An elevator system and a control method, device therefor

Abstract

An elevator system and a control method and device thereof, the elevator system comprising: an elevator car, an image acquisition module arranged in the elevator car, wherein the image acquisition module is configured to acquire an image in the car; the control method of the elevator system comprising: obtaining space occupancy information in the elevator car by using the image acquired by the image acquisition module, and determining whether the obtained space occupancy information conforms to preset space occupancy information; and in response to a determination result that the preset space occupancy information is conformed to, shielding an external call signal. In the case that the occupancy information in the elevator car conforms to the preset space occupancy information, the external call signal can be shielded, thereby saving dispatching resources.

Description

Technical Field

[0001] This article relates to elevator control technology, and more particularly to an elevator system and its control method and device. Background Technology

[0002] With the widespread application of elevator technology, users' demand for intelligent elevator operation is increasing. However, the existing control methods only block external call signals when overload is detected, which is insufficient to cope with the complex and ever-changing real-world elevator riding conditions. Summary of the Invention

[0003] This application provides an elevator system and its control method and device, which can save scheduling resources and more efficiently cope with complex and ever-changing actual elevator riding scenarios.

[0004] The elevator system control method provided in this disclosure is applied to an elevator system, which includes: an elevator car, and an image acquisition module disposed within the elevator car, wherein the image acquisition module is used to acquire images within the car; the elevator system control method includes:

[0005] The image acquisition module is used to acquire images to obtain information on the space occupancy inside the elevator car, and it is determined whether the obtained space occupancy information matches the preset space occupancy information.

[0006] In response to the judgment result that conforms to the preset space occupancy information, the external call signal is blocked.

[0007] In some exemplary embodiments, obtaining information on the space occupancy inside the elevator car using images acquired by the image acquisition module includes:

[0008] The images acquired by the image acquisition module are binarized, and the space occupancy information inside the elevator car is obtained based on the binarization result.

[0009] Alternatively, the images acquired by the image acquisition module can be input into a pre-trained space occupancy recognition model to obtain information on the space occupancy inside the elevator car.

[0010] In some exemplary embodiments, the step of binarizing the image acquired by the image acquisition module and obtaining the space occupancy information inside the elevator car based on the binarization result includes:

[0011] Gaussian blurring and threshold segmentation are applied to the images acquired by the image acquisition module to binarize the images into foreground and background regions; wherein the foreground region refers to the area occupied by passengers or goods, and the background region refers to the area without passengers or goods.

[0012] Obtain the number of pixels in the foreground region and calculate the ratio of the number of pixels in the foreground region to the total number of pixels in the image to obtain the pixel occupancy rate;

[0013] Based on the pre-established correspondence between pixel occupancy rate and space occupancy information, obtain the space occupancy information corresponding to the obtained pixel occupancy rate.

[0014] In some exemplary embodiments, the space occupancy recognition model is trained in the following manner:

[0015] Obtain the set of images inside the elevator car used to construct the space occupancy recognition model, as well as the space occupancy information labeled in each image of the set of images inside the elevator car;

[0016] Using the obtained set of images inside the elevator car and the labeled information on space occupancy, a pre-built machine learning model is trained to obtain the space occupancy recognition model.

[0017] In some exemplary embodiments, the step of training a pre-built machine learning model using the obtained set of images inside the elevator car and the labeled space occupancy information to obtain the space occupancy recognition model includes:

[0018] The obtained set of elevator car interior images is divided into a first set of elevator car interior images for training and a second set of elevator car interior images for verification.

[0019] The images in the first elevator car image set are used as input, and the space occupancy information marked on the images in the first elevator car image set is used as labels. A pre-built machine learning model is trained, and the parameters of the machine learning model are iteratively optimized through the backpropagation algorithm until the preset conditions are met to obtain the initial model.

[0020] The images in the second elevator car image set are input into the initial model to predict the space occupancy information corresponding to the images in the second elevator car image set. The prediction accuracy of the initial model is calculated based on the predicted space occupancy information and the labeled space occupancy information. If the prediction accuracy meets a preset threshold, the obtained initial model is used as the space occupancy recognition model.

[0021] In some exemplary embodiments, obtaining the elevator car interior image set for constructing the space occupancy recognition model includes:

[0022] Multiple images of the elevator car are acquired, and at least one of the following processes is performed on the acquired images: deleting blurry images, deleting images in which the image acquisition module is blocked, deleting images with uneven lighting, and deleting images with specular reflections.

[0023] The remaining images are used as the image set inside the elevator car.

[0024] In some exemplary embodiments, the space occupancy information includes: the degree of space occupancy;

[0025] When the space occupancy information is a degree of space occupancy, determining whether the obtained space occupancy information matches the preset space occupancy information includes:

[0026] Determine whether the obtained space occupancy level is higher than the preset space occupancy level.

[0027] In some exemplary embodiments, the elevator system further includes image acquisition modules respectively disposed on each floor, wherein the image acquisition modules are used to acquire images of a first predetermined area outside the car door on the corresponding floor; the control method of the elevator system further includes:

[0028] In response to the judgment result that does not conform to the preset space occupancy information, it is determined that the space occupancy of the elevator car is not higher than the preset space occupancy. The image collected by the image acquisition module set on the first floor where no elevator call request has been received is obtained, and based on the collected image, it is determined whether there is a situation on the first floor where it is inconvenient to manually call the elevator.

[0029] In response to the determination that there is a situation on any first floor where it is inconvenient to manually call the elevator, an automatic elevator call is triggered, the floor information of the first floor is obtained, and the elevator car is controlled to run to the corresponding first floor according to the obtained floor information, wherein the first floor includes one or more floors of arbitrary height.

[0030] The control device for an elevator system provided in this disclosure is applied to an elevator system. The control device includes a memory and a processor, wherein the memory is configured to store an executable program.

[0031] The processor is configured to read the executable program and execute the elevator system control method as described above.

[0032] The elevator system provided in this embodiment includes an elevator car, an image acquisition module installed inside the elevator car, image acquisition modules installed on each floor outside the elevator car, and a control device for the elevator system as described above, which is electrically connected to the elevator car and each of the image acquisition modules.

[0033] Compared with related technologies, the elevator system and its control method and device provided in this application can block external call signals when the occupancy information in the elevator car matches the preset space occupancy information, so that the elevator no longer responds to external call requests, thereby saving scheduling resources and more efficiently coping with complex and ever-changing actual elevator riding scenarios.

[0034] Other features and advantages of this application will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the application. Other advantages of this application can be realized and obtained by means of the embodiments described in the description and the accompanying drawings. Attached Figure Description

[0035] The accompanying drawings are used to provide an understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0036] Figure 1 This is a flowchart illustrating a control method for an elevator system according to an embodiment of this application;

[0037] Figure 2 This is a schematic diagram of a process for obtaining information on the space occupancy inside an elevator car, according to an embodiment of this application.

[0038] Figure 3 This is a schematic diagram of a process for obtaining a space occupancy recognition model according to an embodiment of this application;

[0039] Figure 4 This is a schematic diagram of the structure of a control device for an elevator system according to an embodiment of this application;

[0040] Figure 5 This is a schematic diagram of the structure of an elevator system according to an embodiment of this application. Detailed Implementation

[0041] This application describes several embodiments, but these descriptions are exemplary and not limiting, and it will be apparent to those skilled in the art that many more embodiments and implementations are possible within the scope of the embodiments described herein. Although many possible combinations of features are shown in the drawings and discussed in the detailed description, many other combinations of the disclosed features are also possible. Unless specifically limited, any feature or element of any embodiment may be used in combination with, or may replace, any feature or element of any other embodiment.

