Intermodal Mobility Assistance for Megacities - Smart Information Systems: Computational Intelligence for Real-Life Applications - page 359

Fig. 13.9 Same start and end point as in Figs. 13.7 and 13.8 (traffic incidents and surveillance
camera information enabled during routing)
which are currently blocked and tries to find the best path while circumventing the
blocked portions. However, one problem remains concerning the suggested route.
Since no surveillance video analysis is used as one of the information sources, the
IMA system is unaware of the sudden formation of a traffic congestion on the sug-
gested route.
The suggested path depicted in Fig. 13.9 does not suffer from this problem. This
map shows the suggested route by the IMA system, planned when considering both
traffic incidents and surveillance video information. This suggested route is certainly
longer than the one in Fig. 13.8 , but the commute will be faster compared to the
previous one, as the sudden traffic congestion is detected by the analysis of video
data gathered from the static traffic surveillance camera and taken into account for
planning.
13.5 Conclusions and Future Work
In this chapter, we presented the video analysis and route planning component of the
IMA system, which is a distributed system aims at providing effective transportation
in order to increase the quality of life in mega cities.
The issue of traffic jam discovery in the viewing range of a static traffic surveillance
camera was addressed with the video analysis component. A holistic video analysis
was performed by extracting particular global features from videos of 5 s length. The
features used in the current system are based on the dense optical flow algorithm
introduced by Farneback. This dense optical flow algorithm was selected due to its
traffic flow classification performance. Dense optical flow vectors provide a rough
overview of the traffic flow in the scene. These low-level features were processed
to construct meaningful motion descriptors. The traffic flow analysis model was
subsequently generated using the constructed motion descriptors. We evaluated our