-D motion interpretations

Real-Time Quantized Optical Flow

“…Temporal occlusion occurs when an object occludes another object in some frame in the past, but disoccludes it in the present due to motion (but not the converse, as described below). The e ect of temporal occlusion can be seen in Figure 17 ( 32]). In Frame T a slow-moving background is being disoccluded by a faster-moving foreground.…”

A Real-time Computer Vision Platform for Mobile Robot Applications

“…* It can be seen that some algorithms (e.g., Fleet & Jepson [13]) may be very accurate but very slow while other algorithms (e.g., Camus [6]) may be very fast but not very accurate. The algorithm of Liu et al [15,16], on the other hand, is very flexible since it can be very accurate for some parameter settings while very fast for other settings.…”

“…Five successive image frames are smoothed and subsampled to 64 ϫ 64 pixels. Image derivatives up to third order are then obtained by applying separable 3-D Hermite polynomial differentiation filters to the neighborhood of each pixel [15,16]. This produces an overdetermined linear system which is then solved using a least squared error (LSE) method.…”

“…The algorithm, developed by Liu et al [15,16], uses up to third-order spatio-temporal derivatives and a generalized motion model that accommodates expansion and translation. In this approach, the spatial and temporal derivatives are fit to a single coherent motion model (i.e., corresponding to a single object), which leads to a linear system of multiple motion constraint equations.…”