This paper presents a vision-based method for fire detection from fixed surveillance smart cameras. The method integrates several well-known techniques properly adapted to cope with the challenges related to the actual deployment of the vision system. Concretely, background subtraction is performed with a context-based learning mechanism so as to attain higher accuracy and robustness. The computational cost of a frequency analysis of potential fire regions is reduced by means of focusing its operation with an attentive mechanism. For fast discrimination between fire regions and fire-coloured moving objects, a new colour-based model of fire's appearance and a new wavelet-based model of fire's frequency signature are proposed. To reduce the false alarm rate due to the presence of fire-coloured moving objects, the category and behaviour of each moving object is taken into account in the decision-making. To estimate the expected object's size in the image plane and to generate geo-referenced alarms, the camera-world mapping is approximated with a GPS-based calibration process. Experimental results demonstrate the ability of the proposed method to detect fires with an average success rate of 93.1% at a processing rate of 10 Hz, which is often sufficient for real-life applications.

This paper presents a system for obstacle detection in railway level crossings from 3D point clouds acquired with tilting 2D laser scanners. Although large obstacles in railway level crossings are detectable with current solutions, the detection of small obstacles remains an open problem. By relying on a tilting laser scanner, the proposed system is able to acquire highly dense and accurate point clouds, enabling the detection of small obstacles, like rocks laying near the rail. During an offline training phase, the system learns a background model of the level crossing from a set of point clouds. Then, online, obstacles are detected as occupied space contrasting with the background model. To reduce the need for manual on-site calibration, the system automatically estimates the pose of the level crossing and railway with respect to the laser scanner. Experimental results show the ability of the system to successfully perform on a set of 41 point clouds acquired in an operational one-lane level crossing.

This paper studies the use of eye tracking in a First-Person Shooter (FPS) game as a mechanism to: (1) control the attention of the player’s avatar according to the attention deployed by the player; and (2) guide the gameplay and game’s procedural content generation, accordingly. This results in a more natural use of eye tracking in comparison to a use in which the eye tracker directly substitutes control input devices, such as gamepads. The study was conducted on a custom endless runner FPS, Zombie Runner, using an affordable eye tracker. Evaluation sessions showed that the proposed use of eye tracking provides a more challenging and immersive experience to the player, when compared to its absence. However, a strong correlation between eye tracker calibration problems and player’s overall experience was found. This means that eye tracking technology still needs to evolve but also means that once technology gets mature enough players are expected to benefit greatly from the inclusion of eye tracking in their gaming experience.

This paper proposes a model for water detection in video sequences, which is a key asset of any robot operating in natural environments. By searching the visual input for the waters typically chaotic dynamic texture, the model is able to filter out the static background and even any dynamic object present in the scene. In this work, the waters signature is defined, mostly, in terms of an entropy measure computed from the optical flow obtained across several frames. To foster the classification of motionless regions in the visual input, usually associated to the far field, a segmentation guided label propagation method is used. The model is experimentally validated on 12 diverse videos, acquired from static and moving cameras.

Air pollution is an important source of morbidity and mortality. It is essential to understand to what levels of air pollution citizens are exposed, especially in urban areas. Low-cost sensors are an easy-to-use option to obtain real-time air quality (AQ) data, provided that they go through specific quality control procedures. This paper evaluates the reliability of the ExpoLIS system. This system is composed of sensor nodes installed in buses, and a Health Optimal Routing Service App to inform the commuters about their exposure, dose, and the transport’s emissions. A sensor node, including a particulate matter (PM) sensor (Alphasense OPC-N3), was evaluated in laboratory conditions and at an AQ monitoring station. In laboratory conditions (approximately constant temperature and humidity conditions), the PM sensor obtained excellent correlations (R²≈1) against the reference equipment. At the monitoring station, the OPC-N3 showed considerable data dispersion. After several corrections based on the k-Köhler theory and Multiple Regression Analysis, the deviation was reduced and the correlation with the reference improved. Finally, the ExpoLIS system was installed, leading to the production of AQ maps with high spatial and temporal resolution, and to the demonstration of the Health Optimal Routing Service App as a valuable tool.