Binocular vision technology adopted commodity identification algorithm for unmanned vending counter

Abstract

A binocular vision technology adopted commodity identification algorithm for an unmanned vending counter comprises the following steps: preparing data; building a training model, wherein the trainingmodel adopts a yolo v3 model based on a TensorFlow framework; verifying the above model. and determining whether a training result meets expectation; if the expectation is not met, adjusting parameters to perform optimization, and repeatedly executing training process to perform retraining; if the expectation is achieved, deploying the trained yolo v3 model to perform identification of a target commodity; and obtaining images of the target commodity to be identified by utilizing binocular cameras on the unmanned vending counter, and inputting the images to the trained yolo v3 model to identifythe target commodity out. Image acquisition is performed by the binocular cameras, and commodity occlusion or perspective conditions are reduced; target detection is performed by the yolo v3 model, the difficulty and the calculation amount of the identification algorithm are reduced; and the identification accuracy is improved by adding different scales for the target detection, and the good effect of identification on small target commodities is achieved.

Description

technical field [0001] The invention relates to the technical field of unmanned vending cabinets and image recognition, in particular to a product recognition algorithm using binocular vision technology for unmanned vending cabinets. Background technique [0002] At present, many smart vending or self-service settlement systems simply identify individual commodities. This method is very simple, but it also brings a lot of inconvenience to users. For example, users need to send commodities one by one to the camera for shooting. area, the wait time is too long. Multi-commodity identification will greatly facilitate users. Users do not need to send products to fixed areas for identification in sequence. They only need to take away the products and the system will automatically settle the payment. However, the current method of using a single camera (monocular) to take pictures and identify multiple commodities usually limits the size of the container where the commodities are ...

AI Technical Summary

Problems solved by technology

The disadvantage of this patent is that a single camera is used to take pictures of the product, and the product recognition accuracy is not high

[0004] Most of the existing unmanned vending containers are small containers. One camera can capture all the goods in the container, and it is not easy to cause perspective distortion. However, when the container is larger, one camera will make it easier to The captured image is severely distorted, which will bring great difficulties to subsequent product identification

To solve this problem, most of the existing technologies use cameras with a wider field of view, which can solve the problem of too large container scenes to a certain extent, but problems such as occlusion and perspective cannot be solved

[0005] In order to solve the problem that the monocular camera is easy to cause image distortion when taking pictures of the goods in the large container, this invention proposes a solution that uses the binocular camera to take pictures and jointly identify the goods

Images