Hartmut Loos

A Review and Comparison of Measures for Automatic Video Surveillance Systems

Todays video surveillance systems are increasingly equipped with video content analysis for a great variety of applications. However, reliability and robustness of video content analysis algorithms remain an issue. They have to be measured against ground truth data in order to quantify the performance and advancements of new algorithms. Therefore, a variety of measures have been proposed in the literature, but there has neither been a systematic overview nor an evaluation of measures for specific video analysis tasks yet. This paper provides a systematic review of measures and compares their effectiveness for specific aspects, such as segmentation, tracking, and event detection. Focus is drawn on details like normalization issues, robustness, and representativeness. A software framework is introduced for continuously evaluating and documenting the performance of video surveillance systems. Based on many years of experience, a new set of representative measures is proposed as a fundamental part of an evaluation framework.

Performance evaluation of a real time video surveillance system

This paper presents a thorough introduction to the real time video surveillance system which has been developed at Bosch Corporate Research considering robustness as the major design goal. A robust surveillance system should especially aim for a low number of false positives since surveillance guards might get distracted by too many alarms caused by, e.g., moving trees, rain, small camera motion, or varying illumination conditions. Since a missed security related event could cause a serious threat for an installation site, the before mentioned criterion is obviously not sufficient for designing a robust system and thus a low number of false negatives should simultaneously be achieved. Due to the fact that the false negative rate should ideally be equal to zero, the surveillance system should be able to cope with varying illumination conditions, low contrast and occlusion situations. Besides presenting the building blocks of our video surveillance system, the measures taken to achieve robustness is illustrated in this paper. Since our system is based on algorithms for video motion detection, which has been described e.g. in M. Mayer et al., (1996), the previous set of algorithms had to be extended to feature a complete video content analysis system. This transition from simple motion detection to video content analysis is also discussed in the following. In order to measure the performance of our system, quality measures calculated for various PETS sequences is presented.

Multi-cue Based Visual Tracking in Clutter Scenes with Occlusions

Object tracking is important for video analysis applications. However, tracking through occlusions is a difficult task due to significant appearance changes of the objects. Approaches based on either global features or one kind of local features can not solve the problem completely. In this paper, a multi-cue based tracking approach is introduced. It combines a corner tracking with a color and a shape model to resolve the object tracking problem through occlusions for most scenes (indoor and outdoor).To obtain an objective evaluation of the proposed method, a set of detection and tracking measures are used to perform a quantitative analysis based on a large sequence dataset with ground-truth annotation. The experimental results show that the proposed approach works robustly under varying conditions.