A method for automatically constructing an overhead line support equipment file based on multi-frame visual sequence alignment
By using multi-frame visual sequence alignment and the Viterbi algorithm, combined with the uniqueness of pole bounding boxes and kilometer markers, the problems of low accuracy and disordered order in the construction of catenary support pole files were solved, and high-precision automatic construction of catenary support pole equipment files was achieved.
Patent Information
- Application Number
- CN202610356981.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-03-23
- Publication Date
- 2026-07-03
AI Technical Summary
Existing technologies struggle to accurately identify the location of overhead contact line supports in high-speed driving scenarios, resulting in low accuracy in constructing overhead contact line support archives and an inability to effectively handle missed detections, false detections, and sequence errors.
A method based on multi-frame visual sequence alignment is adopted, and global optimal alignment is performed through the Viterbi algorithm. Combined with the trajectory of the change in the area of the pole boundary frame and the uniqueness of the kilometer marker, an archive of catenary support equipment is constructed.
It significantly improved the accuracy and automation of the construction of overhead contact line support equipment files, effectively solved the problems of missed detections, false detections, and disordered sequences, and improved the accuracy and robustness of character recognition.
Smart Images

Figure CN122336621A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an automatic construction method for overhead contact line support equipment files based on multi-frame visual sequence alignment, belonging to the field of railway operation and maintenance technology. Background Technology
[0002] With the continuous advancement of railway electrification and intelligent operation and maintenance systems in my country, the demand for digital management of trackside overhead contact lines is constantly increasing. Among these, the "one pole, one file" information—including pole identification, pole number information, and spatial location information such as kilometer markers—is a crucial foundation for achieving trackside asset positioning, section management, hazard tracking, and alignment of multi-source inspection data. Currently, during routine inspections, high-speed trains or track inspection vehicles typically use a front-mounted camera to continuously capture trackside images along the line. Each captured pole and its number constitute a visual record of the overhead contact line assets. By accurately identifying pole numbers and automatically archiving them at the moment of capture, the system can compress inspection data from massive amounts of video into sparse, structured key asset information, significantly reducing storage and management costs.
[0003] Currently, the industry commonly uses a pipelined solution based on single-frame detection and optical character recognition (OCR) for pole number identification, supplemented by the Global Positioning System (GPS) for spatiotemporal reference. However, in high-speed driving scenarios, GPS is susceptible to factors such as insufficient sampling frequency, positioning jitter, and signal attenuation, making it difficult to accurately correspond to specific pole locations. Simultaneously, due to factors such as motion blur, changes in lighting, license plate reflection, and rapid changes in viewing angle, single-frame OCR recognition has poor stability, easily leading to misreading, missed readings, and detection frame drift. Furthermore, the dense arrangement of trackside poles, the overlap of multiple poles, and local variations in pole spacing increase the risk of continuity judgments based on simple geometric rules. Moreover, local OCR errors can propagate along the time dimension, causing jumps or breaks in the pole number chain, making it difficult to generate accurate and reliable contact wire support records. Summary of the Invention
[0004] The purpose of this invention is to provide an automatic construction method for overhead contact line support equipment files based on multi-frame visual sequence alignment. This method compresses multi-frame candidate data identified in inspection videos into pole nodes and uses the Viterbi algorithm to perform global optimal alignment with the actual pole number sequence. This solves the problems of low accuracy in overhead contact line support file construction, inability to effectively handle missed detections, false detections, and disordered sequences in existing technologies.
[0005] To solve the above-mentioned technical problems, the present invention is implemented using the following technical solution.
[0006] This invention provides a method for automatically constructing overhead contact line support equipment archives based on multi-frame visual sequence alignment, comprising:
[0007] Real-time acquisition of trackside inspection videos, and extraction of the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence;
[0008] Based on the area change trajectory of the rod boundary box, under the preset continuous empty frame constraint, the frame data sequence is segmented to obtain several rod trajectories.
[0009] The pole trajectory is compressed into candidate pole nodes, and representative frames, kilometer marker area characters, license plate area characters, and pole type labels are extracted to obtain a sequence of candidate pole nodes arranged by time.
[0010] Obtain pre-stored real structured data from the trackside, and based on the uniqueness of kilometer markers, construct a key point dictionary using pole type labels, interval distances, row types, and pole numbers as indexes to generate a real pole number sequence arranged in line order;
[0011] For each real pole number and candidate pole node, based on the kilometer marker character matching degree, pole number plate character matching degree, and pole type label matching degree, all video frames covered by the candidate pole node are scored frame by frame, and the matching score between the real pole number and the candidate pole node is calculated.
[0012] Based on the matching score, the Viterbi algorithm is used to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence; wherein, during the solution process, skipping candidate pole nodes and / or missing matching of real pole numbers is allowed, and a jump smoothing penalty term based on the node center frame spacing is introduced to constrain the path jump;
[0013] Based on the globally optimal alignment path, the corresponding real pole number is marked and written back to the candidate pole node sequence to generate the contact network support equipment file.
