Detection method of lifting hook anti-falling device

Abstract

The invention discloses a detection method of a lifting hook anti-falling device. The method comprises: obtaining an image sample set; constructing a single-stage target detection model for identifying a crane, a lifting hook and a lifting hook anti-falling device; deploying the lifting hook anti-falling device detection model into a camera of an operation site, and connecting the camera to a deployment platform; obtaining a to-be-detected image; and inputting the to-be-detected image into the lifting hook anti-falling device detection model to determine whether the lifting hook anti-falling device is detected or not, and the like. The method has the advantages that the accuracy of identifying and analyzing whether the lifting hook anti-disengaging device is installed or not can be remarkably improved in the lifting operation scene, abnormal conditions are found in time, and the safety supervision and inspection efficiency of lifting operation is improved. The small target identification accuracy is improved, and the detection accuracy of the lifting hook anti-falling device in the lifting operation is greatly improved. The detection range is wide by using the camera, the detection performance is improved by using the convolutional neural network with high accuracy and adding the selective visual attention mechanism module, and the real-time performance is good by using only one camera.

Description

Technical Field

[0001] The invention belongs to the technical field of hoisting operation safety management, and in particular relates to a detection method for a hook anti-dropping device. Background Art

[0002] Lifting operations are one of the special dangerous operations of oilfield companies. In order to prevent the objects from falling off the crane hook, the hooks in operation must be equipped with hook anti-drop devices. The internal management of the oilfield requires real-time monitoring of whether the cranes in operation are equipped with hook anti-drop devices to minimize the chance of accidents. However, the relevant safety management departments are currently facing the problems of small staff and wide management areas. Even if online inspections are carried out through the existing industrial video surveillance system, it is impossible to achieve full coverage of the inspection of the operation site. Therefore, it is necessary to use target detection technology to analyze and identify the hooks that are not equipped with hook anti-drop devices and alarm them to ensure the safety of the workers and meet the safety management requirements.

[0003] Chinese patent application No. 201320233154.6 discloses a "crane hook monitoring and tracking system", which uses two cameras with different priorities to capture hook images, and judges the height value of the image captured by each camera to infer the position of the hook at the next moment. The advantage of this method is that it detects the hook position from multiple aspects, but the disadvantage is that it uses too many cameras, resulting in poor real-time performance.

[0004] Chinese Patent Application No. 201710854585.7 discloses "A target detection system and method", which obtains the image to be detected, determines the target to be detected in the image to be detected, pre-processes the image to be detected, obtains the illumination and rotation invariant features and inputs them into the target detection classifier corresponding to the target to be detected for target detection. The advantages of this method are: the hook position is detected by the illumination and rotation invariant appearance features of the hook, highlighting the hook features, and the disadvantages are that the position relationship between the hook and the crane is not used to determine the hook position, and the relationship between the crane and the surrounding objects is not modeled.

[0005] At present, the detection technology of hook anti-dropping device mainly includes the following aspects:

[0006] The detection method based on the inter-frame comparison method is based on the principle of comparing two or three adjacent frames of images. By calculating the difference in pixel values ​​between adjacent frames, it is determined whether the hook anti-slip device exists and its specific location. This method has poor accuracy and requires obtaining multiple images for comparison at the same time, and cannot achieve real-time monitoring.

[0007] The detection method based on region selection and feature extraction uses a sliding window to traverse the image pyramid to select regions for an image, and then manually creates and extracts features for these regions to classify the classifier. This method requires repeating the same region of an image, thus wasting computing resources, and the manual features are poor and have low accuracy. Summary of the invention

[0008] In order to solve the above problems, the object of the present invention is to provide a detection method for a hook anti-dropping device.

[0009] In order to achieve the above object, the detection method of the hook anti-dropping device provided by the present invention comprises the following steps performed in sequence:

[0010] 1) Collect various images containing cranes, hooks and hook anti-dropping devices offline, then pre-process and annotate them to obtain image sample sets, and then divide the image sample sets into training sets, verification sets and test sets according to proportion;

[0011] 2) constructing a single-stage target detection model for identifying cranes, hooks, and hook anti-dropping devices, wherein the single-stage target detection model is composed of a backbone network and a selective visual attention mechanism module, wherein the backbone network is composed of multiple convolutional layers and pooling layers; then inputting the above training set into the single-stage target detection model for training, and then using the validation set and the test set to evaluate and test the trained single-stage target detection model, respectively, to obtain a hook anti-dropping device detection model;

