CN-121973379-A - Rolling equipment control method and device based on multi-target detection

CN121973379ACN 121973379 ACN121973379 ACN 121973379ACN-121973379-A

Abstract

The invention provides a rolling equipment control method and device based on multi-target detection, which relate to the technical field of target detection and comprise the steps of acquiring image data acquired by camera equipment aiming at rolling equipment; the method comprises the steps of carrying out multi-target detection on image data through a pre-trained target detection model to obtain a target detection result corresponding to the image data, wherein the target detection result comprises one or more of a living body of an operator, a hand area of the operator and a target edge contour corresponding to a glove area worn by the operator, determining a current state instance corresponding to the target detection result, wherein the current state instance represents a physical monitoring area in which the target edge contour contained in the target detection result falls, and determining a target control strategy according to the equipment state and the current state instance of the rolling equipment so as to execute a control action corresponding to the target control strategy on the rolling equipment. The invention can effectively prevent and intervene in real time in high risk condition in the operation of rolling equipment.

Inventors

XUAN YU
LIU YANFENG
HU JIANGHONG
CAO BIN
XIAO JIAN

Assignees

菲特（天津）检测技术有限公司
菲特(天津)智能科技发展有限公司

Dates

Publication Date: 20260505
Application Date: 20260407

Claims (9)

