58 results about "Active vision" patented technology

The invention provides an intelligent pipeline welding method and system based on data decision, and relates to the technical field of intelligent pipeline welding.The intelligent pipeline welding method comprises the steps that a welding request is responded, a groove image of a to-be-welded pipeline is obtained through an active visual sensor, groove morphology characteristic parameters of the to-be-welded pipeline are determined based on the groove image, and the to-be-welded pipeline is welded according to the groove morphology characteristic parameters. Based on the groove morphology characteristic parameters, corresponding basic layer channel arrangement rules and reference process parameters are determined from a preset layer channel arrangement database, a welding process scheme matched with the current groove is generated in combination with the groove morphology characteristic parameters, and the current groove is welded on the basis of the welding process scheme. And the welding operation is executed in the groove of the to-be-welded pipeline, the technical problems that in the prior art, the global planning capacity is deficient, the deviation correction strategy is locally lagged, and the whole-process accurate control is insufficient are solved, and the effect of improving the stability and reliability of the welding quality of the pipeline is achieved.

The invention discloses a robot grabbing system based on visual perception, and relates to the field of robot grabbing systems, and the robot grabbing system comprises a multi-mode active data perception module, a scene dynamic pre-judgment and decision module, a flexible self-adaptive grabbing execution module and a cross-module cooperative control module. According to the robot grabbing system based on visual perception, the multi-mode active visual perception module effectively solves the problem of inaccurate perception in backlight, shielding, transparent and reflective target scenes through active view angle adjustment and a multi-mode image fusion technology, the target recognition accuracy can be effectively improved, and the robot grabbing system based on visual perception has the advantages that the robot grabbing efficiency is improved. The system can adapt to various complex industrial and life scenes, the cross-module cooperative control module reduces the response delay of the system to a millisecond level through high-precision time sequence synchronization and distributed data interaction, meanwhile, the system has an abnormal self-adaptive adjustment capability, the cooperative lagging and single-point fault risk of an existing linear architecture is avoided, and the system reliability is improved. And the operation stability and reliability of the system are improved.

The invention discloses a weld defect detection method based on multi-feature extraction and hierarchical SVM (Support Vector Machine) fusion, which comprises the following steps of: extracting image features of a weld surface through passive vision and laser vision sensing technologies, and inputting obtained weld parameters as features of a hierarchical SVM; and converting a multi-classification problem into a plurality of dichotomy problems through the structure of the classifier, and identifying and classifying the six welding seam targets to obtain a final result. According to the method, multi-feature extraction of weld defects is introduced, geometric shapes, texture features and spectral information in weld images can be comprehensively captured through the multi-feature extraction technology, so that the defect recognition precision and reliability are remarkably improved, multi-feature extraction of passive vision and active vision is adopted, feature information of the weld defects is obtained from multiple dimensions, and the defect recognition accuracy is improved. And a multi-classification problem is converted into a plurality of dichotomy problems, so that the welding defects on the surface of the workpiece in a welding scene can be quickly classified and identified.

The invention discloses an active visual explanation system for blind person field travel, and belongs to the technical field of man-machine interaction. Based on the system, a user can selectively broadcast navigation obstacle avoidance or scenery description, real-time image information collected by a deep vision detection module is recognized through a deep neural network model, speed and type information of objects in front and below is obtained in combination with a positioning system, and when the user selects to broadcast navigation obstacle avoidance, the navigation obstacle avoidance or scenery description is broadcasted. After the audio instruction sequence corresponding to each sub-state is generated according to the shortest state sequence, the user acts under the guidance of the audio instruction, and each action can be fed back by the system, so that the correctness and efficiency of the traveling forward route operation are greatly improved; when the user selects to broadcast the beautiful scenery description, the AI model generates an audio instruction sequence according to the identified object information and informs the user of the audio instruction sequence; not only is the safety of the user during the on-site travel improved, but also the user can be helped to know the scenic spot in which the user participates, and the whole-process visual supervision is provided for the on-site travel of the user.

The embodiment of the invention discloses a root canal orifice detecting and positioning method, device and equipment based on active vision and a medium, and belongs to the technical field of stomatology and medical image.The method comprises the steps that an image quality evaluation score serves as a judgment index of the active vision; and adjusting the lens according to the image quality scores of the front and back positions in the process before and after focusing until the imaging position with the highest image quality evaluation score is found, and completing automatic focusing. Based on a dental pulp bottom image collected after automatic focusing, extracting sub-pixel coordinates of the edge of a root canal orifice by adopting a gray gradient interpolation refining and high-low threshold screening mode; based on the sub-pixel coordinates of the edge of the root canal orifice, the root canal orifice is regarded as a circular model, and the aperture center coordinate of each root canal orifice is fitted by using a least square method; the problems of tedious doctor operation, low accuracy and high repeatability in the root canal preparation process are solved, the root canal treatment time is saved, and a convenient and efficient auxiliary tool is provided for oral medical treatment.

