Table 1 gathers all occurrences in our vocabulary which relate to a vehicle with a location reference. Consider, for example, the occurrence `to pass a location'. The row corresponding to this occurrence in Table 1 is subdivided into three separate triples of columns, one each for the three agent attributes A_speed, L_course, and L_distance. The first column of each triple represents the stipulated value which this attribute must take on as a `Pre-Condition', i. e. which must be observed prior to the onset of a tentative or acceptable association between the trajectory section at that halfframe time point and the conceptual description by the occurrence `pass location'. The last column of each triple indicates in an analogous manner the attribute value which must be observed when the trajectory section should no longer be associated with the occurrence, i. e. the `Post-Condition'. The center column of each triple, the so-called `Monotonicity-Condition' characterizes the manner in which the attribute value may change while an association between a trajectory section and the occurrence in question remains acceptable. A short line just indicates that the Monotonicity-Condition is irrelevant for the particular association.
Figure 27 gives the fuzzy membership functions which relate the estimated vehicle speed to one of the different discrete values for the attribute `A(gent)_speed'. Analogously, Figure 28 depicts the fuzzy membership function relating the angular difference between the vehicle's velocity vector and the vector linking the current vehicle position with the location in question. The transformation of estimated distance between the vehicle's current position and the location in question into the discrete values for the attribute L_distance is given in Figure 29. Note that the fuzzy relationship between the estimated distance and the attribute values may depend on the estimated speed.
Figure 27: Fuzzy membership functions of the attribute values null, very small, small, normal, fast, and very fast for A_speed as a function of the estimated vehicle speed.
Figure 28: Fuzzy membership functions of the attribute values approaching, passing, and receding as a function of the angular difference between the vehicle velocity vector and the vector joining the actual vehicle position with the location in question.
Figure 29: Fuzzy membership functions of the attribute values null, small, normal, and great for L_distance as a function of the coordinate difference between the actual vehicle position and the location in question. The membership functions of the attribute values small, normal, and great depend, moreover, on the speed v of the agent.
Figure 30: Finite state automat for recognizing motion verbs which are modelled just by a Pre-Condition and a Monotonicity-Condition.
A set of fuzzy recognition automata - one for each occurrence - evaluates Pre-, Monotonicity-, and Post-Condition at each halfframe time point in order to calculate a fuzzy memebership value (degree of certainty) for the association between the vehicle trajectory at that halfframe time point and the respective occurrence. An example for such a recognition automaton is illustrated in Figure 30. The resulting fuzzy membership value is used to visualize the degree of certainty with which the algorithm associates the conceptual description given by a particular occurrence with a vehicle trajectory at the halfframe time point in Figure 26 . An additional example can be found in Figure 31.
Figure: Each occurrence is associated with a color, projected into the image (a) and into the premise map (b). A strip in the color of the associated occurrence is overlaid to a representative frame from the image sequence currently under investigation, parallel to the trajectory to which it belongs. The width of the strip is proportional to , i.e. if the algorithm assigns the maximum certainty to the association between a particular trajectory segment and an occurrence descriptor, the corresponding strip will have the maximum width at the location of this trajectory segment.