[0042] This application includes and contemplates combinations of features and elements known to those skilled in the art. The embodiments, features, and elements disclosed in this application can also be combined with any conventional features or elements to form unique inventive solutions. Any feature or element of any embodiment can also be combined with features or elements from other inventive solutions to form another unique inventive solution. Therefore, it should be understood that any feature shown and / or discussed in this application can be implemented individually or in any suitable combination. Therefore, the embodiments are not limited except by the limitations imposed by the appended claims and their equivalents. Furthermore, various modifications and changes can be made within the scope of the appended claims.

[0043] Furthermore, in describing representative embodiments, the specification may have presented methods and / or processes as a specific sequence of steps. However, the method or process should not be limited to the specific order of steps described herein, to the extent that it does not depend on such a specific order. As will be understood by those skilled in the art, other sequences of steps are also possible. Therefore, the specific order of steps set forth in the specification should not be construed as a limitation of the claims. Moreover, the claims concerning the method and / or process should not be limited to the steps performed in the written order, and those skilled in the art will readily understand that these orders can be varied and still remain within the spirit and scope of the embodiments of this application.

[0044] This disclosure provides a control method for an elevator system, applied to an elevator system, the elevator system including: an elevator car, and an image acquisition module disposed within the elevator car, wherein the image acquisition module is used to acquire images within the elevator car; such as Figure 1 As shown, the control method of the elevator system includes:

[0045] Step 100: Use the image acquisition module to acquire the space occupancy information inside the elevator car, and determine whether the acquired space occupancy information matches the preset space occupancy information;

[0046] Step 101: In response to the judgment result that matches the preset space occupancy information, block the external call signal.

[0047] For example, the image acquisition module can be a high-definition camera or a depth vision sensor (such as an RGB-D camera), installed on the top or corner of the car, to capture images or video streams inside the car in real time.

[0048] For example, space occupancy information can cover multiple dimensions to comprehensively assess the actual available space within the elevator car, thereby enabling intelligent decision-making regarding elevator operation. For instance, the number of passengers can be used as a criterion. When using passenger numbers as space occupancy information, image recognition technology can monitor the actual number of passengers in the car in real time. When the number of passengers reaches or exceeds the rated passenger capacity, a mechanism to automatically trigger the blocking of external call signals is activated. Furthermore, space occupancy information can also be judged based on the dimension of item occupancy. For example, image recognition technology can detect whether there are large luggage, wheelchairs, strollers, or other items in the car, which may occupy key passageways or localized space. The location of these items is located using a semantic segmentation model, and the remaining available space is analyzed using a human body distribution algorithm. If the detection of item occupancy significantly reduces the effective standing / passage area, it is determined to be "space-constrained," triggering a mechanism to block external call signals.

[0049] In addition, the uniformity of spatial distribution is also an important evaluation dimension. Even if the total number of people and the volume of items do not exceed the threshold, if passengers are concentrated in one place (such as all against the wall or concentrated on one side), the space on the other side may be idle but cannot be effectively utilized. Therefore, space occupancy information can also refer to the space occupancy information of a certain area inside the elevator car, that is, the judgment is based on the space occupancy information of a certain area inside the elevator car. For example, if the passenger distribution is detected to cause an abnormally high density in a local area (such as the proportion of people standing on one side exceeding 70%), it is judged as "functionally restricted", triggering the mechanism of blocking external call signals.

[0050] Furthermore, the dynamic adaptation of space occupancy information can be closely integrated with operational needs under different time windows to achieve flexibility and accuracy in elevator scheduling strategies. Taking passenger density per unit area as an example, during peak hours (such as morning and evening commuting peaks), a higher passenger density per unit area (e.g., 0.6-0.7 people / ㎡) can be used as the judgment standard. When image recognition determines that the preset passenger density per unit area has been reached, it is judged as "functionally restricted," triggering the mechanism to block external call signals. During off-peak hours (such as lunch breaks or late at night), a lower passenger density per unit area is used as the judgment standard. When image recognition determines that the preset passenger density per unit area has been reached, it is judged as "functionally restricted," triggering the mechanism to block external call signals.

[0051] To block external call signals, a judgment step can be added to the elevator control logic. This judgment step could be to check whether a "blocking flag" exists. Specifically, when an external call signal is received, the system first checks whether a "blocking flag" exists. If a "blocking flag" exists, the signal is discarded and not added to the dispatch queue. If no "blocking flag" exists, the request is processed normally and added to the dispatch queue.

[0052] The elevator system control method provided in this application embodiment can block external call signals when the occupancy information in the elevator car matches the preset space occupancy information, so that the elevator no longer responds to external call requests, thereby saving scheduling resources and coping with complex and ever-changing actual elevator riding scenarios more efficiently.

[0053] In some exemplary embodiments,

[0054] The space occupancy information includes: the degree of space occupancy;

[0055] When the space occupancy information is a degree of space occupancy, determining whether the obtained space occupancy information matches the preset space occupancy information includes:

[0056] Determine whether the obtained space occupancy level is higher than the preset space occupancy level.

[0057] For example, the degree of space occupancy can be described from two dimensions: occupancy level and comfort state. When describing the degree of space occupancy from the dimension of occupancy level, the occupancy level from high to low can include: high occupancy level, medium occupancy level, and low occupancy level. When describing the degree of space occupancy from the dimension of comfort state, the comfort state from high to low can include: crowded, average, and comfortable.

[0058] Furthermore, space occupancy can also be quantified using space occupancy rate. When the space occupancy rate is obtained through the image acquisition module (e.g., the actual occupancy rate of the car calculated based on image semantic segmentation, such as 65%), a simple numerical comparison can be performed directly to determine whether the current occupancy rate is greater than a preset threshold (e.g., 60%). If it exceeds the threshold, it is determined to be "functionally limited," and external call signals are automatically blocked to ensure operational efficiency. Conversely, when the space occupancy level is obtained through the image acquisition module, it can be checked whether the current space occupancy level (e.g., "crowded") is higher than a preset space occupancy level (e.g., "moderate"). If it is higher than the preset space occupancy level, it is determined to be "functionally limited," and external call signals are automatically blocked to ensure operational efficiency.

[0059] In some exemplary embodiments, obtaining information on the space occupancy inside the elevator car using images acquired by the image acquisition module includes:

[0060] The images acquired by the image acquisition module are binarized, and the space occupancy information inside the elevator car is obtained based on the binarization result.

[0061] Alternatively, the images acquired by the image acquisition module can be input into a pre-trained space occupancy recognition model to obtain information on the space occupancy inside the elevator car.

[0062] For example, the images acquired by the image acquisition module are binarized. This process simplifies the complex scene in the original image into a binary state of "occupied" and "unoccupied," significantly improving the efficiency and accuracy of subsequent analysis. Specifically, the image is first converted to grayscale to eliminate color interference. Then, based on a preset threshold, the image is divided into foreground (e.g., passengers, luggage, obstacles) and background (e.g., elevator walls, floor). This binarized image can quickly highlight the occupancy status of key areas, such as whether the doorway is obstructed or whether passengers are densely distributed, thus providing the system with an intuitive basis for judging space occupancy.

