Fire door state recognition method based on AI vision

Abstract

This invention discloses an AI vision-based method for fire door status recognition, comprising the following steps: collecting and preprocessing data to obtain standardized visual data; extracting structural visual features of the fire door; performing evidence mapping processing to convert it into structural evidence and constructing state support parameters; constructing and correcting structural constraint relationships to obtain constraint-corrected state support parameters; inputting the structural visual features into a deep random vector functional connection network and updating the state support parameters; constructing an evidence set and calculating the degree of constraint violation to obtain evidence conflict measurement results; performing reliability assessment to obtain weight parameters corresponding to each structural piece of evidence; constructing support channels respectively, and incorporating the evidence conflict measurement results for constraint suppression processing to obtain a fire door status judgment value; and outputting the fire door status result. This invention employs an AI vision-based state recognition method based on structural constraint relationships to achieve fire door status determination with high accuracy and robustness.

Description

Technical Field

[0001] This invention relates to the field of pattern recognition technology, and in particular to a method for fire door status recognition based on AI vision. Background Technology

[0002] Fire doors are key components in building fire protection partitions and personnel evacuation. In practical applications, it is usually necessary to continuously monitor their closed, open, ajar, and abnormal states. Currently, fire door status recognition mostly relies on manual inspection, door magnetic sensor detection, or single visual classification methods based on monitoring images. Among these, manual inspection has the problem of insufficient real-time performance, door magnetic sensors have the problems of high installation and maintenance costs and limited status dimensions, while single visual classification methods can automatically identify using images or videos, but usually mainly output the status category based on the overall image features or local features.

[0003] However, existing technologies still have significant shortcomings in complex environments. When fire doors are affected by obstruction, light fluctuations, changes in shooting angle, or abnormal positions of local components, inconsistencies or even contradictions often occur between different structural features. Existing methods usually lack the ability to uniformly model the inherent correlation between door gap width, contact continuity, door leaf posture, door closer position, and door latch state. It is difficult to perform consistency constraint correction, constraint violation degree calculation, and conflict suppression processing on various structural evidences, which can easily lead to misjudgment of closed state, missed judgment of partially closed state, and insufficient stability of abnormal state identification. Summary of the Invention

[0004] One objective of this invention is to propose an AI vision-based fire door status recognition method. This invention employs an AI vision-based status recognition method based on structural constraint relationships to achieve fire door status determination with high accuracy and robustness.

[0005] A fire door status recognition method based on AI vision according to an embodiment of the present invention includes the following steps: The collected data is preprocessed to obtain standardized visual data; Based on standardized visual data, structural-level visual features of fire doors are extracted; Evidence mapping is performed on structural-level visual features to convert each structural-level visual feature into corresponding structural evidence and construct state support parameters. Construct structural constraint relationships between structural evidences and perform consistency constraint correction to obtain the state support parameters after constraint correction; The structural-level visual features are input into the deep random vector functional connection network for nonlinear mapping and feature fusion processing. The state support parameters after constraint correction are further corrected to obtain the updated state support parameters. An evidence set is constructed based on the updated state support parameters, and the degree of constraint violation between each piece of structural evidence is calculated based on the structural constraint relationship to obtain the evidence conflict measurement result. Based on standardized visual data, the reliability of the state support parameters corresponding to each structural evidence is evaluated to obtain the weight parameters corresponding to each structural evidence. Based on the updated state support parameters and weight parameters, closed support channels and open support channels are constructed respectively, and the evidence conflict measurement results are introduced for constraint suppression processing to obtain the fire door state judgment value. The fire door status result is output based on the fire door status judgment value and the evidence conflict measurement result.

[0006] Optionally, the preprocessing includes noise reduction, brightness normalization, and geometric correction.

[0007] Optionally, the structural visual features include door gap width features, door frame and door leaf contact continuity features, door leaf posture angle features, door closer position features, and door lock tongue state features.

[0008] Optionally, the construction of the state support parameter includes: Evidence mapping processing is performed on the door gap width feature to convert the door gap width feature into door gap width structural evidence. State quantization processing is performed based on the difference between the feature value corresponding to the door gap width structural evidence and the door gap width benchmark value. The closed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically decreasing mapping relationship, and the unclosed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically increasing mapping relationship complementary to the closed state support parameter. Evidence mapping processing is performed on the contact continuity features between the door frame and the door leaf to convert the contact continuity features between the door frame and the door leaf into contact continuity structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the contact continuity structure evidence are obtained. Evidence mapping processing is performed on the door leaf attitude angle features to convert the door leaf attitude angle features into door leaf attitude structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door leaf attitude structure evidence are obtained. Evidence mapping processing is performed on the door closer position features to convert them into door closer position structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door closer position structure evidence are obtained. Evidence mapping processing is performed on the door latch state features to convert them into door latch state structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door latch state structure evidence are obtained. The support parameters for the closed state and the support parameters for the unclosed state corresponding to the structural evidence of door gap width, structural evidence of contact continuity, structural evidence of door leaf posture, structural evidence of door closer position, and structural evidence of door lock tongue state are uniformly combined to form the state support parameters.

[0009] Optionally, the generation of the constraint-corrected state support parameters includes: Based on the state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state, a set of structural constraint relationships is constructed. For each set of structural constraint relationships in the set of structural constraint relationships, the constraint deviation value is calculated based on the state support parameter of the corresponding structural evidence. Based on the constraint deviation values ​​corresponding to each structural constraint relationship, a comprehensive constraint deviation value is constructed for each structural evidence. Based on the comprehensive constraint deviation value corresponding to each structural evidence, the state support parameter corresponding to each structural evidence is subjected to consistency constraint correction processing, thereby obtaining the constraint-corrected state support parameter corresponding to each structural evidence. The constraint-corrected state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state are uniformly organized to obtain the constraint-corrected state support parameters.