[0014] Furthermore, the license plate area characters / kilometer marker area characters with the highest confidence level are extracted from each frame to obtain the frame data sequence, including:
[0015] Discretize the trackside inspection video into an ordered frame sequence;
[0016] Based on the ordered frame sequence, target detection and text recognition are performed frame by frame to extract the original frame data sequence, which includes at least the frame number, timestamp, license plate area bounding box, kilometer marker area bounding box, pole bounding box, license plate area characters and kilometer marker area characters.
[0017] Based on the original frame data sequence, the frame data sequence is obtained by extracting the license plate area character / kilometer marker area character with the highest confidence in each frame.
[0018] If the first character of the license plate area character / kilometer marker area character is "K" or "k", it is included in the kilometer marker set, and the character with the highest confidence is selected as the kilometer marker of the current frame;
[0019] If the first character of the license plate area characters / kilometer marker area characters is not "K" or "k", then the license plate area characters / kilometer marker area characters are normalized and included in the candidate pole node set.
[0020] Furthermore, based on the area change trajectory of the rod boundary box, under the preset constraint of continuous empty frames, the frame data sequence is segmented to obtain several rod trajectories, including:
[0021] Based on the frame bounding box of each frame in the frame data sequence, calculate the area of the frame bounding box in chronological order;
[0022] Valid frames are selected based on the area change trajectory of the rod bounding box:
[0023] If the current frame is missing a pole bounding box, or if the area of the pole bounding box in the current frame is less than a preset effective area threshold, then the current frame is marked as a poleless frame.
[0024] The frame data sequence is segmented according to the area change trajectory of the rod boundary box and the preset continuous empty frame constraint condition, and the sequence is traversed from the starting frame of the frame data sequence.
[0025] When the length of consecutive frames marked as poleless exceeds the preset maximum empty frame threshold, or when the ratio of the area of the pole bounding box of the current frame to the historical maximum area of the current pole trajectory is lower than the preset area decrease ratio threshold, the current pole trajectory is determined to end.
[0026] In response to the determination that the current pole trajectory has ended, the current pole trajectory ends and the recording of the next pole trajectory begins;
[0027] Traverse all frames to generate several pole trajectory segments containing consecutive valid frames, thus obtaining several pole trajectories.
[0028] Furthermore, the preset continuous empty frame constraint condition is: allowing the existence of continuous poleless frames not exceeding a first threshold and non-continuous poleless frames not exceeding a second threshold within a single pole trajectory.
[0029] Furthermore, the pole trajectory is compressed into candidate pole nodes, and representative frames, kilometer marker area characters, license plate area characters, and pole type labels are extracted to obtain a time-ordered sequence of candidate pole nodes, including:
[0030] The representative frame is determined based on the area of the bounding box of the pole in the pole trajectory. The frame with the largest bounding box area in the pole trajectory is selected as the representative frame of the current pole trajectory.
[0031] The kilometer marker string is determined based on the kilometer marker region characters of each frame within the pole trajectory. From the frame data sequence belonging to the pole trajectory, all frames that are included in the kilometer marker set and their corresponding kilometer marker region characters are extracted. The frequency of occurrence of each kilometer marker region character is counted, and the character with the highest frequency is selected as the kilometer marker string of the current pole trajectory. If there are no frames included in the kilometer marker set within the current pole trajectory, the kilometer marker string of the current pole trajectory is marked as empty.
[0032] The pole type label of the current pole trajectory is determined based on the pole type label of each frame within the pole trajectory. The pole type labels identified in each frame within the current pole trajectory are traversed to construct a set of type labels.
[0033] If the type label set contains a preset special type label, then the special type label is used as the pole type label of the current pole trajectory;
[0034] If the set of type labels does not contain preset special type labels but contains regular type labels, then the regular type labels will be used as the pole type labels for the current pole trajectory;
[0035] If the set of type labels is empty, then the pole type label of the current pole trajectory will be marked as unknown;
[0036] The license plate area characters are determined based on the license plate area characters of each frame within the pole trajectory. From the frame data sequence belonging to the current pole trajectory, all frames that are included in the candidate pole node set and their corresponding license plate area characters are extracted to construct a pole number candidate set. The character with the highest confidence is selected as the license plate area character of the current pole trajectory. If there are no frames in the current pole trajectory that are included in the candidate pole node set, the license plate area characters of the current pole trajectory are marked as empty.
[0037] Based on the representative frame, kilometer marker string, pole type label, and license plate area characters, candidate pole nodes are constructed and arranged in chronological order to obtain a candidate pole node sequence.