[0063] For example, utilizing a space occupancy recognition model is another alternative method to obtain information on space occupancy within an elevator car. This model is typically trained on a deep learning framework (such as YOLO, Faster R-CNN, or U-Net) and can directly extract key space occupancy features from raw images. Specifically, the acquired images (which may contain scenes of passengers, luggage, obstacles, etc.) are input into the model, and semantic information is extracted layer by layer through a Convolutional Neural Network (CNN). For instance, the model can identify the number of passengers, the distribution of large luggage, whether key paths (such as doorways and handrails) are obstructed, and the spatial distribution of crowd density. Its output typically includes structured space occupancy information, such as passenger counts, heatmaps of occupied areas, and critical path occupancy rates, which can directly serve as the basis for elevator control system decisions.

[0064] In terms of technical implementation, a deep convolutional neural network (CNN) is used as the core algorithm, and the widely validated and stable ResNet (residual network) architecture is selected as the basic model. The specific process includes:

[0065] Transfer learning and fine-tuning: In the PyTorch deep learning framework, transfer learning is performed on a pre-trained ResNet model (trained on the ImageNet dataset) to fully utilize its learning capabilities on general image features.

[0066] Scene adaptation optimization: Fine-tuning is performed on specific scene data of elevator cars (such as passenger standing patterns and item distribution patterns) to significantly improve the model's recognition ability in small samples and specific scenarios.

[0067] Multi-task joint modeling: Combining object detection (e.g., identifying passengers and items) with semantic segmentation (e.g., dividing occupied areas), by sharing the underlying feature extraction layer, it outputs both quantitative indicators (e.g., space occupancy rate) and qualitative judgments (e.g., "crowded").

[0068] In some exemplary embodiments, such as Figure 2 As shown, the step of binarizing the image acquired by the image acquisition module and obtaining the space occupancy information inside the elevator car based on the binarization result includes:

[0069] Step 200: Apply Gaussian blur processing and threshold segmentation to the image acquired by the image acquisition module to binarize the image into a foreground region and a background region; wherein the foreground region refers to the area occupied by passengers or goods, and the background region refers to the area not occupied by passengers or goods.

[0070] For example, performing Gaussian blurring and thresholding on the raw image acquired by the image acquisition module is a fundamental step in extracting spatial occupancy information. Gaussian blurring eliminates noise in the image (such as sensor noise and local interference caused by uneven lighting), smooths image details, and provides a more stable input for subsequent segmentation. Specifically, this can be achieved by using a weighted average of the image through convolutional kernels to make the gray values ​​of adjacent pixels more consistent. For instance, when the gray values ​​of passenger outlines inside the car fluctuate due to backlighting or shadows, Gaussian blurring can reduce the impact of these fluctuations on the segmentation results. Thresholding segmentation sets a gray value threshold to divide the image into foreground (i.e., passenger / cargo occupied areas) and background (i.e., unoccupied areas). For example, if a pixel's gray value is higher than the threshold, it is marked as foreground; otherwise, it is marked as background. However, in complex lighting environments (such as strong backlighting), a fixed threshold may lead to segmentation failure. Therefore, adaptive thresholding segmentation (such as the Otsu algorithm) can be used to automatically calculate the optimal threshold based on the overall gray value distribution of the image, improving robustness.

[0071] Step 201: Obtain the number of pixels in the foreground region and calculate the ratio of the number of pixels in the foreground region to the total number of pixels in the image to obtain the pixel occupancy rate.

[0072] For example, the pixel occupancy rate (POR) is obtained by statistically analyzing the ratio of the number of pixels in the foreground region to the total number of pixels in the image. This POR serves as an indicator of spatial occupancy; a higher POR indicates a larger occupied area within the elevator car. The formula for calculating the pixel occupancy rate is as follows:

[0073] Pixel occupancy rate = (Number of foreground pixels / Total number of pixels in the image) × 100%.

[0074] Step 202: Based on the pre-established correspondence between pixel occupancy rate and space occupancy information, obtain the space occupancy information corresponding to the obtained pixel occupancy rate.

[0075] For example, when space occupancy is quantified using space occupancy rate, the corresponding space occupancy rate can be obtained based on the obtained pixel occupancy rate and the relationship between pixel occupancy rate and space occupancy rate. The relationship between pixel occupancy rate and space occupancy rate can be achieved through steps such as calibration mapping, formula conversion, and dynamic correction. First, based on the resolution, field of view, and actual car size of the image acquisition module, calibration images are captured by laying out a calibration grid, and the actual area corresponding to a single pixel is calculated. Then, this mapping relationship is used to convert the pixel occupancy rate (i.e., the proportion of foreground pixels) into a space occupancy rate. Furthermore, perspective correction is used to eliminate image distortion caused by the camera installation angle (e.g., edge stretching of a wide-angle lens), resulting in errors. In addition, for different elevator types (e.g., the size difference between residential elevators and hospital elevators), separate calibration and adaptation to scene requirements are required.

[0076] In some exemplary embodiments, the space occupancy recognition model is trained in the following manner:

[0077] Obtain the set of images inside the elevator car used to construct the space occupancy recognition model, as well as the space occupancy information labeled in each image of the set of images inside the elevator car;

[0078] Using the obtained set of images inside the elevator car and the labeled information on space occupancy, a pre-built machine learning model is trained to obtain the space occupancy recognition model.

[0079] For example, the training process of a space occupancy recognition model needs to be systematically designed from data collection and annotation standardization to model optimization to ensure its robustness and practicality in complex scenarios. First, high-definition cameras deployed inside the elevator car continuously collect images from multiple scenes, covering different time periods (e.g., peak / off-peak hours), different passenger distribution patterns (e.g., single person, multiple people, carrying luggage), and environmental interference (e.g., strong backlight, smoke). To improve data diversity, data augmentation techniques (e.g., rotation, flipping, brightness adjustment) can be introduced to simulate more real-world conditions. Each image needs corresponding annotation information on space occupancy, including quantitative annotations (when space occupancy is quantified as a percentage, e.g., 33.28%) and qualitative annotations (e.g., "free," "comfortable," "crowded"). The latter can be manually assessed based on passenger density, standing patterns, and item distribution.

[0080] During the model training phase, a suitable machine learning model is selected based on the task requirements. For example, semantic segmentation models (such as U-Net and DeepLabV3) can directly output segmentation maps of the foreground and background, facilitating the calculation of space occupancy; object detection models (such as YOLOv5 and Faster R-CNN) can identify the number of passengers and large objects, and estimate the total occupancy rate by combining the preset single-person occupancy area; classification models (such as ResNet and EfficientNet) output qualitative space occupancy. Simultaneously, transfer learning and fine-tuning techniques are used to optimize elevator car image data based on the pre-trained model, accelerating convergence and improving performance in small sample scenarios.

[0081] In terms of model architecture and training process, ResNet is used as the backbone network, and end-to-end training is performed on a large-scale GPU cluster, utilizing CUDA to accelerate computation. To address the imbalanced sample problem, a class-weighted loss function (such as FocalLoss) is introduced to assign higher weights to the minority class.

[0082] After the trained space occupancy recognition model is deployed to the production environment, it can receive image streams transmitted from cameras inside the elevator in real time via edge computing devices (such as embedded GPUs) and quickly output the predicted car load status (e.g., "empty", "half-loaded", "fully loaded") under low latency. The model's prediction results are interfaced with the elevator central control system via the CAN bus protocol to achieve data linkage and provide dynamic decision-making basis for the central control system. To ensure the efficient operation of the model on edge devices, model pruning (e.g., channel pruning) or quantization techniques are used during the deployment phase to reduce the computational load. For example, quantizing and optimizing the ResNet model using the TensorRT framework can improve inference speed by several times while keeping accuracy loss within a controllable range.