[0012] 3) Deploy the above-mentioned hook anti-dropping device detection model into a camera at the work site, and connect the camera to the deployment platform;

[0013] 4) Using a camera to collect videos of the crane, hook, and hook anti-drop device, and then extracting frames from the videos to obtain JPEG images as images to be detected;

[0014] 5) The above-mentioned image to be detected is input into the hook anti-dropping device detection model for feature extraction to obtain a feature map of the image, and then the feature map is used to detect the hook and the hook anti-dropping device as the target to determine whether the hook anti-dropping device is detected.

[0015] In step 1), the offline collection of various images containing cranes, hooks and hook anti-dropping devices is followed by preprocessing and annotation to obtain an image sample set, and then the specific steps of dividing the image sample set into a training set, a validation set and a test set in proportion are as follows:

[0016] 1.1) Use image acquisition equipment to capture various images containing cranes, hooks and hook anti-drop devices;

[0017] 1.2) Preprocess the above images including denoising, rotation, color gamut conversion, cropping and scaling to improve the quality and consistency of the images and expand the sample data;

[0018] 1.3) Manually annotate the preprocessed images, mark the center coordinates and width and height of the crane, hook and hook anti-drop device image frame, use the annotated images as image samples, and form an image sample set with all image samples. Then divide the image sample set into training set, validation set and test set in a ratio of 7:1:2 for model training, evaluation and testing.

[0019] In step 2), the YOLO or SSD model is selected as the backbone network; the selective visual attention mechanism module includes SENet and CBAM; the training method is to first compare the output of the single-stage target detection model with the annotation and calculate the loss function; then, the single-stage target detection model is trained using the back gradient propagation algorithm, and the model weights are updated using the stochastic gradient descent optimization algorithm;

[0020] Use the validation set to evaluate the performance of the trained single-stage object detection model, calculate indicators including accuracy, recall, and mean average precision (mAP), and tune the model based on the evaluation results;

[0021] Use the test set to test the trained model and test its performance on unseen data.

[0022] In step 3), the deployment platform includes a local server, a cloud platform or an embedded device.

[0023] In step 4), the method of using a camera to collect a video of a crane, a hook, and a hook anti-drop device, and then performing frame extraction on the video to obtain a JPEG image as an image to be detected is:

[0024] Using FFmpeg video frame extraction technology, the video of the crane, hook and hook anti-dropping device captured by the camera is extracted with a certain frame interval to obtain a JPEG image with a high compression rate.

[0025] In step 5), the method of inputting the above-mentioned image to be detected into the hook anti-off device detection model for feature extraction to obtain a feature map of the image, and then detecting the hook and the hook anti-off device as the target on the feature map to determine whether the hook anti-off device is detected is:

[0026] The image to be detected is input into the backbone network of the hook anti-dropping device detection model, and the feature map of the image is extracted by using the forward propagation process;

[0027] Then, target detection is performed on the feature map. During the target detection process, each position in the feature map needs to be classified and regressed to identify the hook and the hook anti-dropping device, and obtain the center point, width and height of the target frame of the hook and the hook anti-dropping device, a total of four coordinate values. Afterwards, the intersection-and-union ratio of the hook and the hook anti-dropping device is calculated using the above four coordinate values, and a threshold is pre-set according to the degree of overlap between the two targets of the hook and the hook anti-dropping device. If the intersection-and-union ratio is less than the threshold, it is deemed that the hook anti-dropping device is not detected; otherwise, it is deemed that the hook anti-dropping device is detected.

[0028] The method for calculating the intersection and union ratio of the hook and the hook anti-dropping device by using the center point and width and height coordinate values ​​of the target frame of the hook and the hook anti-dropping device is as follows:

[0029] 5.1) Calculate the intersection of the two target frames of the hook and the hook anti-drop device, that is, the area of ​​the overlapping part:

[0030] Calculate the coordinates of the upper left corner and lower right corner of the two target boxes;

[0031] Calculate the areas of the two target boxes based on the coordinates of the upper left corner and the lower right corner;

[0032] According to the area of ​​the above target box, the coordinates of the upper left corner and the lower right corner of the intersection are calculated, and then the area of ​​the intersection is calculated;

