Camera Network Deployment Optimization And Application For Visual Tasks

With the development of camera networks in daily life as well as industrial applications,visual task optimization based on camera networks has become a popular research field.As a kind of non-contact field sensor,camera is characterized by small size,low power consumption,reasonable cost,and large amount of captured information,which can accomplish many complex tasks by simulating biological vision.How to control the camera network to optimize the visual task performance has become an important research topic.Aiming at the research topic of visual task optimization,this thesis introduces a visual task optimization framework to complete various visual tasks based on camera networks by studying the mathematical modeling of task performance.This thesis first solves the problem of information fusion between multiple cameras,which enables the camera network to use the captured information more efficiently in the coverage task.Then,in landmark-based visual localization task,the deployment of landmarks is optimized so that the camera can capture more landmark information to improve localization performance.This thesis has also conducted a detailed performance modeling and optimization approach of the stereo camera sensor,and has proposed a simultaneous mapping and coverage(SMAC)method for target objects with unknown map.Extensive simulations and experiments are designed to verify the effectiveness of the proposed methods.The main contributions of the thesis are as follows.(1)The monocular camera network is used to cover the target with known map,the concept of radial coverage strength is proposed to enhance the coverage performance of camera networks for target objects,and the relative position and orientation relationships between target object and cameras are preserved intactly using the vector form in an innovative way.A standard fused coverage strength algorithm is proposed,which fully considered and restored the characteristics of image fusion techniques.A fast fused coverage strength algorithm is given to reduce the computational effort while maintaining the fusion characteristics.A cost function to measure the coverage performance is constructed and optimized using a monotonic genetic algorithm.The image fusion experiment verify that the fused coverage strength algorithm can simulate the characteristics of image fusion technology.The simulation and experiment of the camera network deployment verify that the proposed algorithm can improve the coverage performance for a given number of cameras.(2)A method to optimize the deployment of landmarks to improve the performance of visual localization is proposed for a motion camera.First,the multi-coverage cap is established based on visual criteria,then the concept of multi-coverage probability is proposed.Then the concept of multiple coverage probability is proposed,which describes the probability that a camera can capture a specified number of landmarks at a certain position.Based on this criterion,a cost function is constructed and optimized using elimination genetic algorithm,and the optimized position and orientation of the landmark deployment are obtained.In addition,a method for estimating the orientation probability density of camera is proposed based on Monte Carlo method.Simulations and experiments validate that the proposed landmark deployment optimization method enables the camera to capture more landmark information,which improves the performance of visual localization task effectively.(3)A coverage model for stereo camera sensor with different configurations is built,as well as six performance criteria to evaluate the coverage performance of the stereo camera.The new depth resolution criterion is used to describe the performance of the stereo camera sensor to capture the depth of the target,and the criterion called permissible degree of orientation is used to describe the ability of the camera t o capture the front face of triangle piece.Other criteria are also introduced into the proposed stereo camera model,including visual resolution,field of view,focus,and occlusion.A cost function for the performance of a stereo camera is optimized using a gradient estimation based algorithm.Simulations verify the feasibility of the coverage model and the performance optimization method.(4)A scheme for simultaneous mapping and coverage of targets with unknown map using stereo camera network is proposed,which reduces the motion consumption of the camera network and saves the task time.The concept of virtual coverage strength for stereo camera network is proposed,and the gradient-estimation-based algorithm is used to optimize the coverage performance when the target object is incompletely mappe d.Simulations verify that the proposed method can optimize the coverage performance of the target by the stereo camera network while mapping the target object,and the experiments verify that the proposed method can be applied to real scenarios.