[0010] Optionally, the generation of the updated state support parameter includes: The door gap width feature, the door frame and door leaf contact continuity feature, the door leaf posture angle feature, the door closer position feature, and the door lock tongue state feature are combined in a unified order to form a structural visual feature input sequence. The constraint-corrected state support parameters corresponding to the door gap width structural evidence, contact continuity structural evidence, door leaf posture structural evidence, door closer position structural evidence, and door lock tongue state structural evidence are combined in the same order to form a constraint-corrected state support parameter input sequence. The structural visual feature input sequence is fused with the constraint-corrected state support parameter input sequence to construct a joint input sequence, and the joint input sequence is input into the deep random vector function connection network. The deep stochastic vector functional connection network includes a stochastic mapping layer, a direct connection fusion layer, a multi-layer progressive modeling layer, and a state support correction layer. In the random mapping layer, random weight mapping and nonlinear activation processing are performed on the joint input sequence to obtain random mapping features. In the direct connection fusion layer, the random mapping features and the joint input sequence are directly concatenated to obtain fused features. In the multi-layer progressive modeling layer, the fused features of the current layer are input into the next layer, and random weight mapping, nonlinear activation processing, and direct connection concatenation processing are repeatedly performed to obtain a multi-layer fused feature sequence. In the state support correction layer, the fused features of the final layer are used as correction input to generate the closed state support correction amount and the open state support correction amount corresponding to each structural evidence. Based on the support corrections for the closed and open states corresponding to each structural piece of evidence, constraint coupling update processing is performed on the corresponding constraint-corrected support parameters for the closed and open states. Complementary constraints are then applied to the two types of state support parameters corresponding to the same structural piece of evidence to obtain the updated state support parameters for each structural piece of evidence.

[0011] Optionally, the generation of the evidence conflict measurement result includes: Based on the updated state support parameters corresponding to the structural evidence of door gap width, structural evidence of contact continuity, structural evidence of door leaf posture, structural evidence of door closer position, and structural evidence of door lock tongue state, an evidence set is constructed. For each set of structural constraint relationships in the set of structural constraint relationships, the degree of constraint violation is constructed based on the updated state support parameter of the corresponding structural evidence, and the degree of constraint violation corresponding to the reverse constraint relationship is obtained. For each set of unidirectional constraints in the structural constraint relationship set, the degree of constraint violation corresponding to the unidirectional constraint relationship is obtained; Based on the degree of constraint violation corresponding to each structural constraint relationship, a comprehensive degree of constraint violation is constructed for each structural evidence. The global aggregation process is performed based on the degree of comprehensive constraint violation corresponding to each structural piece of evidence to obtain the evidence conflict measurement result corresponding to the evidence set.

[0012] Optionally, the generation of the weight parameters includes: Information on image sharpness, occlusion degree, and continuous frame state changes is obtained based on standardized visual data. Based on image clarity, occlusion degree and continuous frame state change information, the reliability of the state support parameters corresponding to each structural evidence is evaluated, and the reliability evaluation value of each structural evidence is obtained. The reliability assessment values ​​corresponding to each structural piece of evidence are normalized to obtain the weight parameters corresponding to each structural piece of evidence. The weight parameters corresponding to each structural piece of evidence are uniformly organized to obtain the weight parameters corresponding to each structural piece of evidence.

[0013] Optionally, the generation of the fire door status determination value includes: Based on the updated state support parameters and weight parameters corresponding to each structural piece of evidence, closed support channels and open support channels are constructed respectively. Based on the evidence conflict measurement results, the closed support contribution value and the unclosed support contribution value corresponding to each structural evidence are constrained and suppressed to obtain the constrained and suppressed closed support contribution value and the constrained and suppressed unclosed support contribution value corresponding to each structural evidence. The closed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the closed support value for the closed support channel. The unclosed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the unclosed support value for the unclosed support channel. The fire door status determination value is obtained by calculating the difference between the closed support value and the unclosed support value.

[0014] Optionally, the generation of the fire door status result includes: Obtain the fire door status judgment value and evidence conflict measurement result, and set the closed status judgment threshold, open status judgment threshold and ajar status judgment interval corresponding to the fire door status judgment value, as well as the abnormal status judgment threshold corresponding to the evidence conflict measurement result. The fire door status judgment value is compared with the closed status judgment threshold, the open status judgment threshold and the half-closed status judgment interval, and the evidence conflict measurement result is compared with the abnormal status judgment threshold to form the closed status judgment condition, the open status judgment condition, the half-closed status judgment condition and the abnormal status judgment condition. When the fire door status judgment value is greater than or equal to the closed status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the closed status judgment condition is met and the closed status result is output. When the fire door status judgment value is less than or equal to the open status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the open status judgment condition is met and the open status result is output. When the fire door status judgment value is within the false opening status judgment range and the evidence conflict measurement result is less than the abnormal status judgment threshold, the false opening status judgment condition is met, and the false opening status result is output. When the result of the evidence conflict measurement is greater than or equal to the abnormal state determination threshold, the abnormal state determination condition is met, and the abnormal state result is output. The current output result among the closed state result, open state result, partially closed state result, and abnormal state result is determined as the fire door state result.

[0015] The beneficial effects of this invention are: This invention provides an AI-based vision-based method for fire door status recognition. It extracts features from fire door images or videos, including door gap width, door frame-door leaf contact continuity, door leaf posture angle, door closer position, and door latch state. These features are then converted into corresponding structural evidence. State support parameters are constructed based on this structural evidence, transforming the previous method, which relied on a single classification result, into a state determination method based on multiple structural evidence. Compared to existing technologies, this invention does not directly determine the fire door state based solely on a single image feature or overall visual result. Instead, it first establishes structural-level evidence representation before conducting subsequent constraint analysis and decision processing. Therefore, it can more accurately reflect the structural state of the fire door in real-world scenarios, improving the consistency between the state recognition result and the actual physical closure state.

[0016] Furthermore, this invention constructs structural constraint relationships between structural evidences and applies consistency constraints to the state support parameters corresponding to each structural evidence based on these relationships. It then further refines the constrained state support parameters using a deep random vector functional connection network. This transforms different structural evidences from isolated, independent judgments into a cohesive judgment system with inherent logical connections. Consequently, even when a fire door is affected by obstruction, changes in lighting, local component displacement, or changes in shooting angle, this invention can still correct for abnormal deviations by relying on the relationships between structural evidences. This reduces misjudgments caused by local image distortion or fluctuations in single features, improving the stability and robustness of recognition in complex environments.