[0083] In terms of privacy protection, multiple layers of protection mechanisms can be implemented: using GAN to generate virtual faces or Gaussian blur algorithms to mask the facial areas in the image to prevent the leakage of passenger identity information; at the same time, the data transmission process is encrypted through encryption protocols to ensure secure interaction between the image and the prediction results between the edge device and the central control system.

[0084] Through the above deployment strategy, the model not only achieves efficient inference and real-time response, but also takes into account data privacy and system security, providing accurate and reliable technical support for intelligent elevator control.

[0085] In some exemplary embodiments, such as Figure 3 As shown, the step of training a pre-built machine learning model using the obtained set of images inside the elevator car and the corresponding space occupancy information to obtain the space occupancy recognition model includes:

[0086] Step 300: Divide the obtained elevator car interior image set into a first elevator car interior image set for training and a second elevator car interior image set for verification.

[0087] Step 301: Take the images in the first elevator car image set as input, take the space occupancy information marked in the images in the first elevator car image set as labels, train the pre-built machine learning model, and iteratively optimize the parameters of the machine learning model through the backpropagation algorithm until the preset conditions are met to obtain the initial model.

[0088] Step 302: Input the images in the second elevator car image set into the initial model to predict the space occupancy information corresponding to the images in the second elevator car image set. Calculate the prediction accuracy of the initial model based on the predicted space occupancy information and the labeled space occupancy information. If the prediction accuracy meets a preset threshold, use the obtained initial model as the space occupancy recognition model.

[0089] For example, the training process of a space occupancy recognition model requires a systematic data partitioning and iterative optimization strategy to ensure the model's generalization ability and stability in complex scenarios. First, the collected set of elevator car images is divided into a training set (i.e., the first set of elevator car images) and a validation set (i.e., the second set of elevator car images), typically using a 7:3 or 8:2 ratio to balance the needs of model learning and performance evaluation. The training set is used for model parameter learning; its input is the elevator car image, and the labels are the corresponding space occupancy information (e.g., qualitative labels such as "crowded" or "comfortable," or quantitative representations of space occupancy rates). The validation set is used to evaluate the model's performance on unseen data to avoid overfitting. Based on this, the model parameters are iteratively updated using the backpropagation algorithm: after the input image is processed by the model network and the predicted result is output, the difference between the predicted value and the true label is calculated (e.g., the mean squared error (MSE) for regression tasks or the cross-entropy loss for classification tasks), and the parameters are adjusted using gradient descent (e.g., the Adam optimizer) to gradually reduce the loss value. The training process continues until preset conditions are met (such as reaching the maximum number of training rounds, loss convergence, or validation set accuracy reaching the target).

[0090] In the validation phase, validation set images are input into the trained initial model, which outputs the predicted space occupancy and compares it with the ground truth annotations to calculate the prediction accuracy. If the accuracy meets a preset threshold, the initial model is confirmed as a space occupancy recognition model; otherwise, the model architecture needs to be adjusted (e.g., increasing the number of layers), hyperparameters need to be optimized (e.g., learning rate, batch size), or the dataset needs to be re-partitioned and retrained. Furthermore, to address challenges in real-world scenarios, model training needs to incorporate dynamic optimization strategies: for imbalanced data, oversampling or Focal Loss can be used to adjust class weights.

[0091] In some exemplary embodiments, obtaining the elevator car interior image set for constructing the space occupancy recognition model includes:

[0092] Multiple images of the elevator car are acquired, and at least one of the following processes is performed on the acquired images: deleting blurry images, deleting images in which the image acquisition module is blocked, deleting images with uneven lighting, and deleting images with specular reflections.

[0093] The remaining images are used as the image set inside the elevator car.

[0094] For example, to build a high-precision space occupancy recognition model, the acquisition of the elevator car image set needs to undergo a rigorous quality screening process to ensure the reliability and representativeness of the training data. First, multiple images are continuously acquired by an image acquisition module deployed inside the car, covering different time periods (e.g., peak / off-peak hours), passenger distribution patterns (e.g., single person, multiple people, carrying luggage), and environmental conditions (e.g., strong backlight, smoke) to simulate diverse real-world scenarios. Subsequently, the original images undergo at least one of the following processing steps:

[0095] Blurry image removal: Blurry images are automatically detected using image sharpness assessment algorithms (such as the Laplacian operator or frequency domain analysis). For example, if the gradient variance of an image is below a preset threshold, it is determined to be blurry and removed, avoiding misjudgment by the model due to image defocus.

[0096] Remove occluded images: Use an object detection model (such as YOLO) to identify whether the camera is obstructed (e.g., a passenger holding an object blocking the lens). If the detected obstruction area exceeds 30%, the image is excluded to ensure the integrity of the model's input screen.

[0097] Delete images with uneven lighting: Calculate the standard deviation of the image brightness. If the standard deviation is greater than a preset threshold, it is determined to be an image with uneven lighting.

[0098] Remove specular reflections from images: Detect brightness peaks in a preset specular area. If the peak exceeds a preset threshold and the edge sharpness is higher than the preset threshold, it is determined to be a specular reflection.

[0099] In some exemplary embodiments, the elevator system further includes image acquisition modules respectively installed on each floor, wherein the image acquisition modules are used to acquire images of a first predetermined area outside the car door on the corresponding floor; the control method of the elevator system further includes:

[0100] In response to the judgment result that does not conform to the preset space occupancy information, it is determined that the space occupancy of the elevator car is not higher than the preset space occupancy. The image collected by the image acquisition module set on the first floor where no elevator call request has been received is obtained, and based on the collected image, it is determined whether there is a situation on the first floor where it is inconvenient to manually call the elevator.

[0101] In response to the determination that there is a situation on any first floor where it is inconvenient to manually call the elevator, an automatic elevator call is triggered, the floor information of the first floor is obtained, and the elevator car is controlled to run to the corresponding first floor according to the obtained floor information, wherein the first floor includes one or more floors of arbitrary height.

[0102] In the above embodiments, when the image acquired by the image acquisition device installed inside the car determines that the space occupancy of the car is not higher than a preset space occupancy level, i.e., the car is not fully loaded, the image acquired by the image acquisition device installed outside the car can be further used to determine whether there is a floor where manual calling is inconvenient. If it is determined that there is a floor where manual calling is inconvenient, the elevator car is controlled to move to the corresponding floor based on the obtained floor information. In this way, the determination of whether manual calling is inconvenient and the automatic calling are performed only when the elevator is not fully loaded, which improves the accuracy of calling and achieves energy saving of the elevator.

[0103] For example, the image acquisition module installed on each floor can be a camera, but is not limited to ordinary surveillance cameras. Its specific type and function can be selected and optimized according to the actual application scenario. For instance, the image acquisition module can use a high-definition visible light camera to clearly capture the human form and behavior in the waiting area under normal lighting conditions; it can also use an infrared camera or a low-light camera to ensure effective imaging in nighttime, dim, or backlit environments; in scenarios with higher requirements for spatial perception, a depth camera or RGB-D sensor can also be used to acquire information including distance, contour, and even three-dimensional pose, thereby more accurately determining whether a person is in a situation where it is inconvenient to manually call the elevator. In addition, the camera can integrate intelligent processing capabilities, with a built-in AI chip or edge computing module, to directly determine the situation where it is inconvenient to manually call the elevator on the device side.

[0104] For example, the first predetermined area can refer to the ground space directly opposite the elevator car door, where passengers most frequently linger while waiting for the elevator. This area can be determined based on pedestrian flow analysis, actual usage habits, or building layout. Taking a construction site scenario as an example, due to the transportation needs of building materials, elevator cars are typically configured with oversized dimensions of 1.6m × 2.2m or more, and the car door width can reach 3.2m. Therefore, the first predetermined area is also correspondingly large, ensuring that one side of the first predetermined area at least covers the width of the car door; for example, it could be set to 3.2m × 3m.

