CN-121564803-B - Production behavior detection method and system

CN121564803BCN 121564803 BCN121564803 BCN 121564803BCN-121564803-B

Abstract

The application relates to a production behavior detection method and a system, and relates to the field of production detection, wherein the method comprises the steps of collecting station image information; the method comprises the steps of analyzing and determining a production action frame according to station image information, marking and extracting action characteristics in the production action frame, judging whether the action characteristics are consistent with qualified characteristics in a preset historical qualified action database, marking a product as a processed product if the action characteristics are consistent with the qualified characteristics in the preset historical qualified action database, collecting corresponding product images and giving a prompt if the action characteristics are inconsistent with the qualified characteristics, determining product quality information according to the product images, determining whether the product quality information is qualified based on the product quality information and preset quality parameters, marking the product as a processed product if the product quality information is qualified, marking the product as a product to be processed if the product quality information is unqualified, and rechecking the processed product through magnetic field detection to finish action detection. The application has the effect of ensuring the production quality and efficiency.

Inventors

WANG YINRUI
YANG MINGLUN
CHENG YUAN

Assignees

辛米尔视觉科技(上海)有限公司

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (8)

1. A method for detecting production behavior, comprising: Collecting station image information; Analyzing and determining a production action frame according to the station image information; Labeling in the production action frame and extracting to obtain action characteristics; Judging whether the action characteristics are consistent with the qualified characteristics in a preset historical qualified action database; if the products are consistent, marking the products as processed products; if the images are inconsistent, corresponding product images are collected and a prompt is sent out; Determining product quality information according to the product image, and determining whether the product is qualified or not based on the product quality information and a preset quality parameter; if the product quality information is qualified, marking the product as a processed product; if the product quality information is unqualified, marking the product as a product to be processed; Rechecking the processed product through magnetic field detection to finish motion detection; Labeling and extracting the production action frame to obtain action characteristics comprises the following steps: marking in the production action frame to determine action mark points; Determining the limb shape according to the production action frame and a preset clothing image; determining key action points according to the action mark points and the limb shape marks in the production action frame; forming an action track in series according to the marked key action points; determining the action characteristics of limb swing according to the action track and the limb shape; determining the limb shape according to the production action frame and the preset clothing image comprises the following steps: selecting an area matched with the preset clothing features from the image of the production action frame, and extracting to obtain a clothing image in the area; Extracting the whole contour lines of the clothes according to the clothes image, and determining the clothes contour; extracting a clothing curve of the clothing from the clothing outline, and recording the bending degree and the fitting degree of the curve; determining a degree of bulk from the degree of curvature and the garment profile; and determining the shape of the limb according to the fitting degree and the fluffiness degree.
2. The method of claim 1, wherein marking the determined action mark point in the production action frame comprises: extracting contour features of all targets in a single frame according to the production action frame; Distinguishing human body candidate areas and non-human body areas according to the contour features; Determining a shape integrity rate and a limb movement trend according to the human body candidate region and a preset human body shape; dividing the human body candidate region with the shape integrity rate smaller than a preset shape judgment rate into the non-human body region; determining worker limb characteristics according to the contour characteristics and the limb movement trend of the re-divided human body candidate region; And determining action mark points according to the limb characteristics of the worker.
3. The method of claim 2, wherein determining the action marker points based on the worker limb characteristics comprises: determining a static portion and a dynamic portion of a limb based on the worker limb characteristics; Extracting a dynamic shape according to the dynamic part; determining joint points from the static portion and the dynamic portion; Determining a joint track according to the dynamic shape and the joint fixed point; determining a joint movement point according to the joint track; determining a static fixed point according to the static part and a preset human body shape; and taking the static fixed point, the joint fixed point and the joint moving point as action marking points.
4. The method of claim 1, wherein determining whether the motion characteristics are consistent with the qualifying characteristics in the pre-defined historical qualifying motion database comprises: Determining a track shape according to the action track; determining an action angle according to the track shape; determining motion displacement according to the motion characteristics and a preset human body shape; Determining an action amplitude according to the action angle and the action displacement; determining the action coincidence degree based on the qualified amplitude of qualified features in a preset historical qualified action database and the action amplitude; When the overlap ratio is larger than the preset judging overlap ratio, the action characteristics are consistent with the qualified characteristics in the preset historical qualified action database; When the overlap ratio is not greater than the preset judging overlap ratio, the action characteristics are inconsistent with the qualified characteristics in the preset historical qualified action database.
5. The method of claim 1, wherein the rechecking by magnetic field detection comprises: collecting overall magnetic field signal data of a product part, and marking a magnetic field abnormal region of signal fluctuation; Determining abnormal magnetic field points according to the change fit of the signal data of the abnormal magnetic field region; determining the actual mounting position of the part based on the magnetic field abnormal point and a preset theoretical mounting point; performing difference calculation on the actual installation coordinates of the parts and a preset theoretical installation point to obtain position deviation; When the position deviation amount is larger than a preset position tolerance threshold value, prompting by a preset theoretical installation point; and when the position deviation amount is not larger than a preset position tolerance threshold value, finishing rechecking.
6. The method of claim 1, further comprising, when determining a production action frame from the station image information analysis: Collecting part magnetic field characteristics of a part to be selected; comparing the part magnetic field characteristics with reference part characteristics in a preset standard part magnetic field database, and determining whether the magnetic field strengths of the part magnetic field characteristics and the reference part characteristics are matched; If so, determining part magnetic field distribution based on the part magnetic field characteristics; performing difference value calculation on the part magnetic field distribution and the preset required part magnetic field distribution to obtain a magnetic field deviation value; if the magnetic field deviation value is not smaller than a preset deviation threshold value, the selected part is inaccurate, and part selection prompt is carried out based on the part magnetic field distribution; and if the magnetic field deviation value is smaller than a preset deviation threshold value, determining that the selected part is accurate.
7. The method of claim 1, wherein determining product quality information from the product image comprises: selecting part features required to be processed by a corresponding station from the product image by using a preset part image frame, and defining the position of the frame for selecting the part features as part position points; Determining a part depth of field value, a part shape and a part hole on the part position point; determining the tightness of the part according to the depth of field value of the part; determining a part deviation value according to the part shape and the part hole; determining the flatness of the part according to the shape of the part and the shape of a preset product; And taking the tightness, the part deviation value and the part flatness as product quality information.
8. A production behavior detection system, comprising: the acquisition module is used for acquiring station image information and product images; A memory for storing a program for implementing a production behavior detection method according to any one of claims 1 to 7; and the processor loads and executes the programs in the memory.

