CN-115760952-B - Vision-based luggage attribute analysis method, analysis device and storage medium

CN115760952BCN 115760952 BCN115760952 BCN 115760952BCN-115760952-B

Abstract

The application discloses a vision-based luggage attribute analysis method, which comprises the steps of preparing a training data set for luggage attribute analysis, collecting a first image set containing luggage with various attributes, detecting each image in the first image set by utilizing a luggage type detection model, taking a luggage target area detection result as the training data set, preprocessing the training data set, carrying out interference elimination and data enhancement preprocessing on the luggage target area, and carrying out training based on a deep learning network by utilizing the preprocessed training data set so as to train a luggage attribute identification model for identifying luggage functions, colors and size attributes. The application also provides a luggage attribute analysis device and a computer readable storage medium.

Inventors

CHEN YAQIONG
LUO FUZHANG
ZHU GUANGQIANG
LAI SHIWU
WANG HEPING

Assignees

盛视科技股份有限公司

Dates

Publication Date: 20260508
Application Date: 20221109

Claims (8)

1. A vision-based luggage attribute analysis method, comprising: collecting a first image set containing various attribute baggage, detecting the type of the baggage and the target area of the baggage by using a baggage type detection model for each image in the first image set, and taking the detection result of the target area of the baggage as a training data set; preprocessing of training data sets, preprocessing of target areas of baggage for interference removal and data enhancement, and Training based on the deep learning network by utilizing the preprocessed training data set to train a luggage attribute recognition model for recognizing luggage functions, colors and size attributes, comprising: Sequentially performing interference elimination and data enhancement pretreatment on each luggage target area in the training data set to obtain a third image set; carrying out data enhancement pretreatment on each luggage target area in the data set to obtain a fourth image set; training a luggage color attribute recognition model, namely performing label marking of color attributes on a third image set, and inputting the third image set after label marking into a first deep learning network for training; training a luggage size attribute recognition model, namely performing label marking of size attribute on a fourth image set, and inputting the third image set marked by the label into a second deep learning network for training; Training a luggage functional attribute recognition model, namely performing label marking of functional color attributes on a fourth image set, and inputting the third image set after label marking into a third deep learning network for training; And training the luggage attribute recognition model, namely calculating the accuracy rates of the trained luggage color attribute recognition model, luggage size attribute recognition model and luggage function attribute recognition model, selecting the main branch of the attribute recognition model corresponding to the highest accuracy rate of one of the three attribute recognition models as the main branch of the fourth deep learning network, freezing the first n units of the main branch, carrying out the training of the other two attribute recognition again by using the rest units of the main branch, and outputting the luggage attribute recognition model with single input and three output.
2. The vision-based luggage attribute analysis method according to claim 1, wherein in preparation of the training data set for luggage attribute analysis, Collecting a second image set containing baggage with various attributes, dividing the second image set into training data, verification data and test data according to the proportion of 8:1:1, and marking the baggage types and the baggage areas on the training data in the second image set; Training a baggage type detection model by adding an attention mechanism module to the initial position of each unit module of a backbone network of Yolox models to construct the baggage type detection model; Baggage type and baggage object detection by performing baggage type detection and baggage object detection on each image in a training data set for baggage attribute analysis using a trained baggage type detection model, and obtaining a center point of a baggage object and a detection frame of an area about the baggage object in each image in the training data set for baggage attribute analysis, The first image set is cropped according to the central point of the luggage target and the detection frame to acquire a luggage target area.
3. The vision-based baggage attribute analysis method according to claim 2, wherein a center point coordinate of any baggage object is recorded as (x, y), and a height and a width of a detection frame of the baggage object are respectively H and W, wherein the step of clipping the first image set according to the center point of the baggage object and the detection frame to obtain the baggage object region is: Calculating the coordinates (x 1 ,y 1 ) of the upper left corner and the coordinates (x 2 ,y 2 ) of the lower right corner of the coordinate frame; cutting by taking the upper left corner point and the lower right corner point as two diagonal points to obtain a rectangular luggage target area; Wherein the method comprises the steps of 。
4. The vision-based baggage attribute analysis method of claim 1, wherein n has a value of 10.
5. The method for analyzing vision-based baggage attribute according to claim 1, wherein the types of baggage detected by the baggage type detection model include a rucksack, a backpack, a handbag, a waist bag, a trunk and a handbag, and wherein the batch interference-free preprocessing of the baggage object belonging to the same baggage type is performed according to the baggage type result obtained by the baggage type detection model, specifically: For a luggage target area of a rucksack type, darkening an upper part area of two thirds of the luggage target area in an X-axis direction; For a luggage target area of the backpack type, the upper part area of one third of the luggage target area is blacked out in the X-axis direction; For a handbag type luggage target area, the upper part area of one third of the luggage target area is darkened in the X-axis direction; for a luggage target area of the handbag, darkening an upper part area of one third of the luggage target area in an X-axis direction; for a luggage target area of the luggage type, a half of an upper part area of the luggage target area is darkened in an X-axis direction, and For a bag-type luggage target area, the luggage target area is equally divided into four areas along the Y-axis, and the first and fourth areas on the Y-axis are darkened.
6. The vision-based baggage attribute analysis method of claim 1 wherein the step of data enhancing the baggage object region comprises: unifying the size of each luggage target area in the training data set; Carrying out affine transformation on each luggage target area in the training data set; Performing horizontal overturn on each luggage target area in the training data set; Vectorizing each luggage target area in the training data set; And carrying out normalization processing on each channel of each luggage target area in the training data set.
7. A baggage attribute detection device, based on the vision-based baggage attribute analysis method according to claim 1, comprising: A preparation unit for collecting a first image set containing baggage with multiple attributes, detecting the type of baggage and the target area of the baggage for each image in the first image set by using a baggage type detection model, and taking the result of the detection of the target area of the baggage as a training data set; A preprocessing unit of training data set for performing interference-free and data-enhanced preprocessing on the luggage target area, and And the training unit is used for training the luggage attribute recognition model based on the deep learning network by utilizing the preprocessed training data set so as to train the luggage attribute recognition model for recognizing the luggage function, color and size attribute.
8. A computer readable storage medium, characterized in that the computer readable storage medium stores a computer program which, when executed by a processor, implements the vision-based baggage attribute analysis method of any one of claims 1-6.

