SUNAR Surveillance Network Augmented by Retrieval - Advanced Concepts for Intelligent Vision Systems - page 163

Table 1. Multiple camera tracking results - MOTA

MOTA Brno KuDir
Test Set Average -1.183 -1.400
Track Averaged Mean -2.052 -2.072
Track Averaged Median -1.770 -1.517

Table 2. The primary to Secondary Camera subject Re-Acquisition (SCRA) metric
table

Sec. RA - GT Sec. RA - Brno Sec. RA - KD
12345

12345 12345

1 9 1 0

2 8 2 0

3 7 0 0

4 1 0 0

5 9 0 0

camera (scspt) runs are incomparable to multiple camera runs. In table 1, only
MCSPT results are depicted.
The table 2 also shows that using standard precision/recall metrics, our results
are slightly better than other results [20]. Moreover, using the multiple camera
(summarized binar) metric - the (primary to) Secondary Camera subject Re-
Acquisition (SCRA, [20]) shows that SUNAR slightly outperformed the other
systems in absolute numbers, which may be seen in table 2. The CPSPT task
results were similar to the table above, but we have been the only participants
there. The illustration of the task is in figure 1. In both figures 1 and 4 (an
illustration of a MCSPT tracking trial), the bounding boxes and trajectories are
of five colors. Blue means non- occluding reference (ground truth), yellow an
occluding reference. The Green box and trajectory shows a correct detection,
red represents a missed detection and the orange color is for false alarms.

6 Conclusions

This paper presents a state of the art SUNAR surveillance system based on
visual information retrieval in theory and praxis (using free software). In contrast
to other approaches, we try to collect and index as much information as we
can acquaint and manage it eciently to avoid a continuous human CCTV
monitoring and analysis of massive and low quality recordings in case of an
accident.
The FIT, Brno University of Technology has taken part in many projects and
evaluations concerning the public safety and visual surveillance, however the
AVSS 2009 Multi-Camera Tracking Challenge [20] was the first public evaluation
campaign concerning object tracking in a wide area under surveillance containing
both camera setups - overlapping and non-overlapping field of views.
Although we are convinced the system works really good under certain cir-
cumstances and it outperformed the others especially in the Secondary Camera