1. The utility model provides a rolling equipment control method based on multi-target detection which is characterized in that at least two camera equipment are arranged at the appointed position of the area where the rolling equipment is located, and a physical monitoring area is defined in advance in the vision area of each camera equipment, and the method comprises the following steps: acquiring image data acquired by the image pickup device aiming at the rolling device; Performing multi-target detection on the image data through a pre-trained target detection model to obtain a target detection result corresponding to the image data, wherein the target detection result comprises one or more of a target edge contour corresponding to a living body of an operator, a target edge contour corresponding to a hand region of the operator and a target edge contour corresponding to a glove region worn by the operator; determining a current state instance corresponding to the target detection result, wherein the current state instance represents the physical monitoring area in which a target edge contour included in the target detection result falls; Determining a target control strategy according to the equipment state of the rolling equipment and the current state instance so as to execute a control action corresponding to the target control strategy on the rolling equipment; the target detection model adopts a YOLOv framework-based neural network model, and comprises a main network, a feature fusion network and a multi-target detection head, wherein the multi-target detection is carried out on the image data through a pre-trained target detection model so as to obtain a target detection result corresponding to the image data, and the method comprises the following steps: Extracting basic features corresponding to the image data through the backbone network; Performing feature fusion on the basic features through the feature fusion network to obtain a multi-size fusion feature map matched with the multi-target detection head, and performing background feature filtering on the multi-size fusion feature map to obtain a multi-size target feature map; Performing target detection on the multi-size target feature map based on the associated static living body anchor frame, dynamic hand anchor frame and dynamic glove anchor frame through the multi-target detection head so as to obtain an initial detection result corresponding to the image data; and calling OpenPose a model to extract key points corresponding to the image data, and correcting the initial detection result by using the key points to obtain a target detection result corresponding to the image data.
2. The multi-target detection-based rolling equipment control method according to claim 1, wherein the multi-target detection heads include a biopsy head, a hand detection head, and a glove detection head; The living body detection head is associated with a static living body anchor frame; the hand detection head comprises a plurality of hand sub-detection heads with different sizes, and at least one dynamic hand anchor frame is associated with the plurality of hand sub-detection heads with different sizes; the glove detection heads comprise a plurality of glove detection heads with different sizes, and at least one dynamic glove anchor frame is associated with the glove detection heads with different sizes.
3. The method according to claim 2, wherein the performing, by the multi-target detection head, target detection on the multi-size target feature map based on its associated static living body anchor frame, dynamic hand anchor frame, and dynamic glove anchor frame to obtain an initial detection result corresponding to the image data, comprises: Performing target detection on the multi-size target feature image data based on the associated static living body anchor frame through the living body detection head so as to obtain an initial edge contour corresponding to the living body of the operator; performing target detection on the multi-size target feature map based on the respectively associated dynamic hand anchor frames through a plurality of hand sub-detection heads contained in the hand detection heads so as to obtain initial edge contours corresponding to hand areas of operators; And performing target detection on the multi-size target feature map based on the dynamic glove anchor frames respectively associated with the glove detection heads through a plurality of glove detection heads included in the glove detection heads so as to obtain initial edge contours corresponding to glove areas of operators.
4. The method for controlling a rolling device based on multi-target detection according to claim 3, wherein the step of performing target detection on the multi-size target feature map based on the respective associated dynamic hand anchor frames by using a plurality of hand sub-detection heads included in the hand detection heads to obtain an initial edge profile corresponding to a hand area of an operator comprises: Extracting features of the multi-size target feature map to obtain an interested region corresponding to the hand region of the operator and a size range of the interested region; Performing the following operation by each of the hand sub-detection heads included in the hand detection head: determining the cross-over ratio between the dynamic hand anchor frames associated with the hand detection head and the region of interest corresponding to the hand region of the operator; judging whether the region of interest corresponding to the hand region of the operator is matched with the dynamic hand anchor frame or not according to the intersection ratio and the intersection ratio threshold corresponding to the size range; And under the condition of matching, obtaining an initial edge contour corresponding to the hand area of the operator.
5. The multi-target detection-based rolling equipment control method according to claim 1, wherein invoking OpenPose a model to extract key points corresponding to the image data to correct the initial detection result using the key points comprises: judging whether the initial detection result comprises an initial edge contour corresponding to a hand region of an operator or an initial edge contour corresponding to a glove region of the operator; If so, invoking OpenPose a model, taking the central coordinates of the initial edge contour corresponding to the operator hand region or the initial edge contour corresponding to the operator glove region as a reference, extracting key points from the arm sub-region output by the target detection model, and correcting the initial edge contour corresponding to the operator hand region or the initial edge contour corresponding to the operator glove region by using the key points.
6. The multi-target detection-based rolling equipment control method according to claim 1, further comprising: Acquiring marking image data corresponding to the rolling equipment, wherein the marking image data is marked with an operator living body, an operator hand area and a glove area worn by an operator; extracting sub-images corresponding to the hand region of the operator and the glove region worn by the operator from the marked image data: for the sub-images corresponding to the same type of region, randomly determining a plurality of target sizes from the sizes corresponding to the sub-images corresponding to the type of region; clustering is carried out according to the sub-images corresponding to the target sizes and the sub-images corresponding to other sizes so as to obtain a plurality of target clustering clusters; and determining the matching degree between each size contained in the target cluster and the size of the multi-target detection head, so as to use the sub-image corresponding to each size contained in the target cluster as a dynamic anchor frame associated with the multi-target detection head based on the matching degree, wherein the dynamic anchor frame is the dynamic hand anchor frame or the dynamic glove anchor frame.
7. The method according to claim 6, wherein clustering the sub-image corresponding to the target size with the sub-images corresponding to other sizes to obtain a plurality of target clusters, comprises: Based on the clustering parameters, euclidean distance and cross ratio distance between the sub-image corresponding to the target size and the sub-images corresponding to other sizes, clustering is carried out to obtain a plurality of clustering clusters; Determining a weighted distance variance value corresponding to each cluster; And under the condition that the weighted distance variance value is larger than a preset threshold, carrying out self-adaptive adjustment on the clustering parameter, and carrying out re-partition on the clustering clusters based on the self-adaptive adjusted clustering parameter, the Euclidean distance and the cross-merging ratio distance between the sub-image corresponding to the target size and the sub-images corresponding to other sizes until the preset clustering stop condition is met, so as to obtain a plurality of target clustering clusters.
8. The multi-target detection-based rolling equipment control method according to claim 1, wherein determining a target control strategy from an equipment state of the rolling equipment and the current state instance comprises: Judging whether the current state instance is matched with an early warning/alarming state instance associated with the equipment state of the rolling equipment; if yes, the control strategy corresponding to the matched early warning/alarming state instance is used as the target control strategy corresponding to the rolling equipment.
9. The utility model provides a roll-in equipment controlling means based on multi-target detection which characterized in that, the assigned position department of roll-in equipment place regional has disposed two piece at least camera equipment, every camera equipment's FOV has all previously demarcated physical monitoring area, the device includes: The image acquisition module is used for acquiring image data acquired by the image pickup equipment aiming at the rolling equipment; the multi-target detection module is used for carrying out multi-target detection on the image data through a pre-trained target detection model so as to obtain a target detection result corresponding to the image data, wherein the target detection result comprises one or more of a target edge contour corresponding to an operator living body, a target edge contour corresponding to an operator hand area and a target edge contour corresponding to a glove area worn by the operator; The state instance determining module is used for determining a current state instance corresponding to the target detection result, and the current state instance represents the physical monitoring area in which the target edge contour contained in the target detection result falls; The control strategy determining module is used for determining a target control strategy according to the equipment state of the rolling equipment and the current state instance so as to execute a control action corresponding to the target control strategy on the rolling equipment; The target detection model adopts a YOLOv framework-based neural network model, and comprises a main network, a feature fusion network and a multi-target detection head, wherein the multi-target detection module is specifically used for: Extracting basic features corresponding to the image data through the backbone network; Performing feature fusion on the basic features through the feature fusion network to obtain a multi-size fusion feature map matched with the multi-target detection head, and performing background feature filtering on the multi-size fusion feature map to obtain a multi-size target feature map; Performing target detection on the multi-size target feature map based on the associated static living body anchor frame, dynamic hand anchor frame and dynamic glove anchor frame through the multi-target detection head so as to obtain an initial detection result corresponding to the image data; and calling OpenPose a model to extract key points corresponding to the image data, and correcting the initial detection result by using the key points to obtain a target detection result corresponding to the image data.