Description

Vision-based luggage attribute analysis method, analysis device and storage medium Technical Field The present application relates to the field of image processing, and more particularly, to a vision-based baggage attribute analysis method, analysis apparatus, and storage medium. Background The personnel usually adopt traditional manual registration mode, carry out luggage information collection, register luggage information such as type, function and colour of luggage. The traditional manual method has high cost and low automation degree, and can not meet the related requirements of the development of the smart city. With the rapid development of economy, the consumption level of people is continuously improved, a large amount of baggage can be carried by travel, and the effect of the baggage attribute noninductive analysis technology is particularly important for realizing the informative management of the baggage. Disclosure of Invention Aiming at the prior art, the application solves the technical problem of providing a vision-based non-inductive luggage attribute analysis method, an analysis device and a storage medium which can realize automatic luggage attribute analysis and help to realize intelligent management of passenger luggage. In order to solve the above technical problems, the present application provides a vision-based luggage attribute analysis method, which includes: collecting a first image set containing various attribute baggage, detecting the type of the baggage and the target area of the baggage by using a baggage type detection model for each image in the first image set, and taking the detection result of the target area of the baggage as a training data set; preprocessing of training data sets, preprocessing of target areas of baggage for interference removal and data enhancement, and Training based on the deep learning network is performed by utilizing the preprocessed training data set so as to train a luggage attribute recognition model for recognizing luggage functions, colors and size attributes. In one possible implementation, in the preparation of the training data set for the analysis of the attributes of the baggage, Collecting a second image set containing baggage with various attributes, dividing the second image set into training data, verification data and test data according to the proportion of 8:1:1, and marking the baggage types and the baggage areas on the training data in the second image set; Training a baggage type detection model by adding an attention mechanism module to the initial position of each unit module of a backbone network of Yolox models to construct the baggage type detection model; Baggage type and baggage object detection by performing baggage type detection and baggage object detection on each image in a training data set for baggage attribute analysis using a trained baggage type detection model, and obtaining a center point of a baggage object and a detection frame of an area about the baggage object in each image in the training data set for baggage attribute analysis, The first image set is cropped according to the central point of the luggage target and the detection frame to acquire a luggage target area. In one possible implementation manner, the coordinates of the center point of any luggage object are recorded as (x, y), the height and the width of the detection frame of the luggage object are respectively H and W, and the step of clipping the first image set according to the center point of the luggage object and the detection frame to obtain the luggage object area is as follows: Calculating the coordinates (x 1,y1) of the upper left corner and the coordinates (x 2,y2) of the lower right corner of the coordinate frame; cutting by taking the upper left corner point and the lower right corner point as two diagonal points to obtain a rectangular luggage target area; Wherein the method comprises the steps of In one possible implementation, the step of performing deep learning network based training using the preprocessed training data set includes: Sequentially performing interference elimination and data enhancement pretreatment on each luggage target area in the training data set to obtain a third image set; carrying out data enhancement pretreatment on each luggage target area in the data set to obtain a fourth image set; training a luggage color attribute recognition model, namely performing label marking of color attributes on a third image set, and inputting the third image set after label marking into a first deep learning network for training; training a luggage size attribute recognition model, namely performing label marking of size attribute on a fourth image set, and inputting the third image set marked by the label into a second deep learning network for training; Training a luggage functional attribute recognition model, namely performing label marking of functional color attributes on a fourth image set, and inputting the