Figure 3.

Schematic depiction of the posturo-temporal filter model. The model consists of a body posture level followed by a body motion level. At the body posture level, the posture of a stimulus is analyzed by posture-selective neurons. Each posture-selective neuron represents a particular posture. The population represents all postures of the action. At the body motion level, activity within the posture-selective representation over the course of time is analyzed by posturo-temporal filters. By combining these filters and calculating the difference between the half-squared response of the forward-oriented posturo-temporal filter and the backward-oriented posturo-temporal filter, the model computes body motion energy. An estimate of body motion energy is determined at every point in time. Maximum pooling and summation over the stimulus duration provide an estimate of the motion direction. Similarly, an estimate of the facing direction is obtained by maximum pooling and summing the activities in the body posture representation.