In: Paparoditis N.. Pierrot-Deseilligny M.. Mallet C.. Tournaire O. (Eds). IAPRS. Vol. XXXVIII. Part ЗА - Saint-Mandé, France. September 1-3. 2010 
Figure 7. Detection results (left): black circles show the position 
of manually marked persons, white regions w'ere generated by 
the algorithm; tracking result (right): reference trajectories are 
black, automatically generated trajectories are white. 
from the optical flow which can occasionally lead to a w'rong 
displacement vector. Although the limitations of the reference 
data and the low' contrast of the image sequence coirupt the 
results to a certain amount, the numerical evaluation show's 
clearly the potential for further improvements of the detection 
and tracking system. A more sophisticated detection algorithm 
based on a machine learning approach will certainly reduce the 
effect of clutter and improve the results significantly. 
3.3 Interpretation of trajectories 
Experimental results for the computation of microscopic motion 
parameters in the trajectory analysis module are presented in 
this section. The computation is performed on manual reference 
data used in the image analysis module to focus at this point to 
the new approach of trajectory interpretation. Two example 
scenes of the image sequence are used for event detection, both 
with eight trajectories visualized on the last image of the 
sequence, respectively (Figure 8, Figure 9). 
Figure 8 show's a scene nearby the queue. Table 1 lists the 
related microscopic parameters of the eight trajectories. The 
microscopic parameters are meaningful with regard to the 
characterization of the trajectories. Trajectories 1 and 2 turn out 
to cover the longest distance and to have the smoothest and the 
least twisting path. This facts are clearly depicted by the 
parameters d rd and d ahs , which result in a high value for 
.9 ~ 0,98, as well as by the small deviation a z ~ 10°. Caused by 
the turn, trajectories 3 and 4 receive a smaller s ~ 0,75 and a 
higher deviation o 2 ~ 50°. The shorter trajectories 7 and 8 are 
characterized by small values for s as the paths are very 
twisting. Modeling the motion patterns of trajectories 1-4 result 
in a complex event called “wailing for other person”. because 
the persons 3 and 4 obviously accelerate their motion when 
persons 1 and 2 are next to them, afterwards walking alongside 
each other. Additionally to the shown microscopic parameters, 
evidence for parallelism can be given by mesoscopic 
parameters and, thus, result in the complex event similar 
described in Figure 4. Parallelism is visualized by the 
chronologically colored crosses of frames no. 1, 4, 7, 10, 13 and 
16 in Figure 8. 
Figure 9 shows another more crowded scene 2 representing a 
different event. Table 2 lists the microscopic parameters of 
eight trajectories. This scene is located at the right boundary of 
the queue, depicted by a macroscopic border of high density 
(red), next to a wall derived from GIS data (blue), cf. Figure 9. 
Resulting from the trajectory’s characteristics, microscopic 
motion parameters again receive values as expected. In this 
scene, the motion pattern shows a possibly dangerous 
“bottleneck” event, because the faster walking persons are 
pushed aside by the queue. These faster persons have to 
sidestep to a small gap between the queue and the wall. 
Figure 8. Results of the trajectory analysis module of scene 1: 
high density borders depicted in red; the color bar shows the 
time steps within the sequence. 
ID 
dabs 
(m) 
drel 
(m) 
V 
(km/h) 
s 
*mean 
(°) 
0- 
n 
1 
1 1,22 
10,91 
5.24 
0,972 
13.27 
10,95 
2 
11,17 
10.94 
5,25 
0,980 
13,95 
9,97 
3 
7.05 
5,05 
2,42 
0.715 
51,30 
57.31 
4 
6,55 
5,05 
2.42 
0,771 
42.88 
44,67 
5 
2,50 
1.60 
0,76 
0.641 
29,66 
92,03 
6 
2.14 
1,52 
0,72 
0.710 
67.99 
50,63 
7 
1,71 
0,78 
0,37 
0,454 
56,72 
35,34 
8 
1.67 
0.43 
0,21 
0.257 
31,47 
84,53 
Table 1. Microscopic parameters for trajectories in Figure 8 
Results of the trajectory analysis module of scene 1. 
Figure 9. Results of the trajectory analysis module of scene 2: 
high density border depicted in red, GIS data depicted in blue; 
the color bar shows the time steps within the sequence. 
ID 
dabs 
(m) 
drel 
(m) 
V 
(km/h) 
.9 
<>mcun 
(°) 
0. 
(°) 
1 
8.73 
8,17 
3,92 
0.936 
72,99 
18,31 
2 
8,27 
7,78 
3,74 
0,941 
81,11 
20,75 
3 
3,29 
2,86 
1,38 
0,871 
82,03 
60,34 
4 
9,77 
9,72 
4,66 
0.994 
72,19 
3.84 
5 
2,31 
1,15 
0.55 
0,500 
67,68 
85,22 
6 
1,78 
0,56 
0,27 
0,315 
34,84 
112,24 
7 
2,01 
1,10 
0,52 
0,549 
89.24 
86,78 
8 
1,66 
0.54 
0,26 
0,324 
36,12 
102,24 
Table 2. Microscopic parameters for trajectories in Figure 9.