[0033] 5.2) Calculate the area of ​​the union of the two target boxes:

[0034] According to the coordinates of the upper left corner and lower right corner of the two target boxes, calculate the area of ​​the union;

[0035] 5.3) Using the area of ​​the intersection and the area of ​​the union, calculate the intersection-union ratio:

[0036] Divide the area of ​​the intersection by the area of ​​the union to get the intersection-union ratio;

[0037] Then, it is determined whether the hook anti-dropout device is detected based on the above-mentioned intersection-and-combination ratio. If the intersection-and-combination ratio is less than the threshold value, it means that the hook anti-dropout device is smaller than the hook, and it is regarded that the hook anti-dropout device is not detected; otherwise, it is regarded that the hook anti-dropout device is detected, and an alarm is immediately issued.

[0038] The detection method of the hook anti-drop device provided by the present invention has the following beneficial effects:

[0039] (1) Improve inspection efficiency

[0040] It can significantly improve the accuracy of identification and analysis of whether the hook anti-slip device is installed in the lifting operation scenario, detect abnormal situations in time, and improve the efficiency of safety supervision and inspection of lifting operations. Compared with manual inspection, the number of video channels browsed at a time is limited, and the video patrol cycle is long, which has the risk of missed reports and delayed detection. According to preliminary calculations, taking the inspection of 100 video images as an example, manual inspection takes an average of about 10 minutes. Using this method, the abnormal situation of the hook without anti-slip can be found within 15 seconds, and the subsequent manual verification takes a total of 4 minutes, saving 60% of time and improving the inspection efficiency by 1.5 times.

[0041] (2) Improve the accuracy of detection of violations in lifting operations

[0042] In terms of target detection, the hook anti-drop device detection improves the accuracy of small target recognition by integrating the selective attention mechanism module, which greatly improves the detection accuracy of the hook anti-drop device in lifting operations, with an accuracy rate of up to 90%.

[0043] (3) Since cameras are used for detection, the detection range is wider. At the same time, a convolutional neural network with higher accuracy is used and a selective visual attention mechanism module is added to improve the detection performance. Using only one camera has better real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flow chart of a detection method for a hook anti-dropping device provided by the present invention. DETAILED DESCRIPTION

[0045] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] like Figure 1 As shown, the detection method of the hook anti-dropping device provided by the present invention comprises the following steps performed in sequence:

[0047] 1) Collect various images containing cranes, hooks and hook anti-dropping devices offline, then pre-process and annotate them to obtain image sample sets, and then divide the image sample sets into training sets, verification sets and test sets according to proportion;

[0048] The specific steps are as follows:

[0049] 1.1) Use image acquisition equipment to collect various images containing cranes, hooks and hook anti-drop devices; during the acquisition process, the impact of factors such as light, angle, and distance on image quality should be considered to ensure image clarity and accuracy; changes in light conditions may cause the image to be too dark or too bright, and an adaptive exposure adjustment algorithm can be used to dynamically adjust the exposure according to the image's histogram and content to ensure that the details of the image are properly captured. At the same time, changes in angle and distance may cause image deformation or information loss, and automatic perspective correction can be achieved through geometric transformation technology to adjust the collected image to a standard perspective to maintain image consistency and comparability.

[0050] 1.2) Preprocess the above images including denoising, rotation, color gamut conversion, cropping and scaling to improve the quality and consistency of the images and expand the sample data;

[0051] Deep learning requires a large amount of data to train accurate models. However, real-world image datasets are often limited and suffer from class imbalance and sample scarcity. Through data augmentation, more training samples can be generated by applying various transformations and perturbations, expanding the scale and diversity of the dataset, which can improve the generalization ability of the model and alleviate the overfitting problem.

[0052] 1.3) Manually annotate the preprocessed images, mark the center coordinates and width and height of the crane, hook and hook anti-drop device image frame, use the annotated images as image samples, and form an image sample set with all image samples. Then divide the image sample set into training set, validation set and test set in a ratio of 7:1:2 for model training, evaluation and testing.