[0038] Furthermore, pre-stored trackside real structured data is acquired. Based on the uniqueness of kilometer markers, a key point dictionary is constructed using pole type labels, interval distances, row types, and pole numbers as indexes. This generates a sequence of real pole numbers arranged in line order, including:
[0039] Obtain pre-stored trackside real structured data, which includes at least the section segment, row, pole number, pole type label, and physical pole information of the kilometer marker;
[0040] Based on the uniqueness of kilometer markers, a key point dictionary is constructed using interval segments, row types, and pole numbers as combined indexes. Each combined index uniquely identifies a physical pole, forming a set of real pole number primary keys.
[0041] Based on the primary key set of real pole numbers, extract the kilometer marker and pole type label corresponding to each physical pole, and generate a real pole number sequence in the order of the route. Each real pole number contains a pole number, the corresponding kilometer marker, and a pole type label.
[0042] Furthermore, the matching score between the actual pole number and the candidate pole node is expressed as:
[0043] ;
[0044] In the formula, Indicates the first The first real pole number and the first Matching scores between candidate pole nodes Indicates the first The first real pole number and the first Kilometer marker character matching degree among candidate pole nodes Indicates the first The first real pole number and the first The degree of matching of pole type labels among candidate pole nodes. Indicates the first The first real pole number and the first The degree of character matching of pole number plates among the candidate pole nodes, where... Indicates the first The first real pole number and the first Step index during the matching process of candidate pole nodes;
[0045] The kilometer marker character matching degree is represented as follows:
[0046] ;
[0047] In the formula, Indicates the first The first real pole number and the first Matching score of kilometer markers between candidate pole nodes Preset weights, , This represents the matching function. Indicates the first A real pole number, kilometer marker character, Indicates the first The kilometer marker characters of each candidate pole node, This indicates the score when both the kilometer marker and the actual pole number match simultaneously. The preset weight for the score when both the kilometer marker and the actual pole number match simultaneously;
[0048] The matching degree of the rod type label is expressed as follows:
[0049] ;
[0050] In the formula, Indicates the first The first real pole number and the first Link type label matching score among candidate link nodes Preset weights, , No. A real pole number and pole type label. Indicates the first The pole type label of each candidate pole node. The score indicates that both the pole type label and the actual pole number match.
[0051] The character matching degree of the pole number plate is represented as follows:
[0052] ;
[0053] In the formula, Indicates the first The first real pole number and the first Pocket number matching score between candidate pole nodes Preset weights, , Indicates the first The pole number characters of each candidate pole node Indicates the first A real pole number plate character.
[0054] Further, based on the matching score, the Viterbi algorithm is used to solve for the globally optimal alignment path from the candidate pole node sequence to the actual pole number sequence, including:
[0055] Based on the length of the actual pole number sequence and the length of the candidate pole node sequence, a cumulative score matrix and a corresponding state transition precursor matrix are constructed. The cumulative score at the starting position is initialized to a preset value, and the cumulative score at the remaining positions in the state transition precursor matrix is initialized to negative infinity.
[0056] The frame span covered by the current candidate pole node in the video frame sequence is obtained. Based on the preset base tolerance, span tolerance coefficient, and the distance between the center frame of the previously matched candidate pole node and the center frame of the current candidate pole node, an adaptive jump smoothing penalty term is determined. Wherein, when the distance exceeds the sum of the base tolerance and the span tolerance coefficient multiplied by the frame span, the adaptive jump smoothing penalty term is calculated based on the excess part of the distance and the preset penalty intensity coefficient. If there is no previously matched candidate pole node or the current is the first match of the path, the adaptive jump smoothing penalty term is set to zero.
[0057] Set a maximum allowed number of consecutive candidate pole nodes to be crossed. Iterate through each position in the cumulative score matrix. Under the condition that the difference between the current candidate pole node index and the previous matching node index does not exceed the maximum allowed number of consecutive candidate pole nodes to be crossed, calculate the cumulative score for three states: matching the real pole number with a candidate pole node, skipping a candidate pole node, and missing a match on the real pole number. Record the state action that maximizes the cumulative score and the corresponding predecessor node index. The cumulative score for matching the real pole number with a candidate pole node is the previous cumulative score plus the matching score minus the adaptive jump smoothing penalty. The cumulative score for skipping a candidate pole node is the previous cumulative score minus a preset skip penalty. The cumulative score for missing a match on the real pole number is the previous cumulative score minus a preset missing match penalty.
[0058] The globally optimal cumulative score is obtained at the end of the cumulative score matrix. Based on the path actions recorded in the state transition predecessor matrix, the process is reversed from the end to the starting position to generate a globally optimal alignment path that includes the mapping relationship between the real pole number and the candidate pole node.
