Signal lamp fault detection method and system based on MobileFaceNet
By constructing a traffic light detection model based on MobileFaceNet and combining time-series image processing and state analysis, the problems of low efficiency and poor accuracy in traffic light fault detection in existing technologies are solved, and real-time and accurate traffic light fault detection and alarm reminders are realized.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-28
- Publication Date
- 2026-03-10
AI Technical Summary
Existing traffic light monitoring methods rely on manual inspections, which are inefficient and costly. Traditional image processing algorithms are not optimized for traffic light features and cannot effectively detect traffic light malfunctions, leading to misjudgments or missed detections.
A traffic light fault detection method based on MobileFaceNet is adopted, which combines temporal image processing and state analysis to construct a traffic light detection model, including an initial convolutional layer, a max pooling layer, a depthwise separable convolutional layer, a bottleneck layer, and a classification output layer, for real-time detection and recognition.
It enables real-time detection of multiple fault states of traffic lights, reduces computational burden and model complexity, improves detection accuracy and robustness, is suitable for deployment on edge devices, and reduces single-frame misjudgment and noise interference.
Smart Images
Figure CN121640155A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation technology, specifically a traffic light fault detection method and system based on MobileFaceNet. Background Technology
[0002] With the development of intelligent transportation, urban traffic lights, as a crucial infrastructure for road operation, directly impact traffic safety and efficiency. Currently, traffic light monitoring primarily relies on manual inspection or traditional image processing algorithms based on thresholds and color segmentation. Manual inspection is inefficient and costly; traditional image processing algorithms are not optimized for traffic light characteristics, resulting in coarse classification granularity; they cannot identify typical fault states such as lights not being lit or multiple lights being lit simultaneously, leading to misjudgments or missed detections. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a traffic light fault detection method and system based on MobileFaceNet, which realizes real-time detection and recognition of traffic light images, and combines time-series traffic light fault judgment logic to realize the detection of multiple combinations of traffic light fault states.
[0004] The technical solution of this invention is as follows: A traffic light fault detection method based on MobileFaceNet specifically includes the following steps: (1) Collect traffic light video data, and perform image preprocessing on each frame of the traffic light video data to obtain a dataset of traffic light time sequence images; (2) Construct a traffic light detection model, input the traffic light time sequence image into the traffic light detection model for detection, and output the detection result; the traffic light detection model is based on the MobileFaceNet model, which includes sequentially connected initial convolutional layers, max pooling layers, depthwise separable convolutional layers, bottleneck layers, feature aggregation layers and classification output layers; (3) Arrange the detection results of all traffic light time sequence images in the dataset in time sequence and input them into the state analysis module. The state analysis module analyzes the detection results and time sequence relationship of all traffic light time sequence images and outputs the traffic light state type. (4) When the signal light status type output by the status analysis module is fault status mode, an alarm message is output to the traffic management center to provide an alarm reminder.
[0005] The image preprocessing described herein specifically involves cropping, scaling, and normalizing each frame of the traffic light video data.
[0006] The traffic light detection model performs the following process: First, an initial convolutional layer is used to extract edge and color features from the traffic light time-series image. Then, after continuous downsampling using a max pooling layer, a depthwise separable convolutional layer performs single-channel convolution and cross-channel feature fusion. The fused features output by the depthwise separable convolutional layer are input to a bottleneck layer. The bottleneck layer then progressively enhances semantic information and performs dimensionality compression. The output of the bottleneck layer is input to a feature aggregation layer, which performs feature aggregation and low-dimensional compression. Finally, a classification output layer performs classification detection and recognition on the low-dimensional compressed feature image, outputting six types of detection results.
[0007] The bottleneck layer comprises a five-layer structure connected sequentially. The first and third layers each include three Bottleneck inverse residual modules, the second and fourth layers each include one Bottleneck inverse residual module, and the fifth layer includes two Bottleneck inverse residual modules. Between two adjacent Bottleneck inverse residual modules, the output of one Bottleneck inverse residual module serves as the input of the other. Each Bottleneck inverse residual module includes a first 1×1 convolution, a 3×3 depthwise convolution, a second 1×1 convolution, and a residual connection that are connected sequentially. The first 1×1 convolution is used to expand the number of input channels and increase the feature dimension. The 3×3 depthwise convolution is used for channel-wise convolution to reduce the number of parameters. The second 1×1 convolution is used to compress the number of channels. When the input and output dimensions are the same, a residual connection is added.
