A furniture detection method and device based on deep convolutional neural network

Abstract

The invention discloses a furniture detection method based on a deep convolutional neural network, comprising the following steps: constructing a furniture detection network, the furniture detection network including a feature extractor composed of FPN and ResNet101, and a target detector composed of an SSD model ; Train the furniture detection network to determine the parameters of the furniture detection network to obtain a furniture detection model; use the furniture detection model to sequentially perform feature extraction and target detection on the furniture scene graph to be detected to obtain furniture and furniture categories. A deep convolutional neural network based furniture detection is also disclosed. The furniture detection method and device can quickly and accurately detect different types of furniture with large size differences.

Description

technical field [0001] The invention belongs to the technical field of architectural interior design, and in particular relates to a furniture detection method and device based on a deep convolutional neural network. Background technique [0002] With the hot market of home decoration, especially soft decoration, "what you see is what you get" has become the growing demand of every designer and customer. In the home improvement market, designers often provide customers with a rendering of a home improvement scene to show customers the details of the home improvement design. But from the rendering to the landing, clients often can only follow the arrangement of the design company. This "monopoly" model seriously limits the choice of customers, and also limits the ability of independent designers to implement the scene. [0003] Target detection is one of the traditional tasks of computer vision. Thanks to the features from manual extraction to deep learning features, target...

AI Technical Summary

Problems solved by technology

[0006] (3) The efficiency and accuracy of the existing target detection models are often incompatible. For example, the dual-model detection is Faster RCNN which has high accuracy and also brings a huge amount of calculation; the single-model detector YOLO has almost 30fps calculation speed, but the accuracy is not satisfactory

[0007] As a kind of deep convolutional neural network, the deep residual network ResNet mainly solves the problem that the network cannot be trained when the level is deep.

Images