The embodiment of the invention relates to the technical field of visual feedback, and provides a self-adaptive visual feedback system and method for ensuring material forming quality, the system comprises a front-end framework and a rear-end framework which work cooperatively, the front-end framework comprises a sensing module, a feature extraction module, an error compensation module and an execution control module, the back-end architecture comprises a decision module, a twin simulation module and an evolution module; the system further comprises a quality analysis module. According to the embodiment of the invention, environmental interference is overcome through active vision and event camera collaborative perception, long-term precision is maintained through online calibration compensation, quality quantitative evaluation is realized by combining three-dimensional shape real-time reconstruction and a deep learning method, a digital twin technology is introduced to carry out strategy security verification, and continuous evolution of the system is realized by means of a federated learning architecture. The method is not only suitable for the welding process, but also can effectively meet the high-quality control requirements of various material forming processes such as spraying, cladding and surfacing.

The present invention relates to the field of visual positioning technology, and specifically, to a method for automatically plugging and unplugging a charging gun of a mobile charging robot based on active vision positioning technology. The method includes: capturing an image of a vehicle and inputting it into a cloud large model to automatically identify the vehicle model; using a multi-modal SLAM pose estimation algorithm to locate the vehicle pose, generating a three-dimensional model of the vehicle pose and the position coordinates of the charging gun; capturing an image of the charging port and using an advanced active vision positioning algorithm to real-time locate the position coordinates of the charging port; introducing the influence of the robotic arm vibration, using the RRT algorithm to calculate the movement path of the robotic arm, and completing the plugging and unplugging operation through real-time feedback control of the robotic arm. The method for automatically plugging and unplugging a charging gun of a mobile charging robot based on active vision positioning technology solves the problem of precise positioning of the vehicle pose and the charging gun position through a multi-modal SLAM pose estimation algorithm, and real-time corrects the position of the charging port through an advanced active vision positioning algorithm, calculates the influence of the robotic arm vibration, and realizes precise automatic plugging and unplugging of the charging gun.

The invention discloses an intelligent anti-bird collision warning method and system and a medium, relates to the technical field of electric power safety protection, is improved on the basis of a mature structure and an installation mode of a traditional reflective warning ball, and intelligently upgrades the prior art on the basis of an active vision enhancement and guidance strategy direction. By integrating a sensing system and a light emitting system and based on bird visual ecology, while the advantages of a passive reflection mode are reserved, the capability of emitting bird sensitive spectrums (such as ultraviolet light) and the capability of randomly switching dynamic flickering in multiple preset flickering modes are added, and an all-weather, anti-adaptation and high-efficiency intelligent bird collision prevention solution is formed; the warning device is upgraded from passive reflection to active sensing and light emitting, the recognizable degree of the wire is fundamentally improved, and birds are guided to safely evade.

The invention belongs to the field of robot perception, and provides an active visual perception method and system based on fusion of a concept scene graph and a large model. The method comprises the following steps: based on an image of a set scene, extracting candidate entities in the image and text description and spatial relationship of the candidate entities; extracting a target entity from a user instruction, calculating a vision-language alignment matching score between text description of the target entity and candidate entities, and constructing a concept scene graph by taking the candidate entity with the highest matching score as a main node; and processing the concept scene graph by using a large model embedded with a ternary prompt project, and generating a strategy code of a current task scene in combination with a successful historical task strategy code, so that a robot actively senses implicit information in the scene, and recognition and capture of a shielded target entity with visual confusion are realized.

This invention discloses a navigation-following unmanned surface vessel (USV) formation control method based on active vision, comprising the following steps: constructing a kinematic model of the USV based on a geodetic coordinate system; establishing a gimbal monocular camera model for the gimbal camera mounted on the USV; defining servo tracking errors for the navigation-following USV formation system by combining the USV kinematic model and the gimbal monocular camera model; defining image coordinate errors based on the viewing angle constraints of the gimbal monocular camera; designing a motion controller and adaptive parameter update law for the USV using the Lyapunov synthesis method; and designing a gimbal camera steering controller and adaptive parameter update law using the Lyapunov synthesis method. This method ensures the smooth implementation of visual servo control of the USV by defining servo tracking errors and image coordinate errors, and then incorporating these two errors into the Lyapunov synthesis method to design the USV motion controller, the gimbal camera steering controller, and the adaptive parameter update law.

