CN112766034A - Intelligent monitoring system for workshop operation safety - Google Patents

Abstract

Translated fromChinese

本发明公开了一种面向车间作业安全的智能监控系统，其特征在于包括基于轻量化openpose的人体关键点检测模块、基于YOLOv3的安全帽检测模块、基于交并比的人与安全帽匹配模块以及基于关节与区域位置匹配的危险区域检测模块四部分。该系统包含的模块功能丰富，各模块由轻量化模型搭建，可满足可间的通用监控需求，为车间安全生产提供了保障。

The invention discloses an intelligent monitoring system oriented to workshop operation safety, which is characterized by comprising a human body key point detection module based on lightweight openpose, a safety helmet detection module based on YOLOv3, a human and safety helmet matching module based on cross-parallel ratio, and Four parts of the dangerous area detection module based on joint and area position matching. The modules included in the system are rich in functions, and each module is constructed by a lightweight model, which can meet the general monitoring requirements of the company and provide a guarantee for the safe production of the workshop.

Description

Intelligent monitoring system for workshop operation safety

Technical Field

The invention relates to an intelligent monitoring system for workshop operation safety, and belongs to the field of machine learning, deep learning, image processing and system integration.

Background

With the progress of automation development and intelligent transformation and upgrade of manufacturing industry, more and more automation devices and industrial intelligent robots participate in workshop production operation, gradually replace the work of part of workshop workers, and play an increasingly important role in modern production. Although the number of workshop workers is in fact relatively reduced, this does not mean that the robot can largely or even completely replace the human role in workshop operations in the short term.

At present, most types of automated robots can only perform work according to pre-programmed instructions, such as palletizing robots, welding robots, etc. The robots generally have the requirements or characteristics of large scale, monotonicity, mechanicity and repeatability, and can be used as personnel substitutes for high-strength operation or high-risk severe environment operation; some intelligent robots obtain certain sensing and judging capabilities through technologies such as integration of multi-source sensors and embedded intelligent algorithms, for example, the intelligent robots are used for intelligent sorting robots and cooperative assembly robots in automatic flexible production assembly lines, but the operation intelligentization and flexibility degrees of the robots are still far away from human beings, and high-quality work cannot be finished in some manufacturing occasions such as complex process flows and high requirements on refinement degrees.

The production assembly workshop of automobile products is one of the manufacturing workshops with the highest degree of automation in the current industrial field, the classical production mode of an automatic assembly line comes from the automobile manufacturing industry, and the conditions of standardization and modularization of accessories, advanced evolution of assembly process and the like greatly promote the automation level of the production assembly workshop. However, even if the assembly work still exceeds 20% of the assembly work, the robot still can not reach the quality level of manual work in a short period, and the work which can only be completed by the human (such as debugging of the assembly module and the like) is often the core process link in the production process. In a production workshop of initiating explosive devices, most of the process operations need to be directly contacted with materials such as gunpowder, so that complex electric equipment required by robot operation is a great potential safety hazard, and the initiating explosive device industry is mainly operated manually so far. Therefore, even though the current industrial intelligence level has developed to a considerable height, human beings still remain the highest-level intelligent unit bodies at present, and still dominate most of workshop production operations at the present stage, and especially in some product production lines or assembly lines with discrete process flows, complex process flows, high refinement degrees, high automation cost or high process requirements, manual operations cannot be replaced.

When the robot works within the capacity range, the production output with fixed quality can be realized even if the robot works continuously with high intensity. Compared with the prior art, human workers have physiological defects which cannot be avoided, so that the conditions of distraction, physical and mental fatigue, emotional fluctuation and the like are easy to occur after long-time work, and the working behaviors of the workers become various and uncontrollable. Although the worker is required to strictly comply with the established production behavior specification, the worker may not be able to operate according to the standard in the actual operation process due to the subjective or objective factors, which may cause the problem of non-specification of the operation behavior. The importance of the worker in the production operation is obvious whether the behavior is standard or not: on the one hand, the irregular operation behavior may cause the quality of the product to be reduced to a great extent, even unqualified; on the other hand, in some high-risk workshops, such as the above mentioned initiating explosive device workshops, the non-standardization of the action of the worker is also a potential safety hazard, and high-risk accidents such as blasting and the like may be caused.

