Automatic Layout Of Furniture Based On Reinforcement Learning

With the development of economy,standard of living raised,consumption idea changed and the real estate industry ushered in the golden age of development,which has become increasingly important in the national economy status.Therefore,the need for household design is constant and growing.However,the complexity and time-consuming of household design cannot meet the demand for quantity and quality in the market today.Automation in house decoration become imminent.Decorators need to copy and mark two-dimensional house type manually in the system to generate three-dimensional model,then complete the layout of the furniture by multiple adjustments according to their specialized knowledge,which cannot meet the demand for diversify and personalization.Using computers to realize automatic interior design can save a lot of human and non-human resources and provide more personalized services for more users.However,the efficiency and accuracy of existing floor plan automatic recognition algorithms are very low.Interior design needs to meet the requirements of functionality,aesthetics,and comfort while conforming to ergonomics.Existing automatic furniture layout algorithms cannot be flexibly applied to various complex apartment structures,and cannot be really implemented.In response to the above problems,this paper takes the home improvement industry as the background and realizes the full automation from the floor plan to the three-dimensional layout plan.The main contributions of the paper are as follows:· Based on digital image processing technology,automatic recognition of the floor plan is realized,each component in the floor plan is accurately identified,a three-dimensional model is constructed,and the preprocessing of the floor plan is completed.Experiments prove that the recognition accuracy and efficiency of the algorithm have great advantages.· Using the evaluation criteria of neural network learning designers,a layout plan automatic evaluation scorer is designed,which can automatically realize the evaluation of layout plans,and at the same time provide an automated environmental feedback model for reinforcement learning to guide action decision-making.· An automatic layout algorithm based on strategy iteration is proposed to model each type of furniture separately and generate layout plans sequentially.Not only the applicability,aesthetics,and comfort are considered,but also the door and window information provided by the space outline can be taken into consideration,thereby generating a reason-able lighting plan.The experimental results show the rationality and diversity of layout schemes,and prove the practicability and time-saving of the work.· A furniture automatic layout algorithm based on imitation learning is proposed to directly learn furniture state-action pairs,namely feature-position pairs.This algorithm has solved the problem of sample efficiency and directly learn the existing strategies.The experimental results show the rationality and diversity of layout schemes,and prove the practicability and time-saving of the work.