Embodiments of the present application disclose an active vision-based root canal orifice detection and positioning method, device, equipment and medium, belonging to the field of oral medicine and medical image technology, wherein the method comprises: taking an image quality evaluation score as a judgment index of active vision, adjusting the lens according to the image quality scores of the front and rear positions in the process before and after focusing until the imaging position with the highest image quality evaluation score is found, and completing automatic focusing; based on the pulp bottom image collected after automatic focusing, the sub-pixel coordinates of the root canal orifice edge are extracted by using the method of gray gradient interpolation refinement and high-low threshold screening; based on the sub-pixel coordinates of the root canal orifice edge, the root canal orifice is regarded as a circular model, and the least square method is used to fit the center coordinate of each root canal orifice aperture. The present application solves the problems of complicated operation, low accuracy and high repeatability of doctors in the root canal preparation process, saves the time of root canal treatment, and provides a convenient and efficient auxiliary tool for oral medical treatment.

The application discloses a ranging method for coinciding central viewpoints of binocular PTZ cameras, solves the problem that a calibration plate cannot be used for recalibration in an online environment during three-dimensional detection of active vision, and belongs to the technical field of active vision.The application comprises the following steps: left and right cameras of binocular PTZ cameras are used to shoot target images at respective positions in a spherical cavity with various pitch angles and azimuth angles, and an angle-image encoder is obtained; during ranging, the target is shot by the left and right cameras; according to the shot images, the azimuth angle and the pitch angle corresponding to the image pair shot by the left and right cameras are found in the angle-image encoder; then, the included angles alpha0 and beta0 between the left and right cameras and the target point are obtained in combination with the geometric relationship between the positions of the left and right cameras and the target; and finally, the three-dimensional distance of the target point is solved according to alpha0, beta0 and the distance D between the position points of the rotation centers of the left and right cameras.The application can still be used for ranging when the central viewpoints coincide without self-calibration under the condition that the internal and external parameters change after the focusing and the shooting angle and range are adjusted.

The invention discloses a multi-layer multi-pass welding early warning and path planning method based on active vision, and relates to the technical field of welding robots. The method at least comprises the following steps that S1, laser active vision groove feature point recognition is conducted, a groove of a workpiece to be welded is recognized through a laser active vision recognition technology, and image information is provided for groove width calculation; s2, welding groove width calculation and early warning are conducted, the welding groove width is calculated through image processing and a width algorithm, and then the early warning function is achieved when the groove width exceeds the safe welding range; and S3, multi-layer and multi-pass welding path planning is carried out. On the basis of the laser active visual recognition technology, aiming at multi-layer and multi-pass welding of the welding robot, automatic welding of multi-layer and multi-pass welding of the welding robot is achieved through three methods of laser active visual groove feature point recognition, welding groove width calculation and early warning and multi-layer and multi-pass welding path planning.

The present application belongs to the technical field of intelligent welding, and discloses a multi-layer multi-pass weld tracking method and system based on active and passive vision. The system of the present application comprises an active and passive vision weld tracking sensor, a welding robot and an industrial computer. The active and passive vision weld tracking sensor comprises an active vision device, a passive vision device and a linear laser emitter. The method of the present application generates an original groove model by detecting groove information with a structured light. Then, the active and passive vision are combined to detect the weld surface shape and the weld corner point position during welding, and a groove model of the current weld is established. The groove model of the current weld is compared with the original groove model to determine whether the weld is offset and measure the offset, so as to determine the position of the next weld and plan and generate a preset welding gun posture, thereby avoiding problems caused by heat loss estimation failure, weld offset and poor welding gun posture.

The invention discloses an active visual guidance ton bag sealing and shaping method and device. The device comprises a visual acquisition module, double mechanical arms and a main control unit. According to the method, RGB and depth data of a ton bag sealing area are collected and preprocessed through the camera module, after inner bag sealing line extraction, operation point screening and feasibility inspection are completed through the detection module, the decision control module generates an outer bag adjusting / shaping instruction according to a detection result, and after coordinate mapping, the double mechanical arms are controlled to cooperatively execute operation. And an active visual closed loop of visual acquisition, detection positioning, decision control and operation execution is formed. According to the scheme, the problems of inaccurate inner bag positioning, difficult outer bag posture adjustment and easy collision in existing ton bag sealing and shaping are solved, the shaping precision and the operation stability are improved, and the device is suitable for automatic packaging production lines in the fields of chemical engineering, grains and the like.