To summarize, see: on one hand, in the core production process or procedure links of most industries, the effect of manual operation cannot be replaced; on the other hand, the inherent disadvantages of human beings may cause the body behaviors or the behavior flow to be irregular during the operation process, thereby causing quality or safety problems. In view of this, it can be concluded that: the research on the normative of the manual operation behaviors in the production workshop is necessary, and the research is one of the important research directions of the production management science. The safety helmet as an effective head protection tool is widely applied to production workshops, so that the safety helmet can monitor whether the safety helmet is worn or not and whether workers enter a dangerous area or not so as to improve the operation safety guarantee. In order to realize the purpose, open position is adopted to detect the skeleton of the human body, then the position of the safety helmet is detected by using a YOLOv3 algorithm, and the position of the human body is matched to judge whether the worker wears the safety helmet or not; and finally, judging whether the worker enters a dangerous area or not.

Disclosure of Invention

The invention aims to solve the technical problem that the existing workshop operation safety monitoring system aims at single action, and two operation specification scenes are considered.

In order to solve the technical problem, the technical scheme of the invention is to provide an intelligent monitoring system for workshop operation safety, which is characterized by comprising the following modules:

module 1): a human body key point detection module based on lightweight openposition;

module 2): a helmet detection module based on YOLOv 3;

module 3): a person and safety helmet matching module based on the intersection ratio;

module 4): and a person and danger area matching module based on joint and area positions.

Preferably, said module 1) replaces the 7 × 7 convolution kernel with three successive convolution kernels of size 1 × 1, 3 × 3, 7 × 7 and uses a hole convolution where the 3 × 3 convolution kernel is; the module 1) adopts Mobile net as a skeleton network; the module 1) employs a depth separable operation.

Preferably, the module 2) is implemented by using a YOLOv3 network; the YOLOv3 network related to the module 2) adopts multi-scale features for target prediction.

Preferably, the module 3) firstly detects the operator by the light open position, and secondly matches the detection result of the safety helmet with a boundary frame formed by key points of the head of the operator; the module 3) takes an intersection ratio (IOU) as an index for matching the operator with the safety helmet; the IOU referred to by the module 3) is given by the following formula:

preferably, the IOU threshold referred to by said module 3) is set to 0.2.

Preferably, the joint involved in the module 4) is a foot joint; the module 4) relates to a matching mode of judging whether the foot joints are in a dangerous area; the module 4) relates to matching criteria for warning as dangerous behavior when one or both of the foot joints are in a dangerous area.

Wearing safety helmets and keeping away from dangerous areas are two major keys of workshop operation safety. Compared with the prior art, the invention adopting the technical scheme has the following technical effects:

the invention simultaneously considers two working conditions of whether the safety helmet is worn or not and whether workers enter a dangerous area or not;

the light open position is adopted in the aspect of human skeleton recognition to enhance the real-time property;

in the detection of whether the safety helmet is worn or not, firstly, YOLOv3 is adopted to detect the light weight of the safety helmet, and then the intersection of a boundary frame formed by head key points detected by open position and a safety helmet area detected by YOLOv3 is compared as a basis for judging whether the safety helmet is worn or not;

in the monitoring of the danger zone, whether one or two of the foot joints are in the danger zone is taken as a lightweight matching criterion.

Drawings

Fig. 1 is a structural diagram of an intelligent monitoring system for workshop operation safety provided by the invention.

Detailed Description

In order to make the invention more comprehensible, preferred embodiments are described in detail below with reference to the accompanying drawings.

Examples

An intelligent monitoring system for workshop operation safety is shown in figure 1 and comprises the following modules:

module 1): a human body key point detection module based on lightweight openposition;

module 2): a helmet detection module based on YOLOv 3;

module 3): a person and safety helmet matching module based on the intersection ratio;

module 4): and a person and danger area matching module based on joint and area positions.

Wherein, module 1) uses three consecutive convolution kernels of size 1 x 1, 3 x 3, 7 x 7 instead of the 7 x 7 convolution kernel and uses a hole convolution at the place of the 3 x 3 convolution kernel; the module 1) adopts Mobile net as a skeleton network; module 1) employs a depth separable operation.

