Nowadays surveillance systems are installed to fight criminals. Locations with high surveillance camera densities make people feel unsafe. The initial idea of this project is to use information from surveillance cameras to objectively enhance security by detecting all-day accidents thus helping non-criminals. The project is situated inside the Forschungsinitiative Sicherheit (see university and project website) where interdisciplinary research around the topic security and safety is performed.
Object shapes are extracted in front of a static background which is learned and updated. In the next step, the object appearance is modeled using an object-wise bag-of-features-approach (BoF) in combination with different types of keypoint descriptors like SIFT, SURF or GLOH.
The BoF-Approach allows to identify individual objects even in merge-situations with partial occlusions. Besides, generalizations for the purpose of classifications are easy to train when the BoF is generalization-wide e.g., a "human people BoF".
As an application of the estimated methods, the BMBF-funded project ASEV has been launched.
In the following sequences, object tracking is performed only with the usage of keypoint features. So the object position which is not suitable in many tracking situations was not used.