Abstract: Motion of actors within a scene may be detected based on imaging data, using machine learning tools operating on cameras that captured the imaging data. The machine learning tools process images to perform a number of tasks, including detecting heads of actors, and sets of pixels corresponding to the actors, before constructing line segments from the heads of the actors to floor surfaces on which the actors stand or walk. The line segments are aligned along lines extending from locations of heads within an image to a vanishing point of a camera that captured the image. Trajectories of actors and visual data are transferred from the cameras to a central server, which links trajectories captured by multiple cameras and locates detected actors throughout the scene, even when the actors are not detected within a field of view of at least one camera.

Abstract: Motion of actors within a scene may be detected based on imaging data, using machine learning tools operating on cameras that captured the imaging data. The machine learning tools process images to perform a number of tasks, including detecting heads of actors, and sets of pixels corresponding to the actors, before constructing line segments from the heads of the actors to floor surfaces on which the actors stand or walk. The line segments are aligned along lines extending from locations of heads within an image to a vanishing point of a camera that captured the image. Trajectories of actors and visual data are transferred from the cameras to a central server, which links trajectories captured by multiple cameras and locates detected actors throughout the scene, even when the actors are not detected within a field of view of at least one camera.

Abstract: The invention concerns the detection of vehicles in images of a night time scene. In particular, but not limited to, the invention concerns a traffic surveillance system that is used to detect and track vehicles to determine information about the detected and tracked vehicles. Candidate pair of headlights are identified 900 in an image based on luminance of points in the image. These candidates are then verified 902 by identifying 400i a sub-image of the image sized to include a candidate vehicle having the pair of candidate headlights; and determining whether the candidate vehicle is a vehicle represented in the image by testing 400k the sub-image for the presence of predetermined features of a vehicle other than the headlights. Aspects of the invention include a method, software and computer hardware.

Abstract: The invention concerns a traffic surveillance system that is used to detect and track vehicles in video taken of a road from a low mounted camera. The inventors have discovered that even in heavily occluded scenes, due to traffic density or the angle of low mounted cameras capturing the images, at least one horizontal edge of the windshield is least likely to be occluded for each individual vehicle in the image. Thus, it is an advantage of the invention that the direct detection of a windshield on its own can be used to detect a vehicle in a single image. Multiple models are projected (206) onto an image with reference to different points in the image. The probability of each point forming part of a windshield is determined based on a correlation of the horizontal edges in the image with the horizontal edges of the windshield model referenced at that point (220). This probability of neighboring points is used to possible detect a vehicle in the image (224).

Abstract: The invention concerns a traffic surveillance system that is used to detect and track vehicles in video taken of a road from a low mounted camera. The inventors have discovered that even in heavily occluded scenes, due to traffic density or the angle of low mounted cameras capturing the images, at least one horizontal edge of the windshield is least likely to be occluded for each individual vehicle in the image. Thus, it is an advantage of the invention that the direct detection of a windshield on its own can be used to detect a vehicle in a single image. Multiple models are projected (206) onto an image with reference to different points in the image. The probability of each point forming part of a windshield is determined based on a correlation of the horizontal edges in the image with the horizontal edges of the windshield model referenced at that point (220). This probability of neighbouring points is used to possible detect a vehicle in the image (224).