[0059] Furthermore, the recursive relationship of the state transition predecessor matrix is expressed as follows:
[0060] ;
[0061] In the formula, Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This indicates the operation of retrieving the maximum value. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This represents a preset constant penalty term for skipping candidate pole nodes. This represents a preset constant penalty term for missing matching of the actual pole number. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. Indicates the first The candidate pole node is matched with the first... When the actual pole number is 1, an adaptive jump smoothing penalty term is applied relative to the previous matched node to constrain the jump distance of adjacent matched nodes on the time axis.
[0062] Furthermore, the adaptive jump smoothing penalty term is expressed as:
[0063] ;
[0064] In the formula, Indicates the first The candidate pole node is matched with the first... When the actual pole number is [number], an adaptive jump smoothing penalty term is applied relative to the previous matched node. This represents the preset penalty intensity coefficient. This represents the distance between the center frame of the current candidate pole node and the center frame of the previously successfully matched candidate pole node. , This indicates the center frame number of the current candidate pole node within the frame sequence it covers. This indicates the center frame number of the previously matched candidate pole node. This indicates the permissible inter-frame spacing tolerance. , Indicates the basic tolerance level. This is the span tolerance coefficient. Indicates the first The frame span covered by each candidate pole node in the video frame sequence. Indicates other situations.
[0065] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:
[0066] 1. This invention achieves the solution of the globally optimal alignment path from the candidate pole node sequence to the actual pole number sequence by segmenting the pole trajectory based on multi-frame visual sequence alignment and generating candidate pole node sequences, combined with the key point dictionary index constructed by the uniqueness of kilometer markers, and by introducing a jump smoothing penalty term in the matching score calculation using the Viterbi algorithm. This effectively solves the problems of low accuracy in the construction of catenary support pole files, inability to effectively handle missed detections, false detections and disordered sequences in the existing technology, and significantly improves the accuracy and automation of the construction of catenary support pole equipment files.
[0067] 2. This invention extracts the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence, and intelligently classifies them into the kilometer marker set or candidate pole node set according to whether the first character is "K" or "k". This effectively filters out possible misidentification and noise interference in single-frame recognition, ensuring that subsequent processing is based on the most reliable recognition results, and significantly improves the accuracy and robustness of character recognition in complex trackside environments such as changes in lighting, dirt, and vibration.
[0068] 3. This invention, based on the area change trajectory of the pole bounding box, segments the frame data sequence under the preset constraint of continuous empty frames to obtain several pole trajectories. By utilizing the area change pattern of the pole as it moves from far to near and then away in the video sequence, and allowing a small number of empty frames within the trajectory, it can accurately cluster video frames belonging to the same pole in a continuous video stream into a complete pole trajectory. This avoids trajectory breakage or incorrect merging caused by brief occlusion or detection failure, providing high-quality spatiotemporal segments for subsequent pole recognition.
[0069] 4. This invention uses the Viterbi algorithm based on the matching score to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence, and introduces an adaptive jump smoothing penalty term to constrain the jump distance between adjacent matching nodes on the time axis. It can comprehensively consider character matching degree, pole type matching degree, and the continuity of the time series in a global scope, and solve the optimal matching sequence between candidate nodes and real pole numbers for the entire line in one go. It effectively solves the problem of local matching errors caused by false detection, missed detection, or character similarity, and realizes high-precision automatic alignment between visual detection results and existing ledger data. Attached Figure Description
[0070] Figure 1 This is a flowchart illustrating an automatic construction method for overhead contact line support equipment files based on multi-frame visual sequence alignment, provided by an embodiment of the present invention. Detailed Implementation
[0071] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.
[0072] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.
[0073] Example 1
[0074] like Figure 1 As shown in the figure, this embodiment introduces an automatic construction method for overhead contact line support equipment archives based on multi-frame visual sequence alignment, including:
[0075] Step 1: Real-time acquisition of trackside inspection video, and extraction of the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence.
[0076] This embodiment extracts the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence. In the initial stage of collecting trackside inspection videos, the quality of the single frame recognition results is optimized, effectively filtering out low-confidence misidentifications caused by momentary occlusion, reflection or dirt.
[0077] Step 2: Based on the area change trajectory of the rod boundary box, under the preset continuous empty frame constraint, the frame data sequence is segmented to obtain several rod trajectories.
[0078] This embodiment uses the area change trajectory of the pole bounding box to segment the frame data sequence under the preset constraint of continuous missing frames, obtaining several pole trajectories. By utilizing the area change pattern of the pole from far to near in the video and allowing a small number of detection gaps within the trajectory, it can robustly and accurately segment visual segments belonging to the same pole in continuous frames, effectively avoiding trajectory breaks caused by brief missed detections.
[0079] Step 3: Compress the pole trajectory into candidate pole nodes, extract representative frames, kilometer marker area characters, license plate area characters, and pole type labels to obtain a sequence of candidate pole nodes arranged by time.