The present invention discloses a viewpoint planning method based on deep reinforcement learning. The method includes: obtaining an occlusion image and a camera pose to obtain state information; inputting the state information into a trained DQN network to obtain the camera movement action at the next moment; controlling the manipulator to move to adjust the camera position. Among them, in order to improve the training efficiency of the DQN network, a training strategy of global trend-guided learning is creatively proposed. The so-called global trend-guided learning strategy refers to a method of forming a global trend by recording the viewpoint regions where fruit stalks are successfully detected in the same scene during the learning process, and then using the global trend to guide the network learning process. The present invention realizes viewpoint planning by a brand-new technical means to solve the occlusion problem faced by grape picking robots during the picking operation. The method greatly shortens the viewpoint planning time while maintaining a high picking success rate, and provides a new technical solution for the application field of active vision technology.

The invention discloses a road anomaly detection method, which is characterized by comprising the following steps of: firstly, constructing a road anomaly detection model, and training the road anomaly detection model to obtain a trained road anomaly detection model; inputting a to-be-detected road scene image into the trained road anomaly detection model to obtain a detection result; wherein the road anomaly detection model is composed of a backbone network, a neck network and a detection head which are connected in sequence; the backbone network comprises a CBS layer, a C3 layer, a position modulation attention module and an active visual follow-up multi-scale feature extraction module; the method has the advantages that a long-distance dependency relationship is effectively captured through the position modulation attention module to perform accurate positioning and enhance the scene adaptability, and the weak feature target detection capability under the complex road background is improved; the active vision follow-up multi-scale feature extraction module actively adjusts the sight distance to perform multi-scale feature extraction in a multi-receptive-field mode under the condition of fixing the position of a target to be detected, and the limitation of a single-scale method is relieved.

The invention discloses an active vision-based multi-layer and multi-pass welding seam feature identification method, and relates to the technical field of XX. According to the method, laser images are obtained through active vision, a center line extraction and correction technology is adopted, geometric features of welding seams are extracted, a corner recognition and correction technology is combined, welding points are positioned, the welding process is divided into backing welding, filling welding and cosmetic welding, and center line correction and corner recognition and correction methods are selectively adopted for welding of each layer according to welding characteristics. Therefore, the geometric features of the laser center lines of the multiple welding seams are recognized and corrected, automatic welding of the welding robot in multi-layer and multi-pass welding operation can be achieved, manual participation is reduced, and the welding efficiency is improved.

The invention relates to the field of metallurgical equipment state monitoring, and discloses a method and a device for detecting and judging cracks of a refractory layer of a ladle before full-size iron receiving, and the method comprises the steps: obtaining an image of the inner wall of the ladle; performing first-stage processing on the image, and quickly positioning one or more suspected crack regions by using a convolutional neural network; second-stage processing is carried out on each suspected crack area, and three-dimensional physical parameters such as the physical width and the physical depth of the crack in the area are calculated by actively projecting an optical pattern and analyzing geometric deformation of the optical pattern; and carrying out numerical comparison on the calculated physical parameters and a preset safety standard, and carrying out automatic ladle judgment on the ladle. According to the method, through a two-stage processing strategy, the rapid identification capability of deep learning and the accurate quantification capability of active visual three-dimensional measurement are combined, and accurate acquisition of the physical size of the crack is realized, so that reliable data support is provided for automatic and objective packet judgment decision, and the accuracy and reliability of detection are remarkably improved.

The techniques described herein are directed to fundus cameras, and more particularly, to fundus cameras having a display that projects an active vision alignment stimulus onto an eye of a subject via one or more components of an optical assembly. A fundus camera includes: an optical assembly including one or more optical components; a display adapted to project an active alignment pattern and a reference alignment pattern onto the eye of the subject via the optical assembly; a camera adapted to capture one or more images of the fundus of the subject; an eye tracker adapted to monitor the position of the subject's eye relative to the fundus camera; and a control system operatively coupled to the display, the camera, and the eye tracker and configured to direct the display to guide the subject toward pupil-to-camera alignment by dynamically adjusting the active alignment pattern relative to the reference alignment pattern based on the position of the subject's eye relative to the fundus camera. The active vision alignment stimulus is dynamically adjusted to guide the subject toward optical alignment for fundus imaging.

