Event Title
Implementation of a Camera Orientation Optimization for Mobile Industrial Plants
Location
Clemson, SC
Start Date
16-10-2020 10:45 AM
End Date
16-10-2020 11:10 AM
Presentation Type
Presentation
Description
In the last few years, occupational safety plays an increasingly important role in almost all industrial sectors. In this context the deployment of wide-area visual sensor networks has become practical. A proper placement of visual sensors in the target environment is an important design problem as it has a direct impact on both the cost and the performance of the network. This thesis addresses the camera placement problem for high hazard mobile industrial plants and proposes a solution that maximizes the total coverage of a camera network while satisfying boundary conditions. Camera systems become very complex and difficult to survey with increasing size. For this reason a virtual reality user interface is used in this work and allows the network design in an intuitive manner. The associated workflow is simple and can be used for either the effective planning of a video system or for the realignment of existing systems in dynamic environments. The user has the ability to modify the scene by adding cameras or moving existing objects within the visualization and see the effects on the coverage rate immediately. This makes an illustration of a mobile plant possible. Moreover the scene can be divided into regions, which are weighted according to their importance of observation.
The user gets the natural ability to see what each camera can see and is able to discover dead zones. The effects of opening angle, visibility and mounting height thus can be made understandable. The placement problem is formulated as a combinatorial optimization problem. The number of decision variables increases with the number of feasible camera locations. Particularly for large infrastructures the computational complexity of the problem is a major challenge. To reduce the amount of possible combinations camera orientations are first compared to each other and only the best are used for the optimization. For solving the placement problem a genetic algorithm (GA) is used. The proposed algorithm has been optimized for speed to return good results while still allowing interactive modification of parameters. Characteristics of surveillance targets and cameras, as well as occlusion are considered during optimization.
Implementation of a Camera Orientation Optimization for Mobile Industrial Plants
Clemson, SC
In the last few years, occupational safety plays an increasingly important role in almost all industrial sectors. In this context the deployment of wide-area visual sensor networks has become practical. A proper placement of visual sensors in the target environment is an important design problem as it has a direct impact on both the cost and the performance of the network. This thesis addresses the camera placement problem for high hazard mobile industrial plants and proposes a solution that maximizes the total coverage of a camera network while satisfying boundary conditions. Camera systems become very complex and difficult to survey with increasing size. For this reason a virtual reality user interface is used in this work and allows the network design in an intuitive manner. The associated workflow is simple and can be used for either the effective planning of a video system or for the realignment of existing systems in dynamic environments. The user has the ability to modify the scene by adding cameras or moving existing objects within the visualization and see the effects on the coverage rate immediately. This makes an illustration of a mobile plant possible. Moreover the scene can be divided into regions, which are weighted according to their importance of observation.
The user gets the natural ability to see what each camera can see and is able to discover dead zones. The effects of opening angle, visibility and mounting height thus can be made understandable. The placement problem is formulated as a combinatorial optimization problem. The number of decision variables increases with the number of feasible camera locations. Particularly for large infrastructures the computational complexity of the problem is a major challenge. To reduce the amount of possible combinations camera orientations are first compared to each other and only the best are used for the optimization. For solving the placement problem a genetic algorithm (GA) is used. The proposed algorithm has been optimized for speed to return good results while still allowing interactive modification of parameters. Characteristics of surveillance targets and cameras, as well as occlusion are considered during optimization.