[0008] The feature aggregation layer consists of a sequentially connected feature enhancement module, a learnable spatial pooling module, and a feature compression module. The feature enhancement module consists of sequentially connected 1×1 convolutions, batch normalization, and PReLU activation, which integrates cross-channel information and enhances semantic expression capabilities through channel dimensionality upscaling. The learnable spatial pooling module consists of depthwise convolutions and batch normalization, which is used to aggregate local features in spatial dimension into global features. The feature compression module consists of 1×1 convolutions and batch normalization, which is used to compress high-dimensional global features into a low-dimensional vector space.
[0009] The six types of test results include: red and green traffic lights are on simultaneously; red and yellow traffic lights are on simultaneously; yellow and green traffic lights are on simultaneously; none of the red, yellow, or green traffic lights are on; a single red or green traffic light is on normally; and the yellow light is on normally.
[0010] The state analysis module analyzes the detection results and timing relationships of all traffic light timing images, specifically including the following mode determination steps: S31, Normal State Mode: S311. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the red light is lit normally for multiple consecutive frames and the green light is lit normally for one or more frames, it is determined to be a normal state mode. S312. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the green light is lit normally for multiple consecutive frames, the yellow light is lit normally for three frames, and the red light is lit normally for one or more frames, it is determined to be a normal state mode. The aforementioned multiple frames of normal red light illumination and multiple frames of normal green light illumination are the retrieval results of at least 3-5 consecutive frames of traffic light timing images; S32, Fault Status Mode: S321. When the detection result of three or more frames of signal light timing images arranged in sequence is that the red and green signal lights are on at the same time, the red and yellow signal lights are on at the same time, and the yellow and green signal lights are on at the same time, it is determined to be a fault state mode. S322. If the detection result of five or more signal light timing images arranged in sequence is that the red, yellow, and green signal lights are all off, it is determined to be a fault state mode; if the detection result of ten or more signal light timing images arranged in sequence is that the red, yellow, and green signal lights are all off, it is determined to be a serious fault state mode. S323. If the detection result of the signal light timing image of ten frames or less arranged in chronological order is that the color of a single light is switched more than three times, it is determined to be a fault state mode. S324. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the red light is lit normally for multiple consecutive frames and the yellow light is lit normally for one frame, or the green light is lit normally for multiple consecutive frames and the red light is lit normally for one frame, it is determined to be a fault state mode. S325. If the detection result of the five frames of signal light timing images arranged in sequence is that the yellow light is lit normally for multiple consecutive frames, then it is determined to be a fault state mode.
[0011] The state analysis module determines the normal state mode and the fault state mode based on the sliding window segmentation method, and the number of signal light timing images in the sliding window is 10-15 frames.
[0012] The traffic light detection device based on MobileFaceNet includes an image acquisition module, an image preprocessing module, a traffic light detection module, a status analysis module, and an alarm output module. The image acquisition module is an intersection camera used to collect video data from traffic lights; The image preprocessing module is used to preprocess each frame of the traffic light video data to obtain a dataset of traffic light time sequence images. The traffic light detection module uses the MobileFaceNet model as the base network to detect the status of traffic lights in time-series images and outputs the detection results. The status analysis module is used to analyze the detection results and timing relationships of all traffic light timing images and output the traffic light status type; When the traffic light status type output by the status analysis module is a fault status mode, the alarm output module will output alarm information to the traffic management center to provide an alarm reminder.
[0013] Advantages of this invention: (1) This invention constructs a traffic light detection model based on the MobileFaceNet model. A max pooling layer is added between the initial convolutional layer and the depthwise separable convolutional layer of the existing MobileFaceNet model. The max pooling layer performs continuous downsampling, reducing the spatial size of the feature map, so that all subsequent convolutional operations are performed on a smaller feature map, greatly reducing the number of floating-point operations, and significantly reducing the overall computational burden of the model, which is particularly suitable for edge devices with limited computing power. Due to the significant reduction in computation, the actual inference time on embedded processors can be improved by 60% to 70%, enabling the traffic light detection model to meet the requirements of real-time processing.