[0053] The training set is a dataset used to train a deep learning model. During the training process, the model improves its performance and accuracy by learning and updating parameters on the samples in the training set. The training set should contain a sufficient number and diversity of samples to cover the various situations and changes that the model may encounter. The validation set is used to evaluate and adjust the model's hyperparameters, such as learning rate, regularization parameters, etc. During the training process, the validation set can be used to monitor the performance of the model and make adjustments based on its performance. By evaluating on the validation set, the best model hyperparameter configuration can be selected to improve the generalization ability of the model. The test set is used to evaluate the final performance and generalization ability of the model. After the model is trained and validated, the test set can be used for the final evaluation of the model. The samples in the test set have not been seen by the model before, so they can be used to measure the accuracy and performance of the model on real-world data. The evaluation results of the test set can be used as the final indicator of the model to compare different models or report the performance of the model. The purpose of dividing the training set, validation set, and test set is to ensure that the evaluation of the model is objective and consistent. The use of the validation set allows for model tuning and hyperparameter selection, while the use of the test set provides an unbiased estimate of model performance. This division method helps prevent the model from overfitting to a specific dataset and better measures the performance of the model on unseen samples.

[0054] 2) constructing a single-stage target detection model for identifying cranes, hooks, and hook anti-dropping devices, wherein the single-stage target detection model is composed of a backbone network and a selective visual attention mechanism module, wherein the backbone network is composed of multiple convolutional layers and pooling layers; then inputting the above training set into the single-stage target detection model for training, and then using the validation set and the test set to evaluate and test the trained single-stage target detection model, respectively, to obtain a hook anti-dropping device detection model;

[0055] The specific steps are as follows:

[0056] Select YOLO or SSD models as the neural network architecture. These models have fast detection speed and good detection accuracy. In the feature extraction stage of the model, a selective visual attention mechanism module is introduced to weight the obtained feature map. This module can assign different weight values ​​according to the importance of different areas in the feature map, so that the network pays more attention to the core area and improves the detection ability of the target. Commonly used selective visual attention mechanism modules include SENet (Squeeze-and-Excitation Network) and CBAM (Convolutional Block Attention Module).

[0057] The training method is to first compare the output of the single-stage target detection model with the annotation and calculate the loss function; then, use the backpropagation algorithm to train the single-stage target detection model, and use the stochastic gradient descent optimization algorithm to update the model weights; during the training process, some techniques such as learning rate decay and batch normalization can be used to improve the convergence speed and stability of the model.

[0058] The validation set is used to evaluate the performance of the trained single-stage object detection model, calculate indicators such as accuracy, recall, and mean average precision (mAP), and tune the model based on the evaluation results, such as adjusting the network structure and hyperparameters, to improve the performance of the model in the object detection task.

[0059] The trained model is tested using the test set to test its performance on unseen data. If the model achieves the expected performance level, it can be applied in real-world scenarios for object detection of cranes, hooks, and hook-stop devices.

[0060] 3) Deploy the above-mentioned hook anti-dropping device detection model into a camera at the work site, and connect the camera to the deployment platform;

[0061] Choose a suitable deployment platform, such as a local server, cloud platform, or embedded device, based on actual conditions. Considering that construction sites usually require real-time detection, choose a deployment platform that can provide sufficient computing resources and processing power. Deploy the camera at an appropriate location on the construction site to ensure that it can capture the area that needs to be detected, and ensure that the camera can be connected to the deployment platform to obtain video streams for real-time detection.

[0062] 4) Using a camera to collect videos of the crane, hook, and hook anti-drop device, and then extracting frames from the videos to obtain JPEG images as images to be detected;

[0063] Use FFmpeg video frame extraction technology to extract a number of frames at a certain interval from the video of the crane, hook and hook anti-drop device collected by the camera. The number of extracted frames can be controlled by setting a suitable frame rate (for example, extracting a few frames per second). This can reduce the amount of data to a certain extent and obtain a JPEG image with a higher compression rate.

[0064] 5) Inputting the above-mentioned image to be detected into the hook anti-off device detection model for feature extraction to obtain a feature map of the image, and then detecting the hook and the hook anti-off device as the target on the feature map to determine whether the hook anti-off device is detected;

[0065] In order to better learn image features, backbone network weights pre-trained on large-scale image datasets are usually used. These weights can be downloaded from public model libraries or training platforms, such as ImageNet pre-trained weights.