[0080] This embodiment compresses the pole trajectory into candidate pole nodes, extracts representative frames, kilometer marker area characters, license plate area characters, and pole type labels, and obtains a sequence of candidate pole nodes arranged by time. This step condenses the redundant video frame sequence into structured node information, and determines the optimal characters and pole types from multiple trajectory frames through a statistical voting mechanism. This not only eliminates single-frame random errors, but also transforms the disordered video stream into an ordered list of nodes that can be matched.
[0081] Step 4: Obtain the pre-stored real structured data of the trackside. Based on the uniqueness of the kilometer markers, construct a key point dictionary with pole type labels, interval distances, row types and pole numbers as indexes, and generate a real pole number sequence arranged in the order of the line.
[0082] This embodiment acquires pre-stored real structured data from the trackside, and based on the uniqueness of kilometer markers, constructs a key point dictionary using pole type labels, interval distances, row types, and pole numbers as indexes. It generates a real pole number sequence arranged in the order of the line, transforms existing ledger data into standardized matching templates, and uses multi-dimensional combined indexes to ensure the unique identification of each physical pole.
[0083] Step 5: For each real pole number and candidate pole node, score all video frames covered by the candidate pole node frame by frame based on the matching degree of kilometer marker characters, pole number plate characters, and pole type label, and calculate the matching score between the real pole number and the candidate pole node.
[0084] This embodiment calculates the matching score between the real pole number and the candidate pole node by scoring all video frames covered by the candidate pole node frame by frame based on the matching degree of kilometer marker characters, pole number plate characters, and pole type label. The matching probability is comprehensively and quantitatively evaluated from multiple dimensions such as character content, pole attributes, and visual quality.
[0085] Step 6: Based on the matching score, use the Viterbi algorithm to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence; wherein, during the solution process, skipping candidate pole nodes and / or missing matching of real pole numbers is allowed, and a jump smoothing penalty term based on the node center frame spacing is introduced to constrain the path jump.
[0086] This embodiment uses the Viterbi algorithm based on the matching score to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence. During the solution process, skipping candidate pole nodes and / or missing matching of real pole numbers is allowed, and a jump smoothing penalty term based on the node center frame spacing is introduced to constrain path jumps. This step searches for the optimal matching sequence globally. By introducing the penalty term and jump constraints, it effectively tolerates false detections and missed detections during the detection process, while ensuring the continuity of the matching path on the time axis, thus achieving globally optimal alignment for long sequence data.
[0087] Step 7: Based on the global optimal alignment path, mark the corresponding real pole number and write it back to the candidate pole node sequence to generate the catenary support equipment file.
[0088] This embodiment generates a catenary support equipment file by marking the corresponding real pole number according to the globally optimal alignment path and writing it back to the candidate pole node sequence. This step transforms the abstract optimal path into specific business data, completing the automatic linking from visual detection to existing ledgers, and finally generating a structured file containing accurate pole number information, realizing the automated construction and updating of the catenary support equipment file.
[0089] Example 2
[0090] Based on the same inventive concept as Embodiment 1, this embodiment introduces the implementation steps of an automatic construction method for contact wire support equipment files based on multi-frame visual sequence alignment, including:
[0091] Step 1: Real-time acquisition of trackside inspection video, and extraction of the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence.
[0092] Step 1.1: Discretize the trackside inspection video into an ordered frame sequence.
[0093] Step 1.2: Based on the ordered frame sequence, target detection and text recognition are performed frame by frame to extract the original frame data sequence, which includes at least the frame number, timestamp, license plate area bounding box, kilometer marker area bounding box, pole bounding box, license plate area characters, and kilometer marker area characters.
[0094] Step 1.3: Based on the original frame data sequence, obtain the license plate area character / kilometer marker area character with the highest confidence in each frame to obtain the frame data sequence.
[0095] If the first character of the license plate area character / kilometer marker area character is "K" or "k", it is included in the kilometer marker set, and the character with the highest confidence is selected as the kilometer marker of the current frame;
[0096] If the first character of the license plate area characters / kilometer marker area characters is not "K" or "k", then the license plate area characters / kilometer marker area characters are normalized and included in the candidate pole node set.
[0097] Step 2: Based on the area change trajectory of the rod boundary box, under the preset continuous empty frame constraint, the frame data sequence is segmented to obtain several rod trajectories.
[0098] In this embodiment, the preset continuous empty frame constraint condition is: allowing the existence of continuous poleless frames not exceeding a first threshold and non-continuous poleless frames not exceeding a second threshold within a single pole trajectory.
[0099] Step 2.1: Calculate the area of the rod bounding box in chronological order based on the rod bounding box of each frame in the frame data sequence.
[0100] Step 2.2: Filter valid frames based on the area change trajectory of the rod bounding box:
[0101] If the current frame is missing a pole bounding box, or if the area of the pole bounding box in the current frame is less than a preset effective area threshold, then the current frame is marked as a poleless frame.