[0014] (2) This invention reduces the number of Bottleneck inverse residual modules in the first and third layers of the bottleneck layer in the existing MobileFaceNet model, thereby reducing the size of the traffic light detection model file and making it easier to deploy on embedded devices with limited storage space and load quickly; it also makes the gradient propagation path of the bottleneck layer shorter, reduces the risk of gradient vanishing and gradient exploding, makes the model training more stable, and converges faster; for the traffic light state type retrieval task of this invention, it avoids overfitting on the training set and shows better generalization ability on small and medium-sized datasets.
[0015] (3) This invention adds a classification output layer to the existing MobileFaceNet model, which directly maps the feature vector to the output dimension of the number of categories (the categories of the six detection results), outputs the probability distribution and confidence information of each category, and can output the classification results directly from the input image end-to-end without adding any additional modules or post-processing steps to the back end of the model, and without implementing complex similarity calculation logic, thus reducing the complexity of model calculation and maintenance costs.
[0016] (4) The state analysis module of the present invention analyzes the timing relationship of all traffic light timing images, reduces single-frame misjudgment, can capture the transition of traffic light state, reduces the interference of instantaneous noise on judgment, and improves the robustness and practicality of the detection system. Attached Figure Description
[0017] Figure 1 This is a flowchart of the traffic light fault detection method of the present invention.
[0018] Figure 2 This is a framework diagram of the traffic light detection model of the present invention.
[0019] Figure 3 This is a structural block diagram of the traffic light fault detection system of the present invention. Detailed Implementation
[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] See Figure 1 A traffic light fault detection method based on MobileFaceNet specifically includes the following steps: (1) Collect traffic light video data, and perform image preprocessing on each frame of the traffic light video data to obtain a dataset of traffic light time sequence images; (2) Construct a traffic light detection model, input the traffic light time sequence image into the traffic light detection model for detection, and output the detection results; see Figure 2 The traffic light detection model uses the MobileFaceNet model as the base network, which includes sequentially connected initial convolutional layers, max pooling layers, depthwise separable convolutional layers, Bottleneck bottleneck layers, feature aggregation layers, and classification output layers. The traffic light detection model performs the following process: First, an initial convolutional layer is used to extract edge and color features from the time-series image of the traffic light. Then, after continuous downsampling using a max pooling layer, a depthwise separable convolutional layer performs single-channel convolution and cross-channel feature fusion. The fused features output by the depthwise separable convolutional layer are input to a bottleneck layer, which then progressively enhances semantic information and performs dimensionality compression. The output of the bottleneck layer is input to a feature aggregation layer, which performs feature aggregation and low-dimensional compression. Finally, a classification output layer performs classification detection and recognition on the low-dimensional compressed feature image, outputting six types of detection results. These six types of detection results include: red and green traffic lights are on simultaneously, red and yellow traffic lights are on simultaneously, yellow and green traffic lights are on simultaneously, none of the red, yellow, and green traffic lights are on, a single red or green traffic light is on normally, and the yellow light is on normally. The Bottleneck layer consists of five sequentially connected layers. The first and third layers each contain three Bottleneck inverse residual modules, the second and fourth layers each contain one Bottleneck inverse residual module, and the fifth layer contains two Bottleneck inverse residual modules. Between two adjacent Bottleneck inverse residual modules, the output of one Bottleneck inverse residual module serves as the input of the other. Each Bottleneck inverse residual module includes a first 1×1 convolution, a 3×3 depthwise convolution, a second 1×1 convolution, and a residual connection. The first 1×1 convolution expands the number of input channels and increases the feature dimension. The 3×3 depthwise convolution is used for channel-wise convolution to reduce the number of parameters. The second 1×1 convolution compresses the number of channels. When the input and output dimensions are the same, a residual connection is added. The feature aggregation layer consists of a sequentially connected feature enhancement module, a learnable spatial