[0105] The images used to analyze whether there are situations where manual elevator calls are inconvenient are not those captured by all image acquisition modules, but only those captured by the image acquisition modules set up on floors that have not received elevator call requests. Floors that have not received elevator call requests can refer to floors that have not received external elevator call requests, but not internal ones. Floors that have not received external elevator call requests refer to floors where neither the up nor down elevator call buttons have been pressed by passengers; these floors are referred to as "first floors." For each first floor, images captured by its corresponding image acquisition module are obtained. These images can be acquired in real time or periodically according to a set cycle. Alternatively, images captured by the image acquisition modules of all floors can be obtained first, and then the images belonging to the first floor can be selected from them.

[0106] Once image analysis determines that any first floor is inconvenient for manual call, an automatic call mechanism for that floor is immediately triggered. This mechanism eliminates the need for passengers to actively press the call button; instead, the elevator control logic autonomously generates a call signal equivalent to a manual press. Subsequently, the floor information, such as the floor number (e.g., "3rd floor"), physical location identifier, or communication address, is acquired to determine the elevator car's target stop. The first floor information is usually pre-bound to the installation location of the image acquisition module to ensure a one-to-one correspondence between the identification result and the floor, avoiding misjudgments or misassignment. Finally, based on the acquired floor information, the elevator car is dispatched and runs to the corresponding first floor: if a spare or nearby car is available, it is given priority; if multiple elevators are available, the optimal car is selected based on factors such as operating status, direction, and load. Upon arrival, the elevator doors open automatically, providing timely and seamless service for passengers with difficulty operating the buttons. If it is determined that there are multiple first floors where it is "inconvenient to manually call the elevator", a call signal can be generated for each of these multiple first floors, and the elevator car can arrive at these multiple first floors in sequence according to the floor order.

[0107] For example, while the elevator car has reached the corresponding first floor and the elevator door is open and waiting for passengers to enter, an image acquisition module can be used to acquire images and determine the distance between the passenger or goods and the elevator door during the passenger's entry process. It can also determine whether the elevator door will collide with the passenger or goods when it closes. If it is determined that the passenger or goods will collide, the closing of the elevator door is delayed.

[0108] In some exemplary embodiments, determining whether there is a situation on the first floor where it is inconvenient to manually call the elevator based on the acquired images includes:

[0109] Passenger and cargo identification are performed based on images captured by the image acquisition module located on the first floor.

[0110] When both passengers and goods are detected simultaneously, the movement trajectories of the detected passengers and goods are obtained from the images subsequently captured by the image acquisition module set up on the first floor.

[0111] In response to the identified movement trajectories of passengers and goods both moving towards the car door, it is determined that there is a situation where it is inconvenient to manually call the elevator.

[0112] For example, in passenger identification, object detection technology can be used to locate human bodies in images, and behavioral analysis can be combined to determine whether they are waiting for an elevator. For instance, it not only identifies the presence of a person but also continuously tracks their dynamic characteristics in the image, such as changes in position, direction of movement, duration of stay, and body orientation. For example, if someone lingers in a designated area outside the elevator doors for an extended period, with their body roughly facing the elevator doors, or steadily walks towards the elevator doors from a distance, they can be considered to have a clear intention to use the elevator. Conversely, if someone quickly passes through the area, has their back to the elevator doors, or only stays briefly before leaving, they are more likely to be passing by or not using the elevator. Furthermore, multi-frame temporal information can be combined to analyze the continuity and purposefulness of the movement trajectory, further improving the accuracy of the judgment.

[0113] For cargo recognition, a similar technical approach to human detection can be adopted. Object detection algorithms can be used to locate salient, non-human objects in images, and instance segmentation or semantic segmentation methods can be used to further extract their contours and region information. Based on this, the positional changes of these objects across multiple frames are continuously tracked, and their motion trajectories, speeds, dwell times, and relative relationships with the surrounding environment or other objects are analyzed to determine their state characteristics—for example, whether they are stationary, moving synchronously with people, or existing independently and moving autonomously.

[0114] For example, when a passenger and goods are detected simultaneously in an image of a first floor, no immediate judgment is made. Instead, subsequent images acquired by the image acquisition module for that floor are used to track the identified passenger and goods across frames, thereby obtaining their respective movement trajectories. Based on this, the movement trends of these two types of objects are analyzed: if the passenger's trajectory shows that they are moving towards the elevator door, and the identified goods also exhibit synchronous movement characteristics towards the door (e.g., both maintain a stable relative position and move in the same direction), it can be inferred that the passenger is likely carrying or manipulating the goods towards the elevator. This cooperative movement pattern often indicates a clear intention to use the elevator and an actual need to do so. Although the passenger has not yet pressed the call button, the scenario of "person + object" moving towards the door together suggests that there is a situation where manually calling the elevator is inconvenient. This is not based on direct assumptions about the passenger's physical condition or the type of goods, but rather a reasonable inference based on the consistency and purposefulness of their overall behavior.

[0115] For example, in the process of judging the movement trajectory, the movement speed can also be taken into consideration. That is, the identified movement trajectories of passengers and goods must not only meet the condition of moving towards the car door in a uniform manner, but also meet the condition of moving at the same speed, in order to be judged as a situation where it is inconvenient to manually call the elevator.

[0116] In other embodiments, other rules can be used to determine whether there is a situation where it is inconvenient to call the elevator, such as when a passenger is holding goods and standing in front of the elevator door for more than a certain period of time.

[0117] In some exemplary embodiments, before obtaining the identified passenger and cargo movement trajectories from images subsequently acquired by the image acquisition module located on the first floor, the method further includes:

[0118] The outlines of the passenger and the cargo are detected, and the type of the passenger's handling behavior towards the cargo is obtained based on the detected outlines of the passenger and the cargo.

[0119] The step of obtaining the identified passenger and cargo movement trajectories based on images subsequently acquired by the image acquisition module set up on the first floor includes:

[0120] In response to the judgment result that the passenger's handling behavior of the goods belongs to a preset handling behavior type, the movement trajectory of the identified passenger and goods is obtained based on the images subsequently collected by the image acquisition module set on the first floor.

[0121] For example, the elevator control method provided in this application focuses not only on "whether there are people" or "whether there are objects," but also on "what is happening between people and objects." By accurately extracting the outlines of passengers and goods, that is, their outer boundary shapes and spatial occupancy areas in the image, the relative layout of the two in two-dimensional or even three-dimensional space can be constructed. This outline information is the basis for understanding physical interactions: for example, when a part of the passenger outline is closely fitted with the goods outline, forming a stable contact surface, and the two are synchronously displaced in consecutive frames, it may indicate that the passenger is pushing, supporting, or holding the object; if the goods outline is always located in front of / behind the passenger outline and maintains a fixed distance, it is more likely to be a traction or guiding behavior; and if the goods outline is completely inside or above the passenger outline, it may correspond to actions such as lifting or carrying.