[0017] Furthermore, this invention calculates the degree of constraint violation between structural evidence based on structural constraint relationships to obtain evidence conflict measurement results. It further combines image clarity, occlusion degree, and continuous frame state change information to generate weight parameters corresponding to each structural evidence. The evidence conflict measurement results are introduced into both closed and open support channels for constraint suppression processing, reducing the contribution of structural evidence violating structural constraints to the corresponding support channels. Therefore, this invention not only accurately distinguishes between closed, open, partially obscured, and abnormal states, but also prioritizes the identification of abnormal or unstable states when conflicts occur between structural evidence, avoiding misclassification caused by misleading local features in existing technologies. Thus, it has higher state determination accuracy, better anomaly identification capabilities, and greater engineering application value. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1This is an overall flowchart of a fire door status recognition method based on AI vision proposed in this invention; Figure 2 This is a schematic diagram illustrating the nonlinear mapping and feature fusion processing of the constrained state support parameters by the deep random vector functional connection network in the fire door state recognition method based on AI vision proposed in this invention. Figure 3 This is a schematic diagram illustrating the output of closed, open, ajar, and abnormal states based on the fire door status judgment value and evidence conflict measurement results in the fire door status recognition method proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figures 1-3 A method for fire door status recognition based on AI vision includes the following steps: Collect image or video data of the monitoring area where the fire door is located, and perform noise reduction, brightness normalization and geometric correction on the image or video data to obtain standardized visual data; Based on standardized visual data, structural visual features of fire doors are extracted. These structural visual features include door gap width features, door frame and door leaf contact continuity features, door leaf posture angle features, door closer position features, and door lock tongue status features. Evidence mapping processing is performed on structural visual features to convert each structural visual feature into corresponding structural evidence. State support parameters are constructed based on each structural evidence. The state support parameters are obtained by using a monotonic mapping function from the feature values ​​corresponding to the structural evidence. Structural constraint relationships are constructed between structural evidences, and the state support parameters corresponding to each structural evidence are corrected by consistency constraints based on the structural constraint relationships to obtain the state support parameters after constraint correction. The structural-level visual features are input into the deep random vector functional connection network for nonlinear mapping and feature fusion processing. The state support parameters after constraint correction are further corrected to obtain the updated state support parameters. An evidence set is constructed based on the updated state support parameters, and the degree of constraint violation between each piece of structural evidence is calculated based on the structural constraint relationship to obtain the evidence conflict measurement result. The evidence conflict measurement result is used to characterize the degree of deviation between structural evidence from the structural constraint relationship. Based on standardized visual data, image clarity, occlusion degree and continuous frame state change information are obtained. The reliability of the state support parameters corresponding to each structural evidence is evaluated based on the image clarity, occlusion degree and continuous frame state change information, and the weight parameters corresponding to each structural evidence are obtained. Based on the updated state support parameters and weight parameters, closed support channels and open support channels are constructed respectively. Evidence conflict measurement results are introduced into the closed support channels and open support channels for constraint suppression processing, so that the contribution of structural evidence that violates structural constraints to the corresponding support channels is reduced, and the fire door state judgment value is obtained. The fire door status results are output based on the fire door status judgment value and evidence conflict measurement results. The fire door status results include closed status, open status, ajar status and abnormal status.

[0021] In this embodiment, the preprocessing includes noise reduction, brightness normalization, and geometric correction.

[0022] In this embodiment, the structural visual features include door gap width features, door frame and door leaf contact continuity features, door leaf posture angle features, door closer position features, and door lock tongue state features.

[0023] In this embodiment, the construction of the state support parameter includes: Evidence mapping processing is performed on the door gap width feature to convert the door gap width feature into door gap width structural evidence. State quantization processing is performed based on the difference between the feature value corresponding to the door gap width structural evidence and the door gap width benchmark value. The closed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically decreasing mapping relationship, and the unclosed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically increasing mapping relationship complementary to the closed state support parameter. Evidence mapping processing is performed on the contact continuity features between the door frame and the door leaf to convert the contact continuity features into contact continuity structural evidence. Based on the proportional relationship between the feature values ​​corresponding to the contact continuity structural evidence and the theoretical contact length of the door frame, state quantification processing is performed. The closed state support parameter corresponding to the contact continuity structural evidence is obtained through a monotonically increasing mapping relationship, and the unclosed state support parameter corresponding to the contact continuity structural evidence is obtained through a monotonically decreasing mapping relationship complementary to the closed state support parameter. During evidence mapping processing, the state direction is first determined according to the physical meaning of the structural features. Then, each structural evidence is divided into two categories: positive closure evidence and negative closure evidence. Contact continuity structural evidence and door latch state structural evidence are used as positive closure evidence, while door gap width structural evidence, door leaf posture structural evidence, and door closer position structural evidence are used as negative closure evidence. A transition interval is set for each type of structural evidence so that the feature value corresponds to a gradual change in the state support parameter when approaching the closure boundary, rather than a direct jump. The state support parameters obtained from each structural evidence are constrained at the same scale so that the state support parameters output by different structural evidence are within the same numerical range, and the closed state support parameter corresponding to each structural evidence can be directly compared with the open state support parameter. Evidence mapping processing is performed on the door leaf attitude angle features to convert them into door leaf attitude structure evidence. State quantization processing is performed based on the deviation between the feature values ​​corresponding to the door leaf attitude structure evidence and the attitude angle of the closed state. The support parameter of the closed state corresponding to the door leaf attitude structure evidence is obtained through a monotonically decreasing mapping relationship, and the support parameter of the unclosed state corresponding to the door leaf attitude structure evidence is obtained through a monotonically increasing mapping relationship that is complementary to the support parameter of the closed state. Evidence mapping processing is performed on the door closer position features to convert them into door closer position structure evidence. State quantification processing is performed based on the displacement deviation between the feature value corresponding to the door closer position structure evidence and the door closing termination position. The closed state support parameter corresponding to the door closer position structure evidence is obtained through a monotonically decreasing mapping relationship, and the unclosed state support parameter corresponding to the door closer position structure evidence is obtained through a monotonically increasing mapping relationship complementary to the closed state support parameter. Evidence mapping processing is performed on the door lock tongue state features to convert them into door lock tongue state structure evidence. State quantification processing is then performed based on the correspondence between the feature values ​​corresponding to the door lock tongue state structure evidence and the locking and closing states. The closed state support parameter corresponding to the door lock tongue state structure evidence is obtained through a monotonically increasing mapping relationship, and the unclosed state support parameter corresponding to the door lock tongue state structure evidence is obtained through a monotonically decreasing mapping relationship that is complementary to the closed state support parameter. The support parameters for the closed state and the support parameters for the unclosed state corresponding to the structural evidence of door gap width, structural evidence of contact continuity, structural evidence of door leaf posture, structural evidence of door closer position, and structural evidence of door lock tongue state are uniformly combined to form the state support parameters.

