Dinner plate motion state recognition method based on deep learning

Abstract

The invention discloses a dinner plate motion state recognition method based on deep learning. The motion state of a dinner plate is determined through a convolutional neural network. The method comprises the following steps: training a dinner plate position detection model by utilizing a convolutional neural network; detecting the dinner plate according to the key frame to obtain position information of the dinner plate; and judging the dinner plate position information of the front frame and the rear frame in the key frames, and determining the motion state of the dinner plate. According tothe method, the motion state of the dinner plate can be determined through the camera, the dinner plate needing to be detected does not need to be determined through an additional sensor when the dinner plate is detected, and cost can be reduced while the detection speed is not affected.

Description

technical field [0001] The invention relates to the field of computer vision, in particular to a deep learning-based recognition method for the motion state of a dinner plate. Background technique [0002] With the rapid development of artificial intelligence, more and more fields have applied artificial intelligence in real life. Recently, more and more attention has been paid to smart restaurants, and how to conveniently and quickly checkout payment has become a research hotspot. The existing intelligent payment methods based on deep learning generally use some sensors such as infrared and ultrasonic to judge the movement state of the dinner plate. Result checkout billing. However, the appeal method will increase the cost because of the use of sensors. At the same time, it will bring great inconvenience to the later maintenance and update. Secondly, the dependence of the sensor on the environment will directly affect the subsequent recognition. [0003] Secondly, there ...

AI Technical Summary

Problems solved by technology

However, the appeal method will increase the cost because of the use of sensors. At the same time, it will bring great inconvenience to later maintenance and updates. Secondly, the dependence of sensors on the environment will directly affect subsequent recognition.

[0003] Secondly, there are some problems in the existing methods for judging the motion state of objects based on traditional image processing, such as the inter-frame difference method. There is no suitable range to choose. If the interval is too large or too small, it will affect its detection accuracy. The second most important thing is that it cannot detect stationary objects.

The background difference method is difficult to obtain a unified background in the actual restaurant. Therefore, it is necessary to continuously update the background dynamically, which will greatly reduce the speed of detection, and at the same time, some subtle jitters will interfere with its detection results.

Edge detection method, it is more sensitive to color, so the color of the plate and dishes will have a great impact on it, and at the same time, changes in lighting and shooting angles will interfere with it

[0004] Generally speaking, the reliability of the traditional image processing method for detecting the motion state of objects is relatively poor, and there are problems such as illumination, image shadow and noise, object occlusion and slow detection speed, etc., and are easily affected by the environment, resulting in inaccurate detection results And misjudgment, in practical applications, it is difficult to determine the motion state of the plate

Images