Description

Production behavior detection method and system Technical Field The present invention relates to the field of production detection, and in particular, to a method and a system for detecting production behaviors. Background The production behavior detection is a technical means for judging whether the product is qualified or not and the behavior is compliant by analyzing the operation behaviors and the workflow of the staff. When a worker in a factory inserts or installs semi-finished products on a production line, a production behavior detection technology is often applied to ensure the efficiency and quality of production activities, and in the prior art, a manager generally monitors the working condition of the worker on site to judge whether the production behavior is qualified or not, so that the quality of the products is ensured. Because the field inspection relies on human eye judgment, the number of workers to be supervised by a manager is large, the supervision difficulty is large, when the complexity of an actual product is high or errors caused by manual processing are relatively fine, the problems are difficult to find, the misconnection and misinsertion phenomenon of parts is frequent, the yield is reduced, and further the reworking of the product is caused, so that the quality and the efficiency are seriously affected. Disclosure of Invention In order to ensure the quality and efficiency of production, the invention provides a production behavior detection method and system. In a first aspect, the present invention provides a method for detecting production behavior, which adopts the following technical scheme: a method for detecting production behavior, comprising: Collecting station image information; Analyzing and determining a production action frame according to the station image information; Labeling in the production action frame and extracting to obtain action characteristics; Judging whether the action characteristics are consistent with the qualified characteristics in a preset historical qualified action database; if the products are consistent, marking the products as processed products; if the images are inconsistent, corresponding product images are collected and a prompt is sent out; Determining product quality information according to the product image, and determining whether the product is qualified or not based on the product quality information and a preset quality parameter; if the product quality information is qualified, marking the product as a processed product; if the product quality information is unqualified, marking the product as a product to be processed; And rechecking the processed product through magnetic field detection to finish motion detection. By adopting the technical scheme, the production action frame is determined by collecting the station image information, the fixed-point action characteristics are extracted, whether the processed product is a processed product without reworking is judged through action, and finally, the processed product is subjected to rechecking by magnetic field detection to complete the action detection flow, so that the real-time monitoring of production behaviors and the accurate judgment of product quality are realized, the controllability of the production process and the stability of the product quality are improved, and the production quality and the production efficiency are ensured. Optionally, labeling and extracting the motion features in the production motion frame includes: marking in the production action frame to determine action mark points; Determining the limb shape according to the production action frame and a preset clothing image; determining key action points according to the action mark points and the limb shape marks in the production action frame; forming an action track in series according to the marked key action points; and determining the action characteristics of the swing of the limb according to the action track and the limb shape. By adopting the technical scheme, the action track is obtained through the point position and shape mark linkage, the action characteristics in the production process are more accurately captured and analyzed, and an accurate basis is provided for subsequent action comparison and quality judgment. Optionally, marking the production action frame to determine an action mark point includes: extracting contour features of all targets in a single frame according to the production action frame; Distinguishing human body candidate areas and non-human body areas according to the contour features; Determining a shape integrity rate and a limb movement trend according to the human body candidate region and a preset human body shape; dividing the human body candidate region with the shape integrity rate smaller than a preset shape judgment rate into the non-human body region; determining worker limb characteristics according to the contour characteristics and the limb movement tr