pooling module, and a feature compression module. The feature enhancement module consists of sequentially connected 1×1 convolutions, batch normalization, and PReLU activation. It integrates cross-channel information and enhances semantic expression capabilities through channel dimensionality upscaling. The learnable spatial pooling module consists of depthwise convolutions and batch normalization. It is used to aggregate local features in spatial dimension into global features. The feature compression module consists of 1×1 convolutions and batch normalization. It is used to compress high-dimensional global features into a low-dimensional vector space. The traffic light detection model is trained using the cross-entropy loss function to obtain a trained traffic light detection model, which is then used to detect and identify the status of traffic lights. (3) Arrange the detection results of all traffic light time-series images in the dataset in time sequence and input them into the state analysis module. The state analysis module uses the sliding window segmentation method to divide the time sequence and analyzes the traffic light time-series images (10-15 frames) within the sliding window. Specifically, it includes the following mode determination steps: S31, Normal State Mode: S311. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the red light is lit normally for multiple consecutive frames and the green light is lit normally for one or more frames, it is determined to be a normal state mode. S312. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the green light is lit normally for multiple consecutive frames, the yellow light is lit normally for three frames, and the red light is lit normally for one or more frames, it is determined to be a normal state mode. The search results for multiple frames of red lights illuminating normally and multiple frames of green lights illuminating normally are at least 3-5 consecutive frames of traffic light timing images; S32, Fault Status Mode: S321. When the detection result of three or more frames of signal light timing images arranged in sequence is that the red and green signal lights are on at the same time, the red and yellow signal lights are on at the same time, and the yellow and green signal lights are on at the same time, it is determined to be a fault state mode. S322. If the detection result of five or more signal light timing images arranged in sequence is that the red, yellow, and green signal lights are all off, it is determined to be a fault state mode; if the detection result of ten or more signal light timing images arranged in sequence is that the red, yellow, and green signal lights are all off, it is determined to be a serious fault state mode. S323. If the detection result of the signal light timing image of ten frames or less arranged in chronological order is that the color of a single light is switched more than three times, it is determined to be a fault state mode. S324. When the detection result of the multi-frame signal light timing image arranged in time sequence is that the red light is lit normally for multiple consecutive frames and the yellow light is lit normally for one frame, or the green light is lit normally for multiple consecutive frames and the red light is lit normally for one frame, it is determined to be a fault state mode. S325. If the detection result of the five frames of signal light timing images arranged in sequence is that the yellow light is lit normally for multiple consecutive frames, then it is determined to be a fault state mode. (4) When the signal light status type output by the status analysis module is fault status mode, an alarm message is output to the traffic management center to provide an alarm reminder.
[0022] See Figure 3 A traffic light detection device based on MobileFaceNet includes an image acquisition module, an image preprocessing module, a traffic light detection module, a status analysis module, and an alarm output module. The image acquisition module is an intersection camera used to collect video data from traffic lights; The image preprocessing module is used to preprocess each frame of the traffic light video data to obtain a dataset of traffic light time sequence images. The traffic light detection module uses the MobileFaceNet model as the base network to detect the status of traffic lights in time-series images and outputs the detection results. The status analysis module is used to analyze the detection results and timing relationships of all traffic light timing images and output the traffic light status type; When the traffic light status type output by the status analysis module is a fault status mode, the alarm output module will output alarm information to the traffic management center to provide an alarm reminder.
[0023] (4) When the signal light status type output by the status analysis module is faulty, an alarm message is output to the traffic management center to provide an alarm reminder. Example
[0024] Traffic light video data is obtained from the monitoring system, preprocessed to obtain the timing image of the traffic light to be detected, and then input into the pre-trained traffic light detection model.