Description

Rolling equipment control method and device based on multi-target detection Technical Field The invention relates to the technical field of target detection, in particular to a rolling equipment control method and device based on multi-target detection. Background The rolling equipment is used as key forming equipment in the field of processing high polymer materials such as rubber, plastics and the like, and the working principle of the rolling equipment is that one or more pairs of rollers rotating oppositely apply pressure to materials to enable the materials to be stretched into sheets with required thickness and width. In industrial production, roll equipment typically has a large roll diameter, high line speed and high extrusion pressure, and there are multiple high risk areas during operation, including roll gap nip, drive chain/gear exposure, loading and unloading stations, equipment maintenance channels, and the like. In order to ensure personal safety of operators, a safety protection system combining 'civil air defense' and 'technical defense' is commonly adopted in the prior art. In the aspect of civil air defense, on-duty workers need to receive regular safety training, the content covers the operation rules (including standard processes of start and stop, distance adjustment, roller cleaning and the like) of rolling equipment, daily spot inspection and basic maintenance skills, the standard wearing requirements of personal protective equipment (such as cut-resistant gloves, anti-smash shoes and goggles), site safety identification and recognition, response procedures (such as mechanical sprain treatment and high-temperature scald emergency treatment) under emergency conditions and typical accident case warning education, and meanwhile, enterprises strengthen personnel safety awareness and behavior constraint by establishing management mechanisms such as pre-shift safety confirmation, double mutual protection, hidden danger reporting excitation and the like. In the technical protection aspect, the current mainstream configuration comprises two types of physical protection facilities, namely a pull rope type emergency stop device and a correlation grating, wherein the pull rope is distributed along the whole length of dangerous areas on two sides of rolling equipment and is connected to an emergency stop switch by a high-strength steel wire rope, an operator can pull the pull rope at any position to cut off a main power supply and brake a roller, the correlation grating is arranged in front of a roller gap, at a feeding hole and at a main passing path, an invisible light curtain is formed by an infrared emitter and a receiver, and the interlocking stop is triggered when a human body or a limb enters a set protection surface. However, in actual operation, people's air defense measures are easily affected by factors such as insufficient ambient light, high temperature and high humidity, continuous noise interference, personnel fatigue, emotion fluctuation, individual attention dispersion and the like, so that the operation is neglected or the response is delayed, and when a plurality of people work cooperatively, misoperation can be caused due to unsmooth communication or fuzzy responsibility. In the technical protection layer, the pull rope can be passively triggered only when a danger occurs or is imminent, and the pull rope depends on subjective judgment and instant action of personnel, so that the prevention in advance is difficult to realize; in the rubber calendaring process, the anti-reflective grating has real-time intrusion detection capability, but because the adhesive is high in viscosity, easy to throw and splash and easy to adhere dust, light beams are often blocked by mistake, unplanned shutdown is caused, the protection range is limited to a set plane, three-dimensional spaces above equipment, below roller ends, below an overhaul platform and the like cannot be covered, and more importantly, when an operator is out of the protection range of the grating (such as ascending overhaul) and other post personnel send a start command by mistake, the system cannot identify the abnormal man-machine coexistence state, and serious runaway risk exists. Disclosure of Invention Therefore, the present invention is directed to a method and a device for controlling rolling equipment based on multi-objective detection, which can effectively prevent and intervene in real time in the high risk situation in the rolling equipment operation. In a first aspect, the present invention provides a method for controlling a rolling device based on multi-target detection, where at least two image capturing devices are disposed at a designated position of an area where the rolling device is located, and a physical monitoring area is defined in advance in a viewing area of each image capturing device, the method comprising: acquiring image data acquired by the image pickup device aiming at the rolling device; Perfo