[0122] After identifying the type of passenger's handling behavior, further dynamic tracking and behavior verification steps are taken. Specifically, when contour analysis determines that the passenger's handling behavior belongs to a predefined preset handling behavior type, the next step is taken: using images continuously acquired by the first-floor image acquisition module thereafter, the identified passenger and goods are continuously tracked to obtain their movement trajectories in space. These trajectories reflect their positional changes, direction of travel, speed, and relative motion relationship over a period of time. The significance of this design is that a single momentary handling posture is insufficient to fully confirm the intention to use the elevator or the degree of inconvenience. Through dynamic verification of subsequent trajectories, it is possible to more reliably determine whether the passenger is actively moving towards the elevator door and whether the goods are moving synchronously with them. This two-stage strategy of "first identifying the behavior type, then verifying the movement trend" effectively improves the accuracy of judging actual elevator usage needs and avoids unnecessary automatic elevator calls triggered by static misjudgments or accidental stops.

[0123] In some exemplary embodiments, the preset carrying behavior type includes: a first carrying behavior type in which the passenger uses both hands to carry goods, and the outline of the passenger includes: the outline of the passenger's hands;

[0124] The step of obtaining the passenger's handling behavior type of the goods based on the detected outline of the passenger and the outline of the goods includes:

[0125] Determine whether the outlines of the passenger's hands at least partially overlap with the outline of the cargo;

[0126] In response to the determination that the outlines of the passenger's hands at least partially overlap with the outline of the goods, the passenger's handling behavior of the goods is determined to be the first handling behavior type.

[0127] For example, to accurately identify the type of passenger carrying behavior, it is necessary to focus more on the fine contours of their hands, which is a key basis for determining the type of carrying behavior. The extraction of hand contours usually combines human pose estimation and hand instance segmentation techniques. First, pose estimation algorithms (such as OpenPose, HRNet, or MediaPipe Pose) can be used to locate key points of the passenger's body, especially the left and right wrists and finger areas, thereby determining the approximate position of the hands in the image. Then, a hand segmentation model (such as a neural network based on U-Net or DeepLab architecture) is called within the estimated area to generate a high-precision pixel-level hand mask, ultimately obtaining clear and complete left and right hand contours.

[0128] Based on this, the spatial relationship between the outlines of the hands and the outline of the goods is further analyzed. Specifically, it is determined whether the outlines of the left and right hands overlap with the outline of the goods at least partially on the image plane. Here, "overlap" does not require complete coverage, but rather that in the two-dimensional projection, the areas of the hands and the area of ​​the goods intersect in spatial position, indicating that the passenger is likely grasping, lifting, hugging, or stably controlling the goods with both hands. In this case, the current handling behavior is classified as the first handling behavior type, namely the preset behavior pattern of "using both hands to handle goods".

[0129] In some exemplary embodiments, the passenger uses a second carrying behavior type of carrying large goods with one hand, and the passenger's profile includes: the profiles of the passenger's two hands; obtaining the passenger's carrying behavior type of the goods based on the detected profiles of the passenger and the goods includes:

[0130] Determine whether the outlines of the passenger's hands at least partially overlap with the outline of the cargo:

[0131] In response to the determination that the outline of one of the passenger's hands at least partially overlaps with the outline of the cargo, while the outline of the other hand does not overlap with the outline of the cargo, the outline area of ​​the cargo is calculated based on the outline of the cargo after the cargo enters the second predetermined area; wherein, the second predetermined area is a portion of the first predetermined area near the car door.

[0132] In response to the calculated outline area of ​​the goods being greater than a preset area threshold, the passenger's handling behavior type for the goods is determined to be the second handling behavior type.

[0133] For example, within the second predetermined area, the outline area of ​​the goods can be calculated more accurately, and the accuracy of the calculation of the outline area of ​​the goods is significantly improved compared to the area further away from the car door.

[0134] For example, when there is a spatial relationship in which the outline of one hand at least partially overlaps with the outline of the cargo on the image plane, while the outline of the other hand does not overlap with the outline of the cargo at all, it indicates that the passenger is likely to be operating the cargo using only one arm, such as supporting, pulling, or lifting it from the side.

[0135] However, simply relying on single-handed contact is insufficient to determine whether a situation constitutes "inconvenience in manually calling the elevator," as small items (such as handbags or folders) can also lead to similar behavior. Therefore, a more refined judgment mechanism is introduced: once the aforementioned single-handed handling state is initially identified, the movement trajectory of the goods is continuously tracked; once the goods enter a specific area, namely the second preset area (this area is a sub-area near the elevator car door in the first preset area, usually representing a key waiting position where passengers are about to enter the elevator car), a quantitative assessment of the goods' dimensions is initiated.

[0136] Specifically, based on the visual outline of the goods at this moment, its projected area in the image is calculated (scale normalization can be performed using camera intrinsics to improve the accuracy of area estimation). Then, this outline area is compared with a pre-set area threshold, which is set based on empirical data from real-world application scenarios to distinguish between ordinary personal items and larger items that may affect operational convenience. If the calculated goods outline area is greater than the pre-set threshold, the current scenario is determined to be a case of a passenger using one hand to carry large goods, i.e., it meets the pre-set "second carrying behavior type". This behavior pattern typically means that although only one hand is used, due to the large size of the goods, unstable center of gravity, or the need for continuous control, the passenger still finds it difficult to free up a hand to press the elevator call button.

[0137] In other embodiments, in addition to the first and second transport behavior types described above, other preset transport behavior types and their judgment rules may be included.

[0138] In some exemplary embodiments, the number of elevator cars is multiple, and the step of obtaining the floor information of the first floor and controlling the elevator cars to run to the corresponding first floor based on the obtained floor information includes:

[0139] Obtain the image acquisition module identifier corresponding to the image used to trigger automatic elevator call, and obtain the floor information of the first floor corresponding to it based on the mapping relationship between the image acquisition module identifier and floor information;

[0140] Based on the obtained floor information and the operating status of multiple elevator cars, select one elevator car from the multiple elevator cars that has the shortest travel time to the corresponding first floor, and control the selected elevator car to travel to the corresponding first floor.

[0141] For example, when the image acquisition module detects a situation on a first floor where manual elevator calling is inconvenient and triggers automatic elevator calling, it first needs to determine which floor issued the request. To do this, the identifier of the image acquisition module corresponding to the image that triggered the automatic elevator calling can be obtained (i.e., which floor's camera captured the scene requiring assistance). Each image acquisition module has a unique identifier, and this identifier has a preset mapping relationship with specific floor information. By querying this mapping relationship, the floor information of the corresponding first floor can be accurately obtained.

[0142] Once the specific floor from which the request was made (i.e., the first floor where manual calling is inconvenient) is determined, the next step is to select the most suitable elevator car from the existing fleet. To make the optimal choice, the following factors can be considered: floor information: the location of the target floor; and the operational status of the multiple elevator cars: including the current floor of each car, its direction of travel (up or down), and whether it is fully loaded. Based on this data, a scheduling algorithm can be used to calculate which elevator car requires the shortest travel time to reach the target floor.

[0143] The system selects the elevator car that can reach the designated floor in the shortest time and sends a command to it to adjust its route to that floor. This not only improves response speed but also optimizes the scheduling efficiency of the entire elevator system, reducing waiting time and resource waste.

[0144] In some exemplary embodiments, when the presence of both passengers and goods is simultaneously detected, the method further includes:

[0145] Obtain the relative positional relationship between the passenger and the cargo;

[0146] In response to the situation where the cargo is located on the side of the passenger closest to the car door, during the process of controlling the elevator car to run to the corresponding first floor, it is determined whether the cargo has a tendency to move closer to the car door in a third predetermined area; wherein, the third predetermined area is a portion of the first predetermined area that is closer to the car door;

[0147] In response to the cargo showing a tendency to move closer to the car door within the third predetermined area, the passenger is alerted in a preset manner that the cargo may collide with the car door.