[0025] (I) Experimental Data Details: (1) Training set: Images of traffic lights simultaneously lit (fault): 3095 images; Images of traffic lights simultaneously lit (fault): 2628 images; Images of traffic lights simultaneously lit (fault): 1936 images; Images of traffic lights not lit (fault): 3452 images; Images of traffic lights with one red or green light lit normally: 10819 images; Images of traffic lights lit normally: 2647 images; (2) Validation set: Images of traffic lights simultaneously lit (fault): 773 images; Images of traffic lights simultaneously lit (fault): 657 images; Images of traffic lights simultaneously lit (fault): 484 images; Images of traffic lights not lit (fault): 1842 images; Images of traffic lights with one red or green light lit normally: 2493 images; Images of traffic lights with the yellow light lit normally: 351 images; (II) Processing Procedure: (1) Image preprocessing: Input: RGB image of arbitrary size; Output: Normalized tensor image (1,3,64,64) after converting to three channels, scaling to 64×64 and normalizing. (2) Traffic light detection model performs detection: Input image: (1,3,64,64); Output: 6-dimensional probability vector, categories 0, 1, 2, 3, 4, 5, representing red and green lights on simultaneously, red and yellow lights on simultaneously, yellow and green lights on simultaneously, red, yellow and green lights all off, red or green light on normally, and yellow light on normally, respectively. The detection results of the traffic light detection model are shown in Table 1 below. The evaluation index comparison results of the traffic light detection model of the present invention with the traditional color detection method and the target detection method based on YOLO are shown in Table 2 below.
[0026] Table 1
[0027] Table 2
[0028] As shown in Table 2 above, the traffic light detection model of the present invention has higher accuracy, better fault detection rate, lighter model, stronger accuracy in small target detection, lower deployment cost, and can run the end-to-end model stably on medium-sized GPUs, making it more suitable for classifying traffic lights as normal or faulty.
[0029] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A traffic light fault detection method based on MobileFaceNet, characterized in that: Specifically comprising the following steps: (1) Collecting signal lamp video data, and after image preprocessing of each frame of image of the signal lamp video data, obtaining a data set of signal lamp time sequence images; (2) Constructing a signal lamp detection model, inputting the signal lamp time sequence images into the signal lamp detection model for detection, and outputting a detection result; the signal lamp detection model takes a MobileFaceNet model as a benchmark network, and comprises sequentially connected initial convolutional layers, a maximum pooling layer, a deep separable convolutional layer, a bottleneck layer, a feature aggregation layer and a classification output layer; (3) Arranging the detection results of all signal lamp time sequence images in the data set in time sequence and inputting into a state analysis module, and the state analysis module analyzes according to the detection results and time sequence relationship of all signal lamp time sequence images, and outputs a signal lamp state type; (4) When the signal lamp state type output by the state analysis module is a fault state mode, outputting an alarm information to a traffic management center for alarm reminding.
2. The signal light fault detection method based on MobileFaceNet according to claim 1, wherein: The image preprocessing is specifically clipping, scaling and normalization processing of each frame of image of the signal lamp video data.
3. The signal light fault detection method based on MobileFaceNet according to claim 1, characterized in that: The detection process of the signal lamp detection model is specifically: first, using the initial convolutional layer to extract edge and color features in the signal lamp time sequence image, then using the maximum pooling layer to perform continuous downsampling, then using the deep separable convolutional layer to perform single-channel convolution operation and cross-channel feature fusion, inputting the fusion features output by the deep separable convolutional layer into the bottleneck layer, the bottleneck layer gradually improves semantic information and performs dimension compression, inputting the output of the bottleneck layer into the feature aggregation layer, the feature aggregation layer performs feature aggregation and low-dimensional compression, and finally the classification output layer classifies and detects the low-dimensional compressed feature image, and outputs six types of detection results.
4. The signal light fault detection method based on MobileFaceNet according to claim 3, characterized in that: The bottleneck layer comprises five layers of structures connected in sequence, the first layer structure and the third layer structure each comprise three Bottleneck reverse residual modules, the second layer structure and the fourth layer structure each comprise one Bottleneck reverse residual module, and the fifth layer structure comprises two Bottleneck reverse residual modules, the output of one Bottleneck reverse residual module between two adjacent Bottleneck reverse residual modules is taken as the input of the other Bottleneck reverse residual module, each Bottleneck reverse residual module comprises a first 1x1 convolution, a 3x3 deep convolution, a second 1x1 convolution and a residual connection connected in sequence, the first 1x1 convolution is used to expand the input channel number and increase the feature dimension, the 3x3 deep convolution is used for channel-by-channel convolution and parameter reduction, and the second 1x1 convolution is used to compress the channel number, and when the input and output dimensions are consistent, the residual connection is added.
