The area searched for stop signs is the one on the right side of the road (the green rectangle in the right side of every image in figure above). First, the color primitive is applied to detect red blobs in this area, which are candidates for stop signs. Since other red objects can appear in this region. the color test alone is not enough for detecting the stop signs, being used just as a ``focus of attention'' mechanism to further limit the image area that is analyzed.
Once a red blob is detected, the static feature primitive is applied to determine whether any of the filter responses in that area (red rectangle) matches the previously stored response for a stop sign. If the error (difference) is below some predetermined threshold, a stop sign is reported.
The looming detection routine applies two instances of the dynamic feature primitive (for two equal shifts of opposite signs) on consecutive frames in log-polar coordinates. The figure above illustrates the main steps and some intermediate results for the case when the leading car is approaching (expanding from t-1 to t).
By taking the difference of the two maps (for shift values of +s and -s), the sensitivity to speeds in the region where the distributions for the two shifts overlap is reduced. This is visible in the figure, where features from the building in the background are present in the two preliminary maps, but cancel each other in the final map, which contains only the features corresponding to the car.