[0066] The image to be detected is input into the backbone network in the hook anti-slip device detection model, and the feature map of the image is extracted using the forward propagation process. The backbone network can extract the semantic and spatial features of the image layer by layer. As the network goes deeper, the size of the feature map gradually decreases, but the number of channels increases, and it has a higher level of abstract representation capability. The feature map extracted from the backbone network can undergo some subsequent processing operations to better meet the needs of the target detection task. Common processing includes adjusting the size of the feature map, the number of channels, applying activation functions, and standardization. As needed, feature fusion operations can be performed at different levels of the backbone network to fuse semantic and spatial features at multiple levels, which helps to improve the model's detection capabilities for targets of different scales.

[0067] Then, target detection is performed on the feature map. During the target detection process, each position in the feature map needs to be classified and regressed to identify the hook and the hook anti-dropping device, and obtain the center point, width and height of the target frame of the hook and the hook anti-dropping device, a total of four coordinate values. Afterwards, the intersection-and-union ratio of the hook and the hook anti-dropping device is calculated using the above four coordinate values, and a threshold is pre-set according to the degree of overlap between the two targets of the hook and the hook anti-dropping device. If the intersection-and-union ratio is less than the threshold, it is deemed that the hook anti-dropping device is not detected; otherwise, it is deemed that the hook anti-dropping device is detected.

[0068] The specific steps for calculating the intersection and union ratio of the hook and the hook anti-dropping device through the center point and width and height coordinate values ​​of the hook and the hook anti-dropping device target frame are as follows:

[0069] 5.1) Calculate the intersection of the two target frames of the hook and the hook anti-drop device, that is, the area of ​​the overlapping part:

[0070] Calculate the coordinates of the upper left corner and lower right corner of the two target boxes;

[0071] Calculate the areas of the two target boxes based on the coordinates of the upper left corner and the lower right corner;

[0072] According to the area of ​​the above target box, the coordinates of the upper left corner and the lower right corner of the intersection are calculated, and then the area of ​​the intersection is calculated;

[0073] 5.2) Calculate the area of ​​the union of the two target boxes:

[0074] According to the coordinates of the upper left corner and lower right corner of the two target boxes, calculate the area of ​​the union;

[0075] 5.3) Using the area of ​​the intersection and the area of ​​the union, calculate the intersection-union ratio:

[0076] Divide the area of ​​the intersection by the area of ​​the union to get the intersection-union ratio;

[0077] Then, it is determined whether the hook anti-dropout device is detected based on the above-mentioned intersection-and-combination ratio. If the intersection-and-combination ratio is less than the threshold value, it means that the hook anti-dropout device is smaller than the hook, and it is regarded that the hook anti-dropout device is not detected; otherwise, it is regarded that the hook anti-dropout device is detected, and an alarm is immediately issued.

[0078] It should be noted that the choice of threshold depends on the specific application scenario and requirements. A higher threshold will require a higher degree of overlap of the target box to determine that the hook anti-drop device is detected, while a lower threshold allows for looser overlap. Depending on the actual situation, the appropriate threshold can be selected based on experience or experiments.

Claims

1. A method for detecting a hook anti-drop device, characterized in that: The detection method of the hook anti-drop device comprises the following steps performed in sequence: 1) Collect various images containing cranes, hooks and hook anti-dropping devices offline, then pre-process and annotate them to obtain image sample sets, and then divide the image sample sets into training sets, verification sets and test sets according to proportion; 2) constructing a single-stage target detection model for identifying cranes, hooks, and hook anti-dropping devices, wherein the single-stage target detection model is composed of a backbone network and a selective visual attention mechanism module, wherein the backbone network is composed of multiple convolutional layers and pooling layers; then inputting the above training set into the single-stage target detection model for training, and then using the validation set and the test set to evaluate and test the trained single-stage target detection model, respectively, to obtain a hook anti-dropping device detection model; 3) Deploy the above-mentioned hook anti-dropping device detection model into a camera at the work site, and connect the camera to the deployment platform; 4) Using a camera to collect videos of the crane, hook, and hook anti-drop device, and then extracting frames from the videos to obtain JPEG images as images to be detected; 5) The above-mentioned image to be detected is input into the hook anti-dropping device detection model for feature extraction to obtain a feature map of the image, and then the feature map is used to detect the hook and the hook anti-dropping device as the target to determine whether the hook anti-dropping device is detected.