[0024] In this embodiment, the generation of the constrained state support parameters includes: Based on the state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state, a set of structural constraint relationships is constructed. The set of structural constraint relationships consists of multiple sets of structural constraint relationships. Each set of structural constraint relationships is based on the correlation between two sets of structural evidence. The constraint direction type is determined for each set of structural constraint relationships. The constraint direction type includes reverse constraint relationship and same-direction constraint relationship. For each set of structural constraint relationships in the set of structural constraint relationships, the constraint deviation value is calculated based on the state support parameter of the corresponding structural evidence. The calculation of the constraint deviation value specifically includes: For reverse constraint relationships, the first deviation is obtained by performing absolute value processing on the difference between the support parameter of the closed state corresponding to the first structural evidence and the support parameter of the open state corresponding to the second structural evidence; the second deviation is obtained by performing absolute value processing on the difference between the support parameter of the open state corresponding to the first structural evidence and the support parameter of the closed state corresponding to the second structural evidence; and the constraint deviation value corresponding to the reverse constraint relationship is obtained by weighted combination of the first deviation and the second deviation. For same-direction constraint relationships, the third deviation is obtained by performing absolute value processing on the difference between the support parameters of the closed state corresponding to the two structural evidences; the fourth deviation is obtained by performing absolute value processing on the difference between the support parameters of the open state corresponding to the two structural evidences; and the constraint deviation value corresponding to the same-direction constraint relationship is obtained by weighted combination of the third deviation and the fourth deviation. Based on the constraint deviation values ​​corresponding to each structural constraint relationship, a comprehensive constraint deviation value is constructed for each structural evidence. For each structural evidence, the constraint deviation values ​​corresponding to all structural constraint relationships related to that structural evidence are summed or weighted summed to obtain the comprehensive constraint deviation value corresponding to that structural evidence. Based on the comprehensive constraint deviation value corresponding to each structural evidence, the state support parameter corresponding to each structural evidence is subjected to consistency constraint correction processing. The consistency constraint correction processing adopts a two-way coupling adjustment method, so that when the comprehensive constraint deviation value increases, the support parameter of the closed state of the corresponding structural evidence decreases and the support parameter of the open state increases, and when the comprehensive constraint deviation value decreases, the support parameter of the closed state of the corresponding structural evidence increases and the support parameter of the open state decreases, thereby obtaining the constraint-corrected state support parameter corresponding to each structural evidence. The constraint-corrected state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state are uniformly organized to obtain the constraint-corrected state support parameters.

[0025] In this embodiment, the generation of the updated state support parameter includes: The structural visual feature input sequence is formed by combining the door gap width feature, the door frame and door leaf contact continuity feature, the door leaf posture angle feature, the door closer position feature, and the door lock tongue state feature in a unified order. The constraint-corrected state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state are combined in the same order to form the constraint-corrected state support parameter input sequence, so that the structural visual feature input sequence and the constraint-corrected state support parameter input sequence correspond one-to-one at the structural evidence level. The structural visual feature input sequence is fused with the constraint-corrected state support parameter input sequence to construct a joint input sequence, and the joint input sequence is input into the deep random vector function connection network. The deep stochastic vector functional connection network includes a stochastic mapping layer, a direct connection fusion layer, a multi-layer progressive modeling layer, and a state support correction layer; In the random mapping layer, random weight mapping and nonlinear activation processing are performed on the joint input sequence to obtain random mapping features. In the direct connection fusion layer, the random mapping features and the joint input sequence are directly concatenated to obtain fused features. In the multi-layer progressive modeling layer, the fused features of the current layer are input into the next layer, and random weight mapping, nonlinear activation processing, and direct connection concatenation processing are repeatedly performed to obtain a multi-layer fused feature sequence. In the state support correction layer, the fused features of the final layer are used as correction input to generate the closed state support correction amount and the open state support correction amount corresponding to each structural evidence. Randomized mapping features refer to the feature representation formed after the joint input sequence undergoes randomized weight mapping and nonlinear activation processing in the randomized mapping layer. Randomized weight mapping refers to generating input weights and biases in a randomized initialization manner during the network construction stage, and keeping the input weights and biases unchanged after generation. After the joint input sequence is input into the randomized mapping layer, it is first linearly transformed with fixed input weights, then the corresponding biases are superimposed, and finally nonlinear activation processing is performed to obtain the randomized mapping features. Since the joint input sequence is a fixed input during forward computation, and the input weights and biases are fixed parameters after the network is constructed, the randomized mapping features are a fixed output result under the same network parameter conditions. This deep stochastic vector functional connection network is composed of sequentially connected components in a series structure. The output of the stochastic mapping layer and the joint input sequence are concatenated in the direct connection fusion layer, and the concatenation result is then input into the next layer. The multi-layer progressive modeling layers are stacked in series in the same way, and finally output to the state support correction layer. The input data consists of structural visual feature vectors arranged in a uniform order and state support parameter vectors after constraint correction. The data format is a two-dimensional matrix of sample number multiplied by feature dimension. The output data consists of the closed state support correction amount and the open state support correction amount corresponding to each structural evidence. The training data comes from fire door monitoring videos and image samples. First, the door gap width, contact continuity, door leaf posture angle, door closer position and door lock tongue status are extracted. Then, combined with the manual verification results, they are labeled as closed state, open state, ajar state and abnormal state, and the corresponding target state support parameters are generated simultaneously. During model training, a weighted combination of state support regression loss and state classification loss is used as the loss function. The state support regression loss is used to calculate the error between the closed state support correction and the open state support correction corresponding to each structural evidence and the target state support parameter, and the errors corresponding to each structural evidence are accumulated. The state classification loss is used to calculate the classification error between the model output closed state, open state, masked state, and abnormal state and the manually labeled state. The loss is obtained by weighted summation of the state support regression loss and the state classification loss. The training parameters are set to 100 to 200 training epochs, batch size of 32 to 128, learning rate of 0.001 to 0.0001, and convergence condition is set to stop training when the validation set loss no longer decreases after several consecutive training epochs, or when the total loss decreases below a threshold. Based on the closed-state support correction and open-state support correction for each structural piece of evidence, constraint coupling update processing is performed on the corresponding constraint-corrected closed-state support parameters and open-state support parameters. The closed-state support correction is used to drive the closed-state support parameter to adjust along the closed direction, and the open-state support correction is used to drive the open-state support parameter to adjust along the open direction. Complementary constraints are applied to the two types of state support parameters corresponding to the same structural piece of evidence, so that the enhancement of one type of state support parameter is accompanied by the suppression of the other type of state support parameter, thereby obtaining the updated state support parameters for each structural piece of evidence.