[0148] For example, within the third predetermined area, by determining whether the goods have a tendency to move towards the car door, and issuing a warning when such a tendency exists, the risk of the goods colliding with the car door can be effectively avoided.

[0149] If it is determined that the goods are moving towards the car door, video recording for a preset duration (e.g., 15 seconds) can be triggered, and the collected image data will be stored to provide video evidence for subsequent tracing of responsibility for damage to the car door.

[0150] For example, when both a passenger and cargo are detected simultaneously on a floor, their relative positions are further analyzed. Specifically, it's determined whether the cargo is located on the side of the elevator car closer to the passenger, i.e., between the passenger and the car door, or even closer to the door. This indicates that the passenger's line of sight is somewhat affected by the cargo. If this spatial condition is met, during the process of controlling the elevator car to move to that floor, refined monitoring of the cargo's dynamic behavior is initiated. Within a third predetermined area (a sub-area adjacent to the car door in the "first predetermined area"), it's determined whether the cargo exhibits a tendency to move towards the car door. This movement trend can be determined through changes in the cargo's center of gravity, velocity vector, or trajectory prediction model across consecutive frames.

[0151] Once it is confirmed that the goods are approaching the elevator door within the third predetermined area, it is considered that if the elevator car is moved to the corresponding first floor while the door remains open, there is a risk of the goods colliding with the closed door. At this point, a pre-set warning can be issued to passengers. The warning can be a voice announcement (e.g., "Please note that your items may come into contact with the elevator door"), a notification on the exterior display screen, a floor projection warning, or a gentle intervention via integrated sound and light devices within the building. The aim is to guide passengers to adjust their position or pay attention to the location of the goods, thereby preventing damage to the elevator door from a collision.

[0152] The elevator control method provided in this application not only focuses on "whether anyone is riding the elevator", but also delves into the level of "spatial interaction between people and objects". It proactively predicts potential safety hazards before the elevator car door moves and reminds users to improve the safety and smoothness of the elevator ride. It is especially suitable for scenarios where the movement of large items obstructs the user's view.

[0153] The elevator system may also include a data acquisition module installed inside the elevator car. This module can prevent goods from impacting the rear wall panel (e.g., when a forklift enters the elevator car). In practice, the area of ​​the elevator car floor near the rear wall panel can be divided into two predetermined zones: a fourth and a fifth predetermined zone. Both zones are located on the car floor near the rear wall panel, with the fifth zone being closer to the rear wall panel than the fourth. The image acquisition module inside the car monitors the position of the goods in real time: when goods are detected encroaching on the fourth predetermined zone, a first-level warning mode is activated, providing a gentle reminder through a low-frequency buzzer, a yellow warning icon, and a slowly flashing overhead light strip; if the goods further enter the fifth predetermined zone, a second-level emergency warning is triggered, using multiple interactive methods such as a high-frequency buzzer alarm, a red countdown display, flashing warning lights, and voice announcements to alert passengers that the goods are about to collide with the rear wall panel. This tiered early warning mechanism, combined with cargo trajectory prediction algorithms, can achieve continuous protection from risk prediction to emergency braking, avoiding excessive intervention that affects traffic efficiency and effectively reducing the probability of rear panel collision accidents.

[0154] Both the image acquisition modules installed on each floor and the image acquisition modules installed inside the elevator car can be AI cameras. The image acquisition modules installed on each floor can be called external AI cameras, and the image acquisition modules installed inside the elevator car can be called internal AI cameras.

[0155] The car exterior AI camera should be installed at the top center of the outer side of the elevator door frame, 30cm vertically from the top edge of the door frame (error ±2cm). The lens should be tilted downwards at 10°-15° to ensure the image covers the entire gap around the door edge. The horizontal viewing angle should cover a monitoring area of ​​up to 5m (width) × 3m (height). This camera can directly interface with the elevator external call bus via the CAN 2.0B protocol, with a baud rate set to 500kbps, transmitting status signals to the control system in real time and supporting AI intelligent analysis functions. The car exterior camera is suitable for scenarios with a maximum door width of 5m, and the lens focal length needs to be calibrated (6mm fixed focal length lens recommended) to match the coverage area.

[0156] The AI ​​cameras inside the elevator car can be symmetrically installed at the top corners on both sides of the elevator door (one on each side). For two-way doors, both cameras must provide simultaneous coverage. The lenses should also be tilted downwards at 10°-15°, and the image should cover a horizontal range of 0.5m along the car wall, with a coverage area of ​​3.2m (width) × 0.5m (height).

[0157] All cameras must meet certain protection levels and be adaptable to complex working environments such as humidity and dust. They must also have reserved remote diagnostic interfaces to support firmware upgrades and remote parameter configuration via the CAN bus. The external AI camera communicates directly with the elevator's external call bus via the CAN 2.0B protocol (500kbps baud rate), transmitting door zone status signals in real time and supporting AI analysis functions such as obstacle detection. The internal AI camera outputs its collected signals to the car top plate via the CAN protocol, which then forwards them to the elevator control cabinet, enabling AI recognition functions such as cargo collision warnings and trajectory recognition.

[0158] This disclosure also provides a control device for an elevator system, such as... Figure 4 As shown, it includes a judgment module 41 and a control module 42;

[0159] The judgment module 41 is used to obtain information on the space occupancy inside the elevator car using the image acquired by the image acquisition module, and to determine whether the obtained space occupancy information conforms to the preset space occupancy information.

[0160] The control module 42, in response to the judgment result that conforms to the preset space occupancy information, blocks the external call signal.

[0161] In some exemplary embodiments, the determination module 41 is further configured to:

[0162] The images acquired by the image acquisition module are binarized, and the space occupancy information inside the elevator car is obtained based on the binarization result.

[0163] Alternatively, the images acquired by the image acquisition module can be input into a pre-trained space occupancy recognition model to obtain information on the space occupancy inside the elevator car.

[0164] In some exemplary embodiments, the determination module 41 is further configured to:

[0165] Gaussian blurring and threshold segmentation are applied to the images acquired by the image acquisition module to binarize the images into foreground and background regions; wherein the foreground region refers to the area occupied by passengers or goods, and the background region refers to the area without passengers or goods.

[0166] Obtain the number of pixels in the foreground region and calculate the ratio of the number of pixels in the foreground region to the total number of pixels in the image to obtain the pixel occupancy rate;

[0167] Based on the pre-established correspondence between pixel occupancy rate and space occupancy information, obtain the space occupancy information corresponding to the obtained pixel occupancy rate.

[0168] In some exemplary embodiments, the judgment module 41 trains the space occupancy recognition model in the following manner:

[0169] Obtain the set of images inside the elevator car used to construct the space occupancy recognition model, as well as the space occupancy information labeled in each image of the set of images inside the elevator car;

[0170] Using the obtained set of images inside the elevator car and the labeled information on space occupancy, a pre-built machine learning model is trained to obtain the space occupancy recognition model.

[0171] In some exemplary embodiments, the determination module 41 is further configured to:

[0172] The obtained set of elevator car interior images is divided into a first set of elevator car interior images for training and a second set of elevator car interior images for verification.

[0173] The images in the first elevator car image set are used as input, and the space occupancy information marked on the images in the first elevator car image set is used as labels. A pre-built machine learning model is trained, and the parameters of the machine learning model are iteratively optimized through the backpropagation algorithm until the preset conditions are met to obtain the initial model.