Module 2) is implemented by using a YOLOv3 network; the YOLOv3 network related to the module 2) adopts multi-scale features for target prediction.

The module 3) firstly detects the operator by the light open position, and secondly matches the detection result of the safety helmet with a boundary frame formed by key points of the head of the operator; the module 3) takes the intersection ratio (IOU) as an index matched with the safety helmet by an operator; the IOU referred to by module 3) is given by the following equation:

module 3) involves setting the IOU threshold to 0.2.

The joint related to the module 4) is a foot joint; the module 4) relates to a matching mode that whether the foot joints are in a dangerous area is judged; module 4) relates to matching criteria that both warn as dangerous behavior when one or both of the foot joints are within a dangerous area.

Claims (5)

Translated fromChinese

1.一种面向车间作业安全的智能监控系统，其特征在于，包括以下模块：1. an intelligent monitoring system for workshop operation safety, is characterized in that, comprises following module:模块1)：基于轻量化openpose的人体关键点检测模块；Module 1): Human body key point detection module based on lightweight openpose;模块2)：基于YOLOv3的安全帽检测模块；Module 2): helmet detection module based on YOLOv3;模块3)：基于交并比的人与安全帽匹配模块；Module 3): Person and helmet matching module based on cross-combination ratio;模块4)：基于关节与区域位置的人与危险区域匹配模块。Module 4): Matching module between people and dangerous areas based on joint and area positions.2.根据权利要求1所述的面向车间作业安全的智能监控系统，其特征在于，所述模块1)采用三个连续的1*1，3*3，7*7大小的卷积核代替7*7卷积核且在3*3卷积核的地方采用了空洞卷积；2. The intelligent monitoring system for workshop safety according to claim 1, wherein the module 1) adopts three consecutive convolution kernels of 1*1, 3*3, and 7*7 sizes to replace 7 *7 convolution kernel and the hole convolution is used in the place of 3*3 convolution kernel;所述模块1)采用Mobile net作为骨架网络；Described module 1) adopts Mobile net as skeleton network;所述模块1)采用深度可分离操作。The module 1) adopts depth separable operation.3.根据权利要求1所述的面向车间作业安全的智能监控系统，其特征在于，所述模块2)采用YOLOv3网络实现；3. the intelligent monitoring system for workshop safety according to claim 1, is characterized in that, described module 2) adopts YOLOv3 network to realize;所述模块2)涉及的YOLOv3网络采用多尺度特征进行目标预测。The YOLOv3 network involved in the module 2) uses multi-scale features for target prediction.4.根据权利要求1所述的面向车间作业安全的智能监控系统，其特征在于，4. The intelligent monitoring system for workshop operation safety according to claim 1, characterized in that,所述模块3)先由轻量化open pose对作业人员进行检测，其次将安全帽检测结果与作业人员头部关键点组成的边界框进行匹配；The module 3) firstly uses the lightweight open pose to detect the operator, and then matches the safety helmet detection result with the bounding box composed of the key points of the operator's head;所述模块3)将交并比(IOU)作为作业人员与安全帽匹配的指标；The module 3) takes the intersection-over-union ratio (IOU) as an indicator that the operator matches the helmet;所述模块3)涉及的IOU由以下公式给出：The IOU involved in the module 3) is given by the following formula:

所述模块3)涉及的IOU阈值设置为0.2。The IOU threshold involved in the module 3) is set to 0.2.5.根据权利要求1所述的面向车间作业安全的智能监控系统，其特征在于，5. The intelligent monitoring system for workshop operation safety according to claim 1, characterized in that,所述模块4)涉及的关节为脚部关节；The joints involved in the module 4) are foot joints;所述模块4)涉及的匹配方式为判断脚部关节是否在危险区域内；The matching method involved in the module 4) is to judge whether the foot joint is in the dangerous area;所述模块4)涉及的匹配标准为当一个或者两个脚步关节在危险区域内时，都当作危险行为进行报警。The matching standard involved in the module 4) is that when one or two foot joints are in the dangerous area, they are both regarded as dangerous behaviors to give an alarm.

Images