A Visual Distance Approach for Multicamera Deployment With Coverage Optimization

This paper proposes a visual distance approach for a multicamera deployment through parallel coverage optimization over three-dimensional (3-D) space to inspect 3-D objects. The key idea is the proposal of a new “visual distance” between a single camera and a scene point, which takes into account multiple coverage criteria. Specifically, the effective visual sensing area of a single camera is expressed by a 4-D geometric object, which is composed of a 3-D visual frustum and an extra dimension for the acceptable view angle. An anisotropic projective transformation is applied to transform this 4-D geometric object into a 4-D unit cube, based on which a new visual coverage measure called “visual distance” is proposed using the infinity norm. It is shown that this new measure is capable of characterizing the coverage level of a task point in a quantitative way. Based on the visual distance, a coverage strength model and a new type of scene partition, called visual Voronoi partition, are established, and then a parallel gradient optimization approach is developed to achieve optimal coverage performance of a multicamera network. Comparative simulation and experimental results are provided to show the effectiveness and superior performance of the proposed approach.