[0174] The images in the second elevator car image set are input into the initial model to predict the space occupancy information corresponding to the images in the second elevator car image set. The prediction accuracy of the initial model is calculated based on the predicted space occupancy information and the labeled space occupancy information. If the prediction accuracy meets a preset threshold, the obtained initial model is used as the space occupancy recognition model.

[0175] In some exemplary embodiments, the determination module 41 is further configured to:

[0176] Multiple images of the elevator car are acquired, and the acquired images are processed by at least one of the following: deleting blurry images, deleting images in which the image acquisition module is blocked, deleting images with uneven lighting, and deleting images with specular reflections.

[0177] The remaining images are used as the image set inside the elevator car.

[0178] In some exemplary embodiments, the space occupancy information includes: the degree of space occupancy;

[0179] When the space occupancy information is the degree of space occupancy, the judgment module 41 is also used to determine whether the obtained degree of space occupancy is higher than the preset degree of space occupancy.

[0180] In some exemplary embodiments, the elevator system further includes image acquisition modules respectively installed on each floor, wherein the image acquisition modules are used to acquire images of a first predetermined area outside the car door on the corresponding floor; the control module 42 is further used to:

[0181] In response to the judgment result that does not conform to the preset space occupancy information, it is determined that the space occupancy of the elevator car is not higher than the preset space occupancy. The image collected by the image acquisition module set on the first floor where no elevator call request has been received is obtained, and based on the collected image, it is determined whether there is a situation on the first floor where it is inconvenient to manually call the elevator.

[0182] In response to the determination that there is a situation on any first floor where it is inconvenient to manually call the elevator, an automatic elevator call is triggered, the floor information of the first floor is obtained, and the elevator car is controlled to run to the corresponding first floor according to the obtained floor information, wherein the first floor includes one or more floors of arbitrary height.

[0183] This disclosure also provides a control device for an elevator system, the control device for the elevator system including: a memory and a processor, the memory being configured to store an executable program;

[0184] The processor is configured to read and execute the executable program to implement the elevator system control method as described in any embodiment of this disclosure.

[0185] This disclosure also provides an elevator system, such as... Figure 5 As shown, the system includes: an elevator car 51, an image acquisition module 52 installed inside the elevator car, an image acquisition module 53 installed on each floor outside the elevator car, and a control device 54 of the elevator system described in the above embodiment that is electrically connected to the elevator car 51 and each of the image acquisition modules.

[0186] This disclosure also provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it can implement the control method of the elevator system as described in any of the above embodiments.

[0187] It will be understood by those skilled in the art that all or some of the steps, systems, or apparatuses disclosed above, and their functional modules / units, can be implemented as software, firmware, hardware, or suitable combinations thereof. In hardware implementations, the division between functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed collaboratively by several physical components. Some or all components may be implemented as software executed by a processor, such as a digital signal processor or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit (ASIC). Such software may be distributed on a computer-readable medium, which may include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term "computer storage medium" includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and can be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

Claims

1. A control method for an elevator system, applied in an elevator system, characterized in that, The elevator system includes: an elevator car; an image acquisition module installed inside the elevator car, wherein the image acquisition module is used to acquire images inside the elevator car; the control method of the elevator system includes: The image acquisition module is used to acquire images to obtain information on the space occupancy inside the elevator car, and it is determined whether the obtained space occupancy information matches the preset space occupancy information. In response to the judgment result that matches the preset space occupancy information, the external call signal is blocked; The step of obtaining information on the space occupancy inside the elevator car using images acquired by the image acquisition module includes: The images acquired by the image acquisition module are binarized, and the space occupancy information inside the elevator car is obtained based on the binarization result. Alternatively, the image acquired by the image acquisition module can be input into a pre-trained space occupancy recognition model to obtain information on the space occupancy inside the elevator car; The space occupancy recognition model is trained in the following way: Obtain an image set inside the elevator car used to construct the space occupancy recognition model, and space occupancy information labeled in each image of the image set inside the elevator car; Using the obtained set of images inside the elevator car and the labeled space occupancy information, a pre-built machine learning model is trained to obtain the space occupancy recognition model; The space occupancy information includes: the degree of space occupancy; When the space occupancy information is a degree of space occupancy, the step of determining whether the obtained space occupancy information conforms to preset space occupancy information includes: Determine whether the obtained space occupancy level is higher than the preset space occupancy level.

2. The control method for the elevator system according to claim 1, characterized in that, The step of binarizing the image acquired by the image acquisition module and obtaining the space occupancy information inside the elevator car based on the binarization result includes: The image acquired by the image acquisition module is subjected to Gaussian blur processing and threshold segmentation to binarize the image into a foreground region and a background region; wherein, the foreground region refers to the area occupied by passengers or goods, and the background region refers to the area not occupied by passengers or goods; Obtain the number of pixels in the foreground region and calculate the ratio of the number of pixels in the foreground region to the total number of pixels in the image to obtain the pixel occupancy rate; Based on the pre-established correspondence between pixel occupancy rate and space occupancy information, obtain the space occupancy information corresponding to the obtained pixel occupancy rate.

3. The control method for the elevator system according to claim 1, characterized in that, The process of using the obtained set of images inside the elevator car and the labeled space occupancy information to train a pre-built machine learning model to obtain the space occupancy recognition model includes: The obtained set of elevator car interior images is divided into a first set of elevator car interior images for training and a second set of elevator car interior images for verification. The images in the first elevator car image set are used as input, and the space occupancy information marked on the images in the first elevator car image set is used as labels. A pre-built machine learning model is trained, and the parameters of the machine learning model are iteratively optimized through the backpropagation algorithm until the preset conditions are met to obtain the initial model. The images in the second elevator car image set are input into the initial model to predict the space occupancy information corresponding to the images in the second elevator car image set. The prediction accuracy of the initial model is calculated based on the predicted space occupancy information and the labeled space occupancy information. If the prediction accuracy meets a preset threshold, the obtained initial model is used as the space occupancy recognition model.

4. The control method for the elevator system according to claim 1, characterized in that, The step of acquiring the elevator car image set used to construct the space occupancy recognition model includes: Multiple images of the elevator car are acquired, and the acquired images are processed by at least one of the following: deleting blurry images, deleting images in which the image acquisition module is blocked, deleting images with uneven lighting, and deleting images with specular reflections. The remaining images are used as the image set inside the elevator car.

5. The control method for the elevator system according to claim 1, characterized in that, The elevator system further includes image acquisition modules respectively installed on each floor, wherein the image acquisition modules are used to acquire images of a first predetermined area outside the car door on the corresponding floor; the control method of the elevator system further includes: In response to the judgment result that does not conform to the preset space occupancy information, it is determined that the space occupancy of the elevator car is not higher than the preset space occupancy. The image is acquired by the image acquisition module set on the first floor where no elevator call request has been received, and based on the acquired image, it is determined whether there is a situation on the first floor where it is inconvenient to manually call the elevator. In response to the determination that there is a situation on any first floor where it is inconvenient to manually call the elevator, an automatic elevator call is triggered, the floor information of the first floor is obtained, and the elevator car is controlled to run to the corresponding first floor according to the obtained floor information, wherein the first floor includes one or more floors.

6. A control device for an elevator system, applied in an elevator system, characterized in that, The control device includes: a memory and a processor, the memory being configured to store an executable program; The processor is configured to read the executable program and execute the control method for the elevator system as described in any one of claims 1-5.

7. An elevator system, characterized in that, The system includes an elevator car, an image acquisition module installed inside the elevator car, an image acquisition module installed on each floor outside the elevator car, and a control device for the elevator system as described in claim 6, which is electrically connected to the elevator car and each of the image acquisition modules.

Images