[0102] Step 2.3: Segment the frame data sequence according to the area change trajectory of the rod boundary frame and the preset continuous empty frame constraint conditions, and traverse backward from the starting frame of the frame data sequence.
[0103] Step 2.3: When the length of consecutive frames marked as poleless exceeds the preset maximum empty frame threshold, or when the ratio of the area of the pole bounding box of the current frame to the historical maximum area of the current pole trajectory is lower than the preset area decrease ratio threshold, the current pole trajectory is determined to end.
[0104] Step 2.4: In response to the determination that the current pole trajectory has ended, end the current pole trajectory and start recording the next pole trajectory.
[0105] Step 2.5: Traverse all frames and generate several pole trajectory segments containing consecutive valid frames to obtain several pole trajectories.
[0106] Step 3: Compress the pole trajectory into candidate pole nodes, extract representative frames, kilometer marker area characters, license plate area characters, and pole type labels to obtain a sequence of candidate pole nodes arranged by time.
[0107] Step 3.1: Determine the representative frame based on the area of the bounding box of the pole in the pole trajectory. The frame with the largest bounding box area in the pole trajectory is selected as the representative frame of the current pole trajectory.
[0108] Step 3.2: Determine the kilometer marker string based on the kilometer marker region characters of each frame within the pole trajectory. Extract all frames included in the kilometer marker set and their corresponding kilometer marker region characters from the frame data sequence belonging to the pole trajectory. Count the frequency of occurrence of each kilometer marker region character and select the character with the highest frequency as the kilometer marker string of the current pole trajectory. If there are no frames included in the kilometer marker set within the current pole trajectory, mark the kilometer marker string of the current pole trajectory as empty.
[0109] Step 3.3: Determine the pole type label of the current pole trajectory based on the pole type labels of each frame within the pole trajectory, traverse the pole type labels identified in each frame within the current pole trajectory, and construct a type label set.
[0110] If the type label set contains a preset special type label, then the special type label is used as the pole type label of the current pole trajectory;
[0111] If the set of type labels does not contain preset special type labels but contains regular type labels, then the regular type labels will be used as the pole type labels for the current pole trajectory;
[0112] If the set of type labels is empty, then the pole type label of the current pole trajectory will be marked as unknown.
[0113] Step 3.4: Determine the license plate area characters based on the license plate area characters of each frame within the pole trajectory. Extract all frames that are included in the candidate pole node set and their corresponding license plate area characters from the frame data sequence belonging to the current pole trajectory, construct a pole number candidate set, and select the character with the highest confidence as the license plate area character of the current pole trajectory; if there are no frames included in the candidate pole node set within the current pole trajectory, mark the license plate area characters of the current pole trajectory as empty.
[0114] Step 3.5: Based on the representative frame, kilometer marker string, pole type label, and license plate area characters, construct candidate pole nodes and arrange them in chronological order to obtain a candidate pole node sequence.
[0115] Step 4: Obtain the pre-stored trackside real structured data. Based on the uniqueness of the kilometer markers, construct a key point dictionary using pole type labels, interval distances, row types, and pole numbers as indexes, and generate a real pole number sequence arranged in the order of the line.
[0116] Step 4.1: Obtain pre-stored trackside real structured data, which includes at least the section, row, pole number, pole type label, and physical pole information of the kilometer marker.
[0117] Step 4.2: Based on the uniqueness of the kilometer marker, construct a key point dictionary using the interval segment, row type, and pole number as combined indexes. Each combined index uniquely identifies a physical pole, forming a set of real pole number primary keys.
[0118] Step 4.3: Based on the real pole number primary key set, extract the kilometer marker and pole type label corresponding to each physical pole, and generate a real pole number sequence according to the route order. Each real pole number contains a pole number, the corresponding kilometer marker, and a pole type label.
[0119] Step 5: For each real pole number and candidate pole node, score all video frames covered by the candidate pole node frame by frame based on the matching degree of kilometer marker characters, pole number plate characters, and pole type label, and calculate the matching score between the real pole number and the candidate pole node.
[0120] In this embodiment, the matching score between the actual pole number and the candidate pole node is represented as:
[0121] ;
[0122] In the formula, Indicates the first The first real pole number and the first Matching scores between candidate pole nodes Indicates the first The first real pole number and the first Kilometer marker character matching degree among candidate pole nodes Indicates the first The first real pole number and the first The degree of matching of pole type labels among candidate pole nodes. Indicates the first The first real pole number and the first The degree of character matching of pole number plates among the candidate pole nodes, where... Indicates the first The first real pole number and the first Step index during the matching process of candidate pole nodes;
[0123] In this embodiment, the kilometer marker character matching degree is represented as:
[0124] ;
[0125] In the formula, Indicates the first The first real pole number and the first Matching score of kilometer markers between candidate pole nodes Preset weights, , This represents the matching function. Indicates the first A real pole number, kilometer marker character, Indicates the first The kilometer marker characters of each candidate pole node, This indicates the score when both the kilometer marker and the actual pole number match simultaneously. The preset weight for the score when both the kilometer marker and the actual pole number match simultaneously;
[0126] In this embodiment, the rod type label matching degree is represented as:
[0127] ;
[0128] In the formula, Indicates the first The first real pole number and the first Link type label matching score among candidate link nodes Preset weights, , No. A real pole number and pole type label. Indicates the first The pole type label of each candidate pole node. This indicates the score when both the pole type label and the actual pole number match.