2. The detection method of the hook anti-dropping device according to claim 1, characterized in that: In step 1), the offline collection of various images containing cranes, hooks and hook anti-dropping devices is followed by preprocessing and annotation to obtain an image sample set, and then the specific steps of dividing the image sample set into a training set, a validation set and a test set in proportion are as follows: 1.1) Use image acquisition equipment to capture various images containing cranes, hooks and hook anti-drop devices; 1.2) Preprocess the above images including denoising, rotation, color gamut conversion, cropping and scaling to improve the quality and consistency of the images and expand the sample data; 1.3) Manually annotate the preprocessed images, mark the center coordinates and width and height of the crane, hook and hook anti-drop device image frame, use the annotated images as image samples, and form an image sample set with all image samples. Then divide the image sample set into training set, validation set and test set in a ratio of 7:1:2 for model training, evaluation and testing.

3. The detection method of the hook anti-dropping device according to claim 1 is characterized in that: In step 2), the YOLO or SSD model is selected as the backbone network; the selective visual attention mechanism module includes SENet and CBAM; the training method is to first compare the output of the single-stage target detection model with the annotation and calculate the loss function; then, the single-stage target detection model is trained using the back gradient propagation algorithm, and the model weights are updated using the stochastic gradient descent optimization algorithm; Use the validation set to evaluate the performance of the trained single-stage object detection model, calculate indicators including accuracy, recall, and mean average precision (mAP), and tune the model based on the evaluation results; Use the test set to test the trained model and test its performance on unseen data.

4. The detection method of the hook anti-dropping device according to claim 1, characterized in that: In step 3), the deployment platform includes a local server, a cloud platform or an embedded device.

5. The detection method of the hook anti-dropping device according to claim 1, characterized in that: In step 4), the method of using a camera to collect a video of a crane, a hook, and a hook anti-drop device, and then performing frame extraction on the video to obtain a JPEG image as an image to be detected is: Using FFmpeg video frame extraction technology, the video of the crane, hook and hook anti-dropping device captured by the camera is extracted with a certain frame interval to obtain a JPEG image with a high compression rate.

6. The detection method of the hook anti-dropping device according to claim 1, characterized in that: In step 5), the method of inputting the above-mentioned image to be detected into the hook anti-off device detection model for feature extraction to obtain a feature map of the image, and then detecting the hook and the hook anti-off device as the target on the feature map to determine whether the hook anti-off device is detected is: The image to be detected is input into the backbone network of the hook anti-dropping device detection model, and the feature map of the image is extracted by using the forward propagation process; Then, target detection is performed on the feature map. During the target detection process, each position in the feature map needs to be classified and regressed to identify the hook and the hook anti-dropping device, and obtain the center point, width and height of the target frame of the hook and the hook anti-dropping device, a total of four coordinate values. Afterwards, the intersection-and-union ratio of the hook and the hook anti-dropping device is calculated using the above four coordinate values, and a threshold is pre-set according to the degree of overlap between the two targets of the hook and the hook anti-dropping device. If the intersection-and-union ratio is less than the threshold, it is deemed that the hook anti-dropping device is not detected; otherwise, it is deemed that the hook anti-dropping device is detected.

7. The detection method of the hook anti-dropping device according to claim 6, characterized in that: The method for calculating the intersection and union ratio of the hook and the hook anti-dropping device by using the center point and width and height coordinate values ​​of the target frame of the hook and the hook anti-dropping device is as follows: 5.1) Calculate the intersection of the two target frames of the hook and the hook anti-drop device, that is, the area of ​​the overlapping part: Calculate the coordinates of the upper left corner and lower right corner of the two target boxes; Calculate the areas of the two target boxes based on the coordinates of the upper left corner and the lower right corner; According to the area of ​​the above target box, the coordinates of the upper left corner and the lower right corner of the intersection are calculated, and then the area of ​​the intersection is calculated; 5.2) Calculate the area of ​​the union of the two target boxes: According to the coordinates of the upper left corner and lower right corner of the two target boxes, calculate the area of ​​the union; 5.3) Using the area of ​​the intersection and the area of ​​the union, calculate the intersection-union ratio: Divide the area of ​​the intersection by the area of ​​the union to get the intersection-union ratio; Then, it is determined whether the hook anti-dropout device is detected based on the above-mentioned intersection-and-combination ratio. If the intersection-and-combination ratio is less than the threshold value, it means that the hook anti-dropout device is smaller than the hook, and it is regarded that the hook anti-dropout device is not detected; otherwise, it is regarded that the hook anti-dropout device is detected, and an alarm is immediately issued.

Images