The invention relates to a method and a system for identifying an underwater object on a mobile carrier, and belongs to the technical field of underwater target detection and identification. The method comprises the following steps: detecting the water depth of a mobile carrier through a water depth sensor; when the water depth is greater than a preset threshold value, the industrial personal computer starts and adaptively adjusts the underwater active lighting system; acquiring an image by using an optical imaging system; the industrial personal computer preprocesses the image, and inputs the processed image into a pre-trained deep learning neural network model for target recognition; and finally outputting a target identification result. The system comprises a bottom plate, a water depth sensor, an underwater active lighting system, an optical imaging system and an industrial personal computer. According to the invention, active visual illumination and artificial intelligence identification are combined, the defects of low precision of a traditional acoustic method and poor environmental adaptability of an optical method are overcome, high-precision, self-adaptive and automatic identification of an underwater target on a mobile carrier is realized, and the system has the advantages of small size, low power consumption and high integration level.

The application discloses a fruit autonomous picking method and system based on active perception and visual arm cooperative optimization, and relates to the technical field of robot control. The method comprises the following steps: acquiring a color image and a depth image of a working environment, and performing semantic segmentation on the color image by using a preset zero-sample learning model; performing semantic point cloud extraction on the depth image based on a mask corresponding to a target category, and constructing a fruit tree environment map containing spatial information and semantic information; performing candidate target screening and extraction according to the fruit tree environment map, and generating a candidate grasping pose according to the screened candidate target; performing kinematic reachability analysis on all candidate targets and candidate grasping poses, and calculating a joint score; determining a next grasping object according to the joint score; and picking the next grasping object according to a preset trajectory. The application deeply couples active visual perception and mechanical arm grasping action, and realizes efficient fruit autonomous picking of the robot.

The invention relates to the technical field of video conferences, and discloses a video conference target intelligent tracking method based on a panoramic PTZ camera, and the method comprises the steps: starting a system, and loading a panoramic and PTZ lens space coordinate mapping model which is established through polynomial fitting in advance, the method comprises the following steps: acquiring images through a panoramic lens, detecting and identifying all figure information in an application environment and a scene by utilizing a vision and audio algorithm, extracting active vision, passive vision, active audio and passive audio of each target, performing feature scoring, calculating a comprehensive priority score of each target, selecting the target with the highest score for tracking, and performing target tracking. The method comprises the following steps: calculating the horizontal and vertical rotation angles required by a PTZ lens according to the normalized coordinates of a selected target in a panoramic picture through a mapping model, sending the calculated rotation angles to a PTZ holder, driving the lens to rotate, placing the target in the center of the picture, continuously updating the priority of the target and the PTZ position, and accepting user interaction to adjust a tracking strategy.

The application patent is based on the steel coil band loosening anomaly detection and quantification method of active vision imaging, first, the saliency detection model is constructed to obtain the brightness saliency map of the laser stripe, second, the laser stripe image is segmented by using the serialized threshold to obtain the Boolean graph, the weighted sum of different binary images is obtained to obtain the region stability image. The brightness saliency map and the region stability saliency map are fused by wavelet transform, the mean fusion is adopted for the brightness saliency map, the maximum value fusion mode is adopted for the region stability saliency map, and the adaptive maximum entropy segmentation is carried out on the fused saliency map, finally, the stripe normal field is obtained by calculating the stripe gradient vector, and the stripe center line is extracted based on the stripe distribution normal and the fusion gray gravity center method. The quantification of the loosening band is realized by calculating the neighborhood difference value and the curvature radius of the stripe center, and the application prospect is wide.

The invention relates to a multi-mode active vision following method and system. The method comprises the following steps: acquiring an original image comprising an RGB image and a single line point cloud and point cloud data information through a camera and a single line laser radar; performing dimension reduction processing on the original image and the point cloud data information, processing the original image information through a lightweight target detection model to obtain an image key information structure, performing PCA offline learning on the original point cloud data to obtain fixed parameters, and performing weighted fusion on the distance, the intensity and the angle according to the fixed parameter information to obtain point cloud dimension reduction fusion data; constructing a quantitative relation expression of visual entropy and radar entropy according to the image key information structure and the point cloud dimension reduction fusion data; performing multi-source entropy information fusion to obtain an optimal following target; converting the position information into decision information and transmitting the decision information to a control steering engine of a physical execution layer; according to the invention, the target identification, positioning and tracking precision and reliability of the mobile intelligent agent in a multi-target dynamic environment are significantly improved.

