Computing feature motion without feature detectors: a model for terminator motion without end-stopped cells

Pointlike object features such as line-endings, have a privileged position in the computation of the veridical direction of object motion. Experiments confirm that the human visual system relies heavily on such features if they are present. It has been proposed that units such as end-stopped cells might be necessary for the computation of feature motion instead of the simple cells used in plaid motion models. Conventional plaid motion models have not been applied to feature motion. We present here a model, based on ordinary simple cells, using two parallel pathways (Fourier and non-Fourier) for the computation of the direction of two dimensional motion. Although similar in structure to popular models of plaid motion, our model includes a novel scheme for contrast normalisation and incorporates spatial pooling at the level of MT cells. The model predictions are consistent with psychophysical results for plaids. Furthermore, it computes directions within 5° of the physical motion of line-endings. It is shown that the non-Fourier signal is necessary for the computation of veridical motion.