[0026] In this embodiment, the generation of the evidence conflict measurement result includes: Based on the updated state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door latch state, an evidence set is constructed. The evidence set includes the updated closed state support parameters and the updated unclosed state support parameters corresponding to each structural evidence, while maintaining the correspondence between each structural evidence. For each set of structural constraint relationships in the set of structural constraint relationships, the degree of constraint violation is constructed based on the updated state support parameter of the corresponding structural evidence, and the degree of constraint violation corresponding to the reverse constraint relationship is obtained. The construction of the constraint violation degree specifically includes: for a reverse constraint relationship, the first violation amount is obtained by performing absolute value processing on the difference between the updated closed state support parameter corresponding to the first structural evidence and the updated unclosed state support parameter corresponding to the second structural evidence; the second violation amount is obtained by performing absolute value processing on the difference between the updated unclosed state support parameter corresponding to the first structural evidence and the updated closed state support parameter corresponding to the second structural evidence; and the first violation amount and the second violation amount are weighted and combined to obtain the constraint violation degree corresponding to the reverse constraint relationship. For each set of unidirectional constraints in the set of structural constraints, the third violation quantity is obtained by performing absolute value processing on the difference between the updated closed state support parameters corresponding to the two structural evidences, and the fourth violation quantity is obtained by performing absolute value processing on the difference between the updated unclosed state support parameters corresponding to the two structural evidences. The third and fourth violation quantities are weighted and combined to obtain the degree of constraint violation corresponding to the unidirectional constraint. Based on the degree of constraint violation corresponding to each structural constraint relationship, a comprehensive degree of constraint violation is constructed for each structural evidence. For each structural evidence, the degree of constraint violation corresponding to all structural constraint relationships related to that structural evidence is accumulated or weighted to obtain the comprehensive degree of constraint violation corresponding to that structural evidence. Global aggregation is performed based on the comprehensive constraint violation degree corresponding to each structural evidence. The comprehensive constraint violation degree corresponding to each structural evidence is accumulated or weighted to obtain the evidence conflict measurement result corresponding to the evidence set. The evidence conflict measurement result is used to uniformly characterize the overall deviation degree of structural constraint relationship between each structural evidence. The global aggregation process specifically includes: first, obtaining the comprehensive constraint violation degree corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door latch state; then, according to a unified aggregation rule, summarizing and calculating these comprehensive constraint violation degrees to form an evidence conflict measurement result that can comprehensively reflect the deviation of all structural evidence of the fire door from the structural constraint relationship.

[0027] In this embodiment, the generation of weight parameters includes: Based on standardized visual data, image sharpness, occlusion degree and continuous frame state change information are obtained. Image sharpness is quantified by the continuity of the fire door area edge, the sharpness of the outline and the degree of texture detail preservation. Occlusion degree is quantified by the proportion of the occluded part in the fire door area. Continuous frame state change information is quantified by the change amplitude of the state support parameter corresponding to each structural evidence in adjacent frames. Based on image clarity, occlusion degree, and continuous frame state change information, the reliability of the state support parameter corresponding to each structural evidence is evaluated, and the reliability evaluation value corresponding to each structural evidence is obtained. Among them, when the image clarity increases, the reliability evaluation value corresponding to each structural evidence increases; when the occlusion degree increases, the reliability evaluation value corresponding to each structural evidence decreases; when the change amplitude of the state support parameter corresponding to continuous frame state change information increases, the reliability evaluation value corresponding to each structural evidence decreases. The reliability assessment values ​​corresponding to each structural evidence are normalized so that the reliability assessment values ​​corresponding to each structural evidence are within a uniform numerical range, and the sum of the normalized reliability assessment values ​​corresponding to each structural evidence satisfies a fixed ratio constraint, thus obtaining the weight parameters corresponding to each structural evidence. The weight parameters corresponding to each structural piece of evidence are uniformly organized to obtain the weight parameters corresponding to each structural piece of evidence.

