Technology
Area coverage is a critical factor in video surveillance and especially in video analytics. In many environments, the number of cameras deployed as a part of the system is high. Hence, a network of cameras deployed has many possible configurations depending on the various combinations of camera parameters. Depending on the placement and coverage, the cost of the system deployed may vary dramatically. Centralized planning of the camera network placement and configuration will result in considerable increase in efficiency of the deployed system in terms of coverage and quality of coverage. However in most scenarios the camera parameters are set manually, depending on the individual judgment of the personnel involved. The availability of a planning tool for the automatic set-up of the sensing infrastructure given the map of the environment to be covered, would allow optimizing the configuration of the camera network, minimizing both the number of sensors and the black spots.
Mobics Optimal Planning Tool, provides a user friendly solution to optimize the configuration of camera network in terms of coverage and cost, given a number of constraints. The constraints are usually financial and spatial. Some of them are user-dependent (maximum cost / number of cameras, minimum area coverage) while other are area-dependent (terrain morphology) or problem-dependent (minimum number of pixel per meter to detect an object).
Optimal Planning Tool requires a Digital Elevation Model (DEM) of the area of interest in GeoTIFF format. Given the DEM and user-specified constraints, a highly parallel algorithm is able to decide about the number of cameras as well as its optimal placement parameters (installation location, height, pan, tilt, zoom). The user is being informed about the progress of the algorithm in real-time. In the end of the processing, a comprehensive report can be downloaded in .PDF format.
Several flame and smoke features are checked by Meleagros including motion, color information, geometry attributes and temporal/spatial characteristics. The system operates on user defined areas/tiles that are obtained through scene characterization and image segmentation, thus resulting in texture and distance uniform areas. Meleagros offers day/night operation as luminance conditions are taken into account by selecting the appropriate algorithms and adjusting detection thresholds.
Meleagros implements a unique fusion scheme of different, heterogeneous sources for better assessment of the field observations, and for developing safer conclusions about the crisis and risk. The system is designed to incorporate a large set of fusion algorithms (e.g. Dempster-Shafer approximate reasoning - mathematical theory of evidence). The fusion scheme allows for the scaling of the mechanism and the effective implementation of various versions of the Meleagros system (large scale / prefectures, local authorities, private installations). The combination of data from multiple peripheral components allows the safer decision about potential outbreak of fire and eliminates false alarms.
Fire and smoke detection in wide forest areas is a challenging problem. Mobics Advanced Computer Vision Algorithm provides a solution using visible spectrum cameras. In such wide areas, the information about the detected event is not always enough. In order to handle a situation, fire-fighters need to know the exact location of the event. Mobics Advanced Camera Localization Solution, provides two different methods to tackle this problem and limit the searching area.
The first method uses a single-camera setup. In this method a transformation is used in order to map points from camera's image coordinate system to geographic coordinates (latitude, longitude). To achieve this, a number of reference points are defined both in image coordinate system (pixels) and geographic coordinate system (latitude, longitude). This method provides a rough approximation of the event's location since it depends a lot on terrain's morphology.
The second method uses a multi-camera setup. That method requires 3 geographic points (latitude, longitude) for each camera. These 3 points form a triangle which represents camera's Horizontal Field Of View. The first point is the actual camera's installation position and the other two points can be determined using Google maps in order to simulate camera's FOV and maximum effective range. Alternatively, those points can be estimated, if lens's focal length, sensor's height and camera's orientation parameters are known. Thereafter, the intersection of cameras can be determined, using triangulation, defining the area where the event has been detected in world coordinates.
Meleagros simplifies the remote configuration and management of a deployed system. The configuration tool enables management of multiple independent systems through an intuitive Web user interface. Furthermore, it allows remote monitoring of system status through heartbeat mechanism and self-testing operations.
Meleagros implements an image segmentation method for scene characterization. Image segmentation is the process of partitioning a digital image into multiple segments of common characteristics. The goal of segmentation is to simplify the representation of an image into something that is more meaningful and easier to analyze. Finer characterization of the scene (with almost homogeneous texture) regarding the fuel type or distance, speeds up the detection process while considerably reducing the false alarm rate. Meleagros offers an easy to use tool of predefined scene types (e.g. forest, sky, etc.) for image segmentation that is based on the watershed transformation method.
Meleagros solution offers a simulation engine that is based on the well known fire simulation software FARSITE, developed by the U.S. Forest Service. FARSITE uses spatial information on topography and fuel, as well as meteorological data (temperature, humidity, wind speed and direction). It requires a GIS (Geographic Information System) for processing of spatial information. Meleagros simulation engine includes capabilities of spreading models for ground and crown fire, spot fire burning behind the fire front and acceleration of fire in a two-dimensional spread model.