[0129] In this embodiment, the character matching degree of the pole number plate is represented as:
[0130] ;
[0131] In the formula, Indicates the first The first real pole number and the first Pocket number matching score between candidate pole nodes Preset weights, , Indicates the first The pole number characters of each candidate pole node Indicates the first A real pole number plate character.
[0132] Step 6: Based on the matching score, use the Viterbi algorithm to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence; wherein, during the solution process, skipping candidate pole nodes and / or missing matching of real pole numbers is allowed, and a jump smoothing penalty term based on the node center frame spacing is introduced to constrain the path jump.
[0133] Step 6.1: Based on the length of the actual pole number sequence and the length of the candidate pole node sequence, construct the cumulative score matrix and the corresponding state transition precursor matrix, initialize the cumulative score at the starting position to a preset value, and initialize the cumulative scores at the remaining positions in the state transition precursor matrix to negative infinity.
[0134] In this embodiment, the recursive relationship of the state transition predecessor matrix is expressed as follows:
[0135] ;
[0136] In the formula, Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This indicates the operation of retrieving the maximum value. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This represents a preset constant penalty term for skipping candidate pole nodes. This represents a preset constant penalty term for missing matching of the actual pole number. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. Indicates the first The candidate pole node is matched with the first... When the actual pole number is 1, an adaptive jump smoothing penalty term is applied relative to the previous matched node to constrain the jump distance of adjacent matched nodes on the time axis.
[0137] Step 6.2: Obtain the frame span covered by the current candidate pole node in the video frame sequence. Based on the preset basic tolerance, span tolerance coefficient, and the distance between the center frame of the previously matched candidate pole node and the center frame of the current candidate pole node, determine the adaptive jump smoothing penalty term.
[0138] In this embodiment, when the spacing exceeds the sum of the basic tolerance and the span tolerance coefficient multiplied by the frame span, the adaptive jump smoothing penalty term is calculated based on the excess portion of the spacing and a preset penalty intensity coefficient; if there is no candidate pole node that was successfully matched in the previous match or the current is the first match of the path, the adaptive jump smoothing penalty term is set to zero.
[0139] In this embodiment, the adaptive jump smoothing penalty term is expressed as:
[0140] ;
[0141] In the formula, Indicates the first The candidate pole node is matched with the first... When the actual pole number is [number], an adaptive jump smoothing penalty term is applied relative to the previous matched node. This represents the preset penalty intensity coefficient. This represents the distance between the center frame of the current candidate pole node and the center frame of the previously successfully matched candidate pole node. , This indicates the center frame number of the current candidate pole node within the frame sequence it covers. This indicates the center frame number of the previously matched candidate pole node. This indicates the permissible inter-frame spacing tolerance. , Indicates the basic tolerance level. This is the span tolerance coefficient. Indicates the first The frame span covered by each candidate pole node in the video frame sequence. Indicates other situations.
[0142] Step 6.3: Set the maximum allowed number of consecutive candidate pole nodes to be crossed. Traverse each position in the cumulative score matrix. Under the condition that the difference between the current candidate pole node index and the previous matching node index does not exceed the maximum allowed number of consecutive candidate pole nodes to be crossed, calculate the cumulative score corresponding to the three states of matching the real pole number with the candidate pole node, skipping the candidate pole node, and missing the matching of the real pole number. Record the state of action that maximizes the cumulative score and the corresponding predecessor node index. Among them, the cumulative score of matching the real pole number with the candidate pole node is the previous cumulative score plus the matching score minus the adaptive jump smoothing penalty term. The cumulative score of skipping the candidate pole node is the previous cumulative score minus the preset skip penalty term. The cumulative score of missing the matching of the real pole number is the previous cumulative score minus the preset missing matching penalty term.
[0143] Step 6.4: Obtain the global optimal cumulative score at the end of the cumulative score matrix. Based on the path actions recorded in the state transition predecessor matrix, backtrack from the end to the starting position to generate a globally optimal alignment path that includes the mapping relationship between the real pole number and the candidate pole node.
[0144] Step 7: Based on the global optimal alignment path, mark the corresponding real pole number and write it back to the candidate pole node sequence to generate the catenary support equipment file.
[0145] Example 3
[0146] Based on the same inventive concept as other embodiments, this embodiment describes a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the steps of the methods of Embodiment 1 or 2 described above.