5. The signal light fault detection method based on MobileFaceNet according to claim 3, characterized in that: The feature aggregation layer is composed of sequentially connected feature enhancement modules, learnable spatial pooling modules and feature compression modules, the feature enhancement module is composed of sequentially connected 1*1 convolution, batch normalization and PReLU activation, the channel dimension is increased to integrate cross-channel information and strengthen semantic expression ability, the learnable spatial pooling module is composed of deep convolution and batch normalization, which is used for aggregating local features in spatial dimension into global features, and the feature compression module is composed of 1*1 convolution and batch normalization, which is used for compressing high-dimensional global features into low-dimensional vector space.
6. The signal light fault detection method based on MobileFaceNet according to claim 3, characterized in that: The six types of detection results include simultaneous lighting of red and green signal lights, simultaneous lighting of red and yellow signal lights, simultaneous lighting of yellow and green signal lights, non-lighting of red, yellow and green signal lights, single normal lighting of red or green signal light and normal lighting of yellow light.
7. The MobileFaceNet-based signal light fault detection method of claim 6, wherein: The state analysis module analyzes the detection results and time sequence relationship of all signal light time sequence images, and specifically includes the following mode determination steps: S31, normal state mode: S311, when the detection results of the multiple signal light time sequence images arranged in time sequence are continuous multiple red light normal lighting and one or more green light normal lighting, the normal state mode is determined; S312, when the detection results of the multiple signal light time sequence images arranged in time sequence are continuous multiple green light normal lighting, three yellow light normal lighting and one or more red light normal lighting, the normal state mode is determined; The multiple red light normal lighting and multiple green light normal lighting are at least continuous 3-5 frame signal light time sequence image search results; S32, fault state mode: S321, when the detection results of three or more signal light time sequence images arranged in time sequence are simultaneous lighting of red and green signal lights, simultaneous lighting of red and yellow signal lights and simultaneous lighting of yellow and green signal lights, the fault state mode is determined; S322, when the detection results of five or more signal light time sequence images arranged in time sequence are non-lighting of red, yellow and green signal lights, the fault state mode is determined; when the detection results of ten or more signal light time sequence images arranged in time sequence are non-lighting of red, yellow and green signal lights, the serious fault state mode is determined; S323, when the detection results of ten or less signal light time sequence images arranged in time sequence are color switching of single light lighting more than three times, the fault state mode is determined; S324, when the detection results of multiple signal light time sequence images arranged in time sequence are continuous multiple red light normal lighting and one yellow light normal lighting, or continuous multiple green light normal lighting and one red light normal lighting, the fault state mode is determined; S325, when the detection results of five signal light time sequence images arranged in time sequence are continuous multiple yellow light normal lighting, the fault state mode is determined.
8. The signal light failure detection method according to claim 7, characterized by: The state analysis module determines the normal state mode and the fault state mode based on a sliding window segmentation method, and the number of signal light time sequence images in the sliding window is 10-15 frames.
9. A signal light detection device for implementing the signal light failure detection method according to claim 1, characterized by: The system comprises an image acquisition module, an image preprocessing module, a signal light detection module, a state analysis module and an alarm output module; The image acquisition module is a road intersection camera, which is used for acquiring signal light video data; The image preprocessing module is configured to perform image preprocessing on each frame of image of the signal lamp video data to obtain a data set of signal lamp time sequence images; The signal lamp detection module, taking a MobileFaceNet model as a benchmark network, is configured to detect a signal lamp state in the signal lamp time sequence images and output a detection result; The state analysis module is configured to analyze the detection result and a time sequence relationship of all the signal lamp time sequence images and output a signal lamp state type; When the signal lamp state type output by the state analysis module is a fault state mode, the alarm output module outputs alarm information to a traffic management center for alarm reminding.