Method for identifying advertising machine frame by using synthetic training picture

Abstract

The invention discloses a method for identifying an advertising machine frame by using a synthetic training picture, which comprises the following steps: establishing a 3D model of a material by using Blender, importing the 3D model into Unity for rendering and picture generation, and generating annotation information of the picture while generating the picture; identifying the advertising machine frame in the picture through a deep learning algorithm, extracting an advertising picture played in the advertising machine frame, and removing deformation of the advertising picture through projection transformation. According to the invention, the deep learning model is trained by adopting batch synthetic pictures, manual annotation is not needed, a large amount of time and labor cost are saved, and the generated data annotation can be ensured to have no error by using the synthetic pictures; the advertisement player frame in the picture is detected by using a deep learning model, and the advertisement picture is taken out and corrected, so that interference caused by environment and shooting deformation is eliminated, and the accuracy of advertisement checking is improved; the error between the images is greatly reduced, and the accuracy of advertisement checking is greatly improved.

Description

technical field [0001] The invention belongs to the field of applying image processing technology to information dissemination, and in particular relates to a method for identifying the frame of an advertising machine by using synthesized training pictures. Background technique [0002] In recent years, deep learning technology has achieved great results in many fields, such as: image processing, natural language processing, speech recognition, machine translation and so on. The reason why deep learning has developed by leaps and bounds in recent years is due to several reasons: First, the generation of massive data in the Internet era has provided enough training data for the neural network; Widely used; of course, there are many algorithmic improvements to avoid overfitting and gradient disappearance. In the field of image processing, the training of the model requires a large number of labeled pictures. For example, the standard data set MNIST for the introduction of cla...

AI Technical Summary

Problems solved by technology

[0003] The main problems in the existing technology include: most image detection and recognition technologies belong to supervised learning, and there must be enough labeled training data to train a good model

These data sets are sufficient for theoretical research in most scenarios, but they are powerless when applied in practice, because the needs of each company are different, and the required data are also very different. It is impossible to pass public datasets to meet all needs; on the other hand, many datasets have legal risks, for example, some face datasets are no longer available due to infringement issues

In addition, during the advertising inspection, the pictures captured in the actual scene are affected by the environment (lighting, background, etc.) and the shooting angles are different (different shooting angles will cause different deformations in the advertising screen), so they can be used directly There will be a large error in the comparison of the algorithm

Images