[0028] In this embodiment, the generation of the fire door status determination value includes: Based on the updated state support parameters and weight parameters corresponding to each structural evidence, closed support channels and open support channels are constructed respectively. The updated closed state support parameters and corresponding weight parameters corresponding to each structural evidence are combined to obtain the closed support contribution value corresponding to each structural evidence. The updated open state support parameters and corresponding weight parameters corresponding to each structural evidence are combined to obtain the open support contribution value corresponding to each structural evidence. Based on the evidence conflict measurement results, the closed support contribution value and the unclosed support contribution value corresponding to each structural evidence are constrained and suppressed. The evidence conflict measurement results are used as a constraint suppression factor to introduce the closed support contribution value and the unclosed support contribution value corresponding to each structural evidence, so that the closed support contribution value and the unclosed support contribution value corresponding to the structural evidence that violates the structural constraint relationship are reduced simultaneously according to the suppression ratio, and the constrained closed support contribution value and the constrained unclosed support contribution value are obtained for each structural evidence. The closed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the closed support value for the closed support channel. The unclosed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the unclosed support value for the unclosed support channel. The fire door status determination value is obtained by calculating the difference between the closed support value and the unclosed support value.

[0029] In this embodiment, the generation of the fire door status result includes: Obtain the fire door status judgment value and evidence conflict measurement result, and set the closed status judgment threshold, open status judgment threshold and ajar status judgment interval corresponding to the fire door status judgment value, as well as the abnormal status judgment threshold corresponding to the evidence conflict measurement result. The fire door status judgment value is compared with the closed status judgment threshold, the open status judgment threshold and the half-closed status judgment interval, and the evidence conflict measurement result is compared with the abnormal status judgment threshold to form the closed status judgment condition, the open status judgment condition, the half-closed status judgment condition and the abnormal status judgment condition. When the fire door status judgment value is greater than or equal to the closed status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the closed status judgment condition is met and the closed status result is output. When the fire door status judgment value is less than or equal to the open status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the open status judgment condition is met and the open status result is output. When the fire door status judgment value is within the false opening status judgment range and the evidence conflict measurement result is less than the abnormal status judgment threshold, the false opening status judgment condition is met, and the false opening status result is output. When the result of the evidence conflict measurement is greater than or equal to the abnormal state determination threshold, the abnormal state determination condition is met, and the abnormal state result is output. The current output result among the closed state result, open state result, partially closed state result, and abnormal state result is determined as the fire door state result.

[0030] Example 1: To verify the feasibility of this invention in practice, it was applied to a fire door status monitoring scenario in a large commercial complex. This complex includes shop areas, equipment rooms, underground parking areas, and multiple evacuation routes connecting different floors. Due to frequent personnel flow and continuous logistics activities, some fire doors are prone to issues such as incomplete closure, insufficient door closer return, ineffective door latch engagement, and discontinuous contact between the door frame and door leaf during daily use. Simultaneously, monitoring equipment installed on the top and walls of the passageways is limited by factors such as changes in ambient lighting, obstructions from pedestrians and trolleys, short-term stacking of goods, and differences in camera viewing angles, often resulting in incomplete local structural information in the acquired images. In such cases, relying solely on a single image classification method to determine whether a fire door is closed or open often leads to misclassification of a near-closed state with potential safety hazards as a normal closure, and misclassification of short-term abnormal images caused by obstruction as open or abnormal states. This fails to meet the requirements of continuity, accuracy, and stability in actual fire inspections.

[0031] In this scenario, visual acquisition devices are first deployed above and to the sides of the fire door to cover the door gap area, the contact area between the door frame and the door leaf, the door closer position area, and the door latch area. During daily operation, the visual acquisition devices continuously acquire images and video data of the fire door and transmit the data to an edge analysis device deployed in a local computer room. The edge analysis device first preprocesses the acquired images to eliminate uneven brightness caused by changes in aisle lighting, reduce the impact of image noise on door edge recognition, and correct perspective deviations caused by the camera installation angle, thereby obtaining standardized visual data. Based on this standardized visual data, this invention does not directly perform a single classification of the entire image. Instead, it first extracts structural-level visual features closely related to the physical closure state of the fire door, including door gap width features, door frame and door leaf contact continuity features, door leaf posture angle features, door closer position features, and door latch state features. Subsequently, these structural-level visual features are converted into corresponding structural evidence, and state support parameters are further constructed.

[0032] In practical applications, the wider the door gap, the further the door is from a closed state; the better the continuity of contact between the door frame and the door leaf, the closer the door is to a stable closure; the further the door leaf's posture angle deviates from the closed position, the more likely the fire door is to be in an open or ajar state; if the door closer does not return to the end position, it often means that the door has not returned to the closed end point; if the door latch does not enter the engaging position, it means that although the door may appear to be close to closed, it has not actually formed an effective lock.

[0033] After these disparate structural information pieces are uniformly converted into structural evidence, the system then processes them in conjunction with the inherent relationships that the structural evidence should satisfy. The consistency constraints of the state support parameters corresponding to each structural evidence are corrected, and the corrected state support parameters are further processed through a deep random vector function connection network to obtain the updated state support parameters.

[0034] After obtaining the updated state support parameters, this invention further constructs an evidence set and calculates the degree of constraint violation between each piece of structural evidence based on structural constraint relationships, obtaining the evidence conflict measurement result. Subsequently, the system combines image clarity, occlusion degree, and continuous frame state change information to perform a reliability assessment on the state support parameters corresponding to each piece of structural evidence, generates weight parameters, and constructs closed support channels and open support channels respectively. The evidence conflict measurement result is introduced into the two support channels for constraint suppression processing, finally obtaining the fire door state judgment value and the fire door state result.

[0035] To demonstrate that the present invention can achieve the expected beneficial effects, in the above-mentioned scenario, images and video data of fire doors were recorded over a long period of time, from different floors and from different shooting angles during continuous operation. Fire safety management personnel then marked and preserved the actual state of the fire doors based on on-site verification. The recorded data includes not only the door closing process under normal passage conditions, but also the state changes under various scenarios such as frequent personnel passage, obstruction by goods transport, changes in passage illumination, slow door return, obstructed door closer return, and incomplete door latch engagement.

