Research On Automatic Layout Of Indoor Furniture Based On Deed Learning

With the continuous development of society,the living standards of the people are improving,and people’s living quality is also increasing.The decoration industry has ushered in a golden period of rapid development.The market size of the decoration industry is expected to reach 3.7817 trillion in 2025.Many enterprises have join in this field and all want to carve up this four trillion market.However,the process of decoration design is relatively complex,and there are still a large number of technical problems in the field of decoration,such as high communication costs between designers and customers,insufficient user friendliness of decoration Internet products,high learning costs,inefficient design and other problems.The design process consists of two parts.Firstly,the designer manually processes the data information on the floor and successively inputs it into the CAD software for drawing;Then build the three-dimensional model of furniture and room,constantly adjust the layout according to the design needs of users,and provide a small number of design versions for users to choose.The output efficiency ratio is extremely low,which seriously reduces the user experience.In view of the advantages of deep learning in image processing,more and more researchers apply deep learning algorithm to the field of decoration design,and the concept of "AI + decoration" came into being.At present,there are still some deficiencies in the relevant smart decoration design products on the market,such as the low accuracy of automatic identification of plan,the operation is complicated,the effect of automatic furniture layout is poor,the design is not flexible,etc.In response to the above problems,based on deep learning,this paper designs indoor furniture layout methods for different plan types,and realizes the automatic generation from two-dimensional plan to three-dimensional furniture layout.The main innovations are as follows:1.A stylized method of plan based on YOLO v5 object detection is proposed.As the preprocessing stage of plan recognition,the plan with different styles is transformed into a unified plan.The experimental results show that the speed of processing a picture is 0.2s,and the accuracy rate is 99.2%,which reflects the universality and efficiency of the stylized technology.2.A key point-based plan recognition method is proposed.By matching the two ends of the door,window,and wall,plus detecting other furniture on the plan and detecting the house name through OCR,a picture is converted into a series of vector data to form a threedimensional model,and an evaluation function is designed to verify the integrity of the plan recognition.After experimental statistics,the average processing time of a plan is 7.6s,and the comprehensive accuracy rate is 97.6%,which proves that its speed and accuracy are better than traditional algorithms.3.An automatic furniture layout algorithm based on convolutional neural networks is proposed,and the placement position of furniture is confirmed step by step through iterative thinking.The lightweight model structure is adopted to optimize the reasoning speed and make it easy to use.The experiment proves that the KL distance between the generated furniture layout plan and the 3D-FRONT dataset is 0.0037 on average,which proves the diversity and reliability of its layout algorithm.Based on these advantages,this paper realizes a set of automated process,which is composed of three stages: converting plan into unified plan,converting stylized plan into structured data and automatic layout of indoor furniture.This process can help designers to quickly draw plan and improve users’ fluency.