Motion Estimation and Object Recovery with Time-Sequentially Sampled Imagery*

Abstract

Algorithms for estimating the motion of a moving object from digital imagery acquired in real time generally rely on two important assumptions: 1) the field of view is sampled frame-instantaneously, i.e. a record of the radiation integrated over a time interval that is short compared to the frame interval is stored at the focal plane, and this frozen image is read out serially during the interval between frames; and 2) there is relatively little change in the image between consecutive frames. With many digital imaging systems, the field of view is sampled time-sequentially during the entire interval between frames; and the radiation sensed at the focal plane changes continuously during this time. Typically, the field of view is sampled in a lexicographic (LEX) or 2:1 line-interlaced (2:1-LI) fashion. These patterns are illustrated in Table 1 which shows the order in which a 4 × 4 array of points would be sampled. With both these patterns, consecutive samples are clustered spatially. Table 1 also shows three dot-interlaced sampling patterns which disperse consecutive samples over the entire field of view. The 2:1 dot-interlaced (2:1-DI) and bit-reversed dot-interlaced (BRV-DI) patterns have been considered for bandwidth compression in video systems. The congruential dot-interlaced (CGR-DI) pattern has been shown to be optimal for sampling bandlimited signals [1],[2]. It also simplifies certain signal processing operations [3]. This particular pattern generates a 5 × 4 array of samples.