[0036] By comparing the identification results of this invention with on-site verification records, it can be seen that this invention demonstrates good adaptability in identifying fire doors in closed, partially ajar, and abnormal states. Especially in scenarios with locally inconsistent and contradictory structural features, this invention does not rely solely on a single image representation as in conventional methods. Instead, it employs a combined mechanism of structural evidence, structural constraint relationships, evidence conflict measurement results, and dual-channel constraint suppression processing to more reliably identify the true state. This demonstrates the good feasibility of this invention in real-world building fire monitoring environments.

[0037] Table 1 Comparison of Core Performance of Fire Door Status Recognition

[0038] As shown in Table 1, the overall state recognition accuracy of traditional single-frame visual classification methods is 89.6%, while the multi-feature weighted fusion method improves it to 91.1%, and the method of this invention further improves it to 92.4%. This change indicates that while relying solely on whole-image classification can complete basic recognition, it is easily affected by the visual performance of a single region when there is channel occlusion, insufficient repositioning, or inconsistencies in local structures. The multi-feature weighted fusion method has improved the overall recognition level by introducing multiple structural information elements. The method of this invention further establishes structural evidence, state support parameters, and structural constraint relationships, forming a verifiable and correctable decision chain between different structural information elements, thus resulting in a more stable overall recognition result.

[0039] Looking at the most challenging indicators that best demonstrate the value of this invention, the accuracy rate for identifying occlusion states increased from 72.8% to 89.6%, the accuracy rate for identifying abnormal states increased from 75.4% to 90.8%, and the accuracy rate for complex occlusion scenes increased from 68.9% to 85.5%. This is because although the multi-feature weighted fusion method uses multiple local features, these features are still used in parallel, lacking a unified constraint on the inherent structural relationship between door gaps, contact continuity, posture, door closer position, and door latch state. In contrast, this invention, through consistency constraint correction and evidence conflict measurement results calculation, has a stronger ability to determine occlusion and abnormal scenes.

[0040] From the perspective of risk control indicators, the false positive rate and false negative rate of traditional single-frame visual classification methods are 8.9% and 9.7%, respectively. The multi-feature weighted fusion method reduces these rates to 5.1% and 6.0%, while the method of this invention further reduces them to 7.7% and 3.2%. This indicates that this invention does not simply improve the detection capability of a single state, but rather achieves a better balance between reducing false positives and false negatives. The main reason for this is that after generating the updated state support parameters, this invention further generates weight parameters based on image clarity, occlusion degree, and continuous frame state change information. Then, it introduces evidence conflict measurement results in both closed and open support channels for constraint suppression processing, reducing the contribution of structural evidence that violates structural constraints. Therefore, it is less likely to be misled by local reflections, short-term occlusion, or local component offsets.

[0041] In terms of operational efficiency, the average single-shot decision latency of traditional single-frame visual classification methods is 1.43 seconds, and that of multi-feature weighted fusion methods is 1.06 seconds. The method of this invention reduces this to 0.81 seconds. This result demonstrates that the present invention not only achieves higher accuracy but also responds faster under continuous edge monitoring conditions. This is because the present invention reduces the additional time overhead caused by complex iterative optimization while retaining the nonlinear correction capability at the structural evidence level, thus maintaining both recognition accuracy and timeliness. In summary, the data in Table 1 illustrates that the method of this invention outperforms the two compared methods in terms of conventional state recognition, complex scene recognition, anomaly risk control, and real-time decision-making.

[0042] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for fire door status recognition based on AI vision, characterized in that, Includes the following steps: The collected data is preprocessed to obtain standardized visual data; Based on standardized visual data, structural-level visual features of fire doors are extracted; Evidence mapping is performed on structural-level visual features to convert each structural-level visual feature into corresponding structural evidence and construct state support parameters. Construct structural constraint relationships between structural evidences and perform consistency constraint correction to obtain the state support parameters after constraint correction; The structural-level visual features are input into the deep random vector functional connection network for nonlinear mapping and feature fusion processing. The state support parameters after constraint correction are further corrected to obtain the updated state support parameters. An evidence set is constructed based on the updated state support parameters, and the degree of constraint violation between each piece of structural evidence is calculated based on the structural constraint relationship to obtain the evidence conflict measurement result. Based on standardized visual data, the reliability of the state support parameters corresponding to each structural evidence is evaluated to obtain the weight parameters corresponding to each structural evidence. Based on the updated state support parameters and weight parameters, closed support channels and open support channels are constructed respectively, and the evidence conflict measurement results are introduced for constraint suppression processing to obtain the fire door state judgment value. The fire door status result is output based on the fire door status judgment value and the evidence conflict measurement result.

2. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The preprocessing includes noise reduction, brightness normalization, and geometric correction.

3. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The structural visual features include door gap width features, door frame and door leaf contact continuity features, door leaf posture angle features, door closer position features, and door lock tongue status features.

4. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The construction of the state support parameters includes: Evidence mapping processing is performed on the door gap width feature to convert the door gap width feature into door gap width structural evidence. State quantization processing is performed based on the difference between the feature value corresponding to the door gap width structural evidence and the door gap width benchmark value. The closed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically decreasing mapping relationship, and the unclosed state support parameter corresponding to the door gap width structural evidence is obtained through a monotonically increasing mapping relationship complementary to the closed state support parameter. Evidence mapping processing is performed on the contact continuity features between the door frame and the door leaf to convert the contact continuity features between the door frame and the door leaf into contact continuity structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the contact continuity structure evidence are obtained. Evidence mapping processing is performed on the door leaf attitude angle features to convert the door leaf attitude angle features into door leaf attitude structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door leaf attitude structure evidence are obtained. Evidence mapping processing is performed on the door closer position features to convert them into door closer position structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door closer position structure evidence are obtained. Evidence mapping processing is performed on the door latch state features to convert them into door latch state structure evidence, and the support parameters of the closed state and the support parameters of the unclosed state corresponding to the door latch state structure evidence are obtained. The support parameters for the closed state and the support parameters for the unclosed state corresponding to the structural evidence of door gap width, structural evidence of contact continuity, structural evidence of door leaf posture, structural evidence of door closer position, and structural evidence of door lock tongue state are uniformly combined to form the state support parameters.

5. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the constraint-corrected state support parameters includes: Based on the state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state, a set of structural constraint relationships is constructed. For each set of structural constraint relationships in the set of structural constraint relationships, the constraint deviation value is calculated based on the state support parameter of the corresponding structural evidence. Based on the constraint deviation values ​​corresponding to each structural constraint relationship, a comprehensive constraint deviation value is constructed for each structural evidence. Based on the comprehensive constraint deviation value corresponding to each structural evidence, the state support parameter corresponding to each structural evidence is subjected to consistency constraint correction processing, thereby obtaining the constraint-corrected state support parameter corresponding to each structural evidence. The constraint-corrected state support parameters corresponding to the structural evidence of door gap width, contact continuity, door leaf posture, door closer position, and door lock tongue state are uniformly organized to obtain the constraint-corrected state support parameters.

6. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the updated state support parameters includes: The door gap width feature, the door frame and door leaf contact continuity feature, the door leaf posture angle feature, the door closer position feature, and the door lock tongue state feature are combined in a unified order to form a structural visual feature input sequence. The constraint-corrected state support parameters corresponding to the door gap width structural evidence, contact continuity structural evidence, door leaf posture structural evidence, door closer position structural evidence, and door lock tongue state structural evidence are combined in the same order to form a constraint-corrected state support parameter input sequence. The structural visual feature input sequence is fused with the constraint-corrected state support parameter input sequence to construct a joint input sequence, and the joint input sequence is input into the deep random vector function connection network. The deep stochastic vector functional connection network includes a stochastic mapping layer, a direct connection fusion layer, a multi-layer progressive modeling layer, and a state support correction layer. In the random mapping layer, random weight mapping and nonlinear activation processing are performed on the joint input sequence to obtain random mapping features. In the direct connection fusion layer, the random mapping features and the joint input sequence are directly concatenated to obtain fused features. In the multi-layer progressive modeling layer, the fused features of the current layer are input into the next layer, and random weight mapping, nonlinear activation processing, and direct connection concatenation processing are repeatedly performed to obtain a multi-layer fused feature sequence. In the state support correction layer, the fused features of the final layer are used as correction input to generate the closed state support correction amount and the open state support correction amount corresponding to each structural evidence. Based on the support corrections for the closed and open states corresponding to each structural piece of evidence, constraint coupling update processing is performed on the corresponding constraint-corrected support parameters for the closed and open states. Complementary constraints are then applied to the two types of state support parameters corresponding to the same structural piece of evidence, thereby obtaining the updated state support parameters for each structural piece of evidence.

7. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the evidence conflict measurement results includes: Based on the updated state support parameters corresponding to the structural evidence of door gap width, structural evidence of contact continuity, structural evidence of door leaf posture, structural evidence of door closer position, and structural evidence of door lock tongue state, an evidence set is constructed. For each set of structural constraint relationships in the set of structural constraint relationships, the degree of constraint violation is constructed based on the updated state support parameter of the corresponding structural evidence, and the degree of constraint violation corresponding to the reverse constraint relationship is obtained. For each set of unidirectional constraints in the structural constraint relationship set, the degree of constraint violation corresponding to the unidirectional constraint relationship is obtained; Based on the degree of constraint violation corresponding to each structural constraint relationship, a comprehensive degree of constraint violation is constructed for each structural evidence. The global aggregation process is performed based on the degree of comprehensive constraint violation corresponding to each structural piece of evidence to obtain the evidence conflict measurement result corresponding to the evidence set.

8. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the weight parameters includes: Information on image sharpness, occlusion degree, and continuous frame state changes is obtained based on standardized visual data. Based on image clarity, occlusion degree and continuous frame state change information, the reliability of the state support parameters corresponding to each structural evidence is evaluated, and the reliability evaluation value of each structural evidence is obtained. The reliability assessment values ​​corresponding to each structural piece of evidence are normalized to obtain the weight parameters corresponding to each structural piece of evidence. The weight parameters corresponding to each structural piece of evidence are uniformly organized to obtain the weight parameters corresponding to each structural piece of evidence.

9. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the fire door status determination value includes: Based on the updated state support parameters corresponding to each structural evidence and the weight parameters corresponding to each structural evidence, closed support channels and open support channels are constructed respectively. Based on the evidence conflict measurement results, the closed support contribution value and the unclosed support contribution value corresponding to each structural evidence are constrained and suppressed to obtain the constrained and suppressed closed support contribution value and the constrained and suppressed unclosed support contribution value corresponding to each structural evidence. The closed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the closed support value for the closed support channel. The unclosed support contribution values ​​after constraint suppression for each structural evidence are summed or weighted to obtain the unclosed support value for the unclosed support channel. The fire door status determination value is obtained by calculating the difference between the closed support value and the unclosed support value.

10. The method for fire door status recognition based on AI vision according to claim 1, characterized in that, The generation of the fire door status result includes: Obtain the fire door status judgment value and evidence conflict measurement result, and set the closed status judgment threshold, open status judgment threshold and ajar status judgment interval corresponding to the fire door status judgment value, as well as the abnormal status judgment threshold corresponding to the evidence conflict measurement result. The fire door status judgment value is compared with the closed status judgment threshold, the open status judgment threshold and the half-closed status judgment interval, and the evidence conflict measurement result is compared with the abnormal status judgment threshold to form the closed status judgment condition, the open status judgment condition, the half-closed status judgment condition and the abnormal status judgment condition. When the fire door status judgment value is greater than or equal to the closed status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the closed status judgment condition is met and the closed status result is output. When the fire door status judgment value is less than or equal to the open status judgment threshold and the evidence conflict measurement result is less than the abnormal status judgment threshold, the open status judgment condition is met and the open status result is output. When the fire door status judgment value is within the false opening status judgment range and the evidence conflict measurement result is less than the abnormal status judgment threshold, the false opening status judgment condition is met, and the false opening status result is output. When the result of the evidence conflict measurement is greater than or equal to the abnormal state determination threshold, the abnormal state determination condition is met, and the abnormal state result is output. The current output result among the closed state result, open state result, partially closed state result, and abnormal state result is determined as the fire door state result.

Images