This paper considers the problem of automatically learning an activity-based semantic scene model from a stream of video data. A scene model is proposed that labels regions according to an identifiable activity in each region, such as entry/exit zones, junctions, paths, and stop zones. We present several unsupervised methods that learn these scene elements and present results that show the efficiency of our approach. Finally, we describe how the models can be used to support the interpretation of moving objects in a visual surveillance environment.

In this paper a novel evaluation framework for measuring the performance of real-time action recognition methods is presented. The evaluation framework will extend the time-based event detection metric to model multiple distinct action classes. The proposed metric provides more accurate indications of the performance of action recognition algorithms for games and other similar applications since it takes into consideration restrictions related to time and consecutive repetitions. Furthermore, a new dataset, G3D for real-time action recognition in gaming containing synchronised video, depth and skeleton data is provided. Our results indicate the need of an advanced metric especially designed for games and other similar real-time applications.

The paper investigates the unsupervised learning of a model of activity for a multi-camera surveillance network that can be created from a large set of observations. This enables the learning algorithm to establish links between camera views associated with an activity. The learning algorithm operates in a correspondence-free manner, exploiting the statistical consistency of the observation data. The derived model is used to automatically determine the topography of a network of cameras and to provide a means for tracking targets across the "blind" areas of the network. A theoretical justification and experimental validation of the methods are provided.

Human activity recognition has attracted the attention of researchers around the world. This is an interesting problem that can be addressed in different ways. Many approaches have been presented during the last years. These applications present solutions to recognize different kinds of activities such as if the person is walking, running, jumping, jogging, or falling, among others. Amongst all these activities, fall detection has special importance because it is a common dangerous event for people of all ages with a more negative impact on the elderly population. Usually, these applications use sensors to detect sudden changes in the movement of the person. These kinds of sensors can be embedded in smartphones, necklaces, or smart wristbands to make them “wearable” devices. The main inconvenience is that these devices have to be placed on the subjects' bodies. This might be uncomfortable and is not always feasible because this type of sensor must be monitored constantly, and can not be used in open spaces with unknown people. In this way, fall detection from video camera images presents some advantages over the wearable sensor-based approaches. This paper presents a vision-based approach to fall detection and activity recognition. The main contribution of the proposed method is to detect falls only by using images from a standard video-camera without the need to use environmental sensors. It carries out the detection using human skeleton estimation for features extraction. The use of human skeleton detection opens the possibility for detecting not only falls but also different kind of activities for several subjects in the same scene. So this approach can be used in real environments, where a large number of people may be present at the same time. The method is evaluated with the UP-FALL public dataset and surpasses the performance of other fall detection and activities recognition systems that use that dataset.