The present invention discloses an active vision-based positioning device and method for a tunnel boring machine. The device comprises a tunnel boring machine, an active vision module, an artificial feature object, a total station, and a positioning control system. The active vision module is mounted on the tunnel boring machine, and the artificial feature object is positioned between the active vision module and the total station. The positioning control system is used to control the active vision module to acquire images of the artificial feature object at different positions and viewing angles, and directly estimate the pose parameters of an industrial camera in the active vision module relative to the artificial feature object based on the images. Combining the pose state of the artificial feature object in the tunnel and the pose state of the industrial camera relative to the tunnel boring machine, the absolute six-degree-of-freedom pose of the tunnel boring machine in the tunnel is obtained through homogeneous coordinate transformation. The total station is used to reposition the artificial feature object after it is moved, thereby achieving autonomous and continuous positioning of the tunnel boring machine. The present invention has the advantages of high robustness, high precision, good real-time performance, low cost, and simple system construction.

The invention discloses a space robot sensing system based on an active visual strategy, which relates to the technical field of robot sensing and comprises an instance segmentation module, a point cloud generation module, a shape clustering module, a shape fitting module, a missing surface visual angle synthesis module, an exploration sampling module and a utility measurement module. The invention further discloses a space robot sensing method based on the active vision strategy. The method comprises the steps that S100, a target part image is collected; s200, extracting a mask of the part instance; s300, generating a three-dimensional point cloud with color information; s400, clustering the shapes of the part instances; s500, fitting parameters of the shape of the part instance; s600, generating an exploratory view angle; s700, synthesizing a target view angle; s800, obtaining an optimal observation visual angle; and S900, finishing active vision strategy planning. According to the method, an optimal visual angle synthesis method is combined, it is ensured that the mechanical arm rapidly and accurately completes part grabbing, and the part recognition accuracy is improved.

The present invention belongs to the field of visual navigation technology, and specifically relates to an active visual detector posture measurement method, medium, and system, comprising the following steps: defining that the optical centers of all laser rangefinders are coplanar, and that the optical center of any laser rangefinder is perpendicular to the line connecting the optical centers of two adjacent laser rangefinders; all laser rangefinders are arranged along the axis of the detector; setting the optical center of any laser rangefinder as the origin and establishing a detector coordinate system; determining the optical center of the laser rangefinder; determining the three-dimensional coordinate information of the contact point; determining the normal vector of the basic landing plane; determining the landing plane coordinate system; determining the posture; determining the translation; and optimizing the adjustment. Compared with the prior art, the present invention solves the problems of the prior art in the detector's insufficient perception of complex environments and poor landing navigation performance. This solution achieves highly robust and stable detector navigation posture estimation through a low-complexity detector posture measurement method.

This invention relates to the field of metallurgical equipment condition monitoring, and discloses a method and apparatus for detecting and judging cracks in the refractory layer of a full-size molten iron ladle before it receives hot iron. The method includes: acquiring an image of the inner wall of the molten iron ladle; performing a first-stage processing on the image to quickly locate one or more suspected crack areas using a convolutional neural network; performing a second-stage processing on each suspected crack area by actively projecting an optical pattern and analyzing its geometric deformation to calculate the three-dimensional physical parameters of the crack, such as its physical width and depth; and comparing the calculated physical parameters with preset safety standards to automatically judge the molten iron ladle. This invention, through a two-stage processing strategy, combines the rapid recognition capability of deep learning with the precise quantification capability of active visual three-dimensional measurement, achieving accurate acquisition of the physical dimensions of cracks. This provides reliable data support for automated and objective ladle judgment decisions, significantly improving the accuracy and reliability of detection.

The present invention provides a multi-robot collaborative grasping method and system based on a metaheuristic algorithm. The method comprises: performing camera calibration on at least two robots to obtain geometric model parameters; performing active visual detection to select the optimal grasping object and obtain visual features; calculating the target posture deviation of the robot manipulator; performing deep reinforcement learning in an offline and online manner to optimize and converge the target posture deviation of the manipulator; inputting the deviation into a PID controller to output the manipulator's torque, then using the target posture deviation as the input signal of a self-organizing neural network and the manipulator's torque as the target parameter of a deep belief network, simultaneously training two serial neural networks to construct a deep learning controller; and finally, guiding the manipulator to a designated position to grasp the target through the negative feedback mechanism of the feedback controller. The present invention achieves effective active visual recognition and metaheuristic optimization control, improving system stability.