[0147] Example 4
[0148] Based on the same inventive concept as other embodiments, this embodiment introduces a computer program product, including computer instructions that, when executed by a processor, implement the steps of the methods described in Embodiment 1 or 2 above.
[0149] In summary, this invention achieves a globally optimal alignment path solution from the candidate pole node sequence to the actual pole number sequence by segmenting the pole trajectory based on multi-frame visual sequence alignment and generating candidate pole node sequences, combined with a key point dictionary index constructed based on the uniqueness of kilometer markers, and by introducing a jump smoothing penalty term in the matching score calculation using the Viterbi algorithm. This effectively solves the problems of low accuracy in the construction of catenary support pole files, inability to effectively handle missed detections, false detections, and disordered sequences in existing technologies, and significantly improves the accuracy and automation of the construction of catenary support pole equipment files.
[0150] This invention extracts the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence, and intelligently classifies them into the kilometer marker set or candidate pole node set based on whether the first character is "K" or "k". This effectively filters out possible misidentification and noise interference in single-frame recognition, ensuring that subsequent processing is based on the most reliable recognition results, and significantly improves the accuracy and robustness of character recognition in complex trackside environments such as changes in lighting, dirt, and vibration.
[0151] This invention segments the frame data sequence based on the area change trajectory of the pole bounding box under the pre-set constraint of continuous missing frames, obtaining several pole trajectories. By utilizing the area change pattern of the pole as it moves from far to near and then away in the video sequence, and allowing a small number of missing frames within the trajectory, it can accurately cluster video frames belonging to the same pole in a continuous video stream into a complete pole trajectory. This avoids trajectory breakage or erroneous merging caused by brief occlusion or detection failure, providing high-quality spatiotemporal segments for subsequent pole recognition.
[0152] This invention uses the Viterbi algorithm based on the matching score to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence, and introduces an adaptive jump smoothing penalty term to constrain the jump distance between adjacent matching nodes on the time axis. It can comprehensively consider character matching degree, pole type matching degree, and the continuity of the time series in a global scope, and solve the optimal matching sequence between candidate nodes and real pole numbers for the entire line in one go. It effectively solves the problem of local matching errors caused by false detection, missed detection, or character similarity, and realizes high-precision automatic alignment between visual detection results and existing ledger data.
[0153] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0154] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0155] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0156] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0157] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.
Claims
1. A method for automatically constructing overhead contact line support equipment archives based on multi-frame visual sequence alignment, characterized in that, include: Real-time acquisition of trackside inspection videos, and extraction of the license plate area characters / kilometer marker area characters with the highest confidence in each frame to obtain the frame data sequence; Based on the area change trajectory of the rod boundary box, under the preset continuous empty frame constraint, the frame data sequence is segmented to obtain several rod trajectories. The pole trajectory is compressed into candidate pole nodes, and representative frames, kilometer marker area characters, license plate area characters, and pole type labels are extracted to obtain a sequence of candidate pole nodes arranged by time. Obtain pre-stored real structured data from the trackside, and based on the uniqueness of kilometer markers, construct a key point dictionary using pole type labels, interval distances, row types, and pole numbers as indexes to generate a real pole number sequence arranged in line order; For each real pole number and candidate pole node, based on the kilometer marker character matching degree, pole number plate character matching degree, and pole type label matching degree, all video frames covered by the candidate pole node are scored frame by frame, and the matching score between the real pole number and the candidate pole node is calculated. Based on the matching score, the Viterbi algorithm is used to find the globally optimal alignment path from the candidate pole node sequence to the real pole number sequence; wherein, during the solution process, skipping candidate pole nodes and / or missing matching of real pole numbers is allowed, and a jump smoothing penalty term based on the node center frame spacing is introduced to constrain the path jump; Based on the globally optimal alignment path, the corresponding real pole number is marked and written back to the candidate pole node sequence to generate the contact network support equipment file.
2. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 1, characterized in that, Extract the license plate area characters / kilometer marker area characters with the highest confidence level from each frame to obtain the frame data sequence, including: Discretize the trackside inspection video into an ordered frame sequence; Based on the ordered frame sequence, target detection and text recognition are performed frame by frame to extract the original frame data sequence, which includes at least the frame number, timestamp, license plate area bounding box, kilometer marker area bounding box, pole bounding box, license plate area characters and kilometer marker area characters. Based on the original frame data sequence, the frame data sequence is obtained by extracting the license plate area character / kilometer marker area character with the highest confidence in each frame. If the first character of the license plate area character / kilometer marker area character is "K" or "k", it is included in the kilometer marker set, and the character with the highest confidence is selected as the kilometer marker of the current frame. If the first character of the license plate area characters / kilometer marker area characters is not "K" or "k", then the license plate area characters / kilometer marker area characters are normalized and included in the candidate pole node set.
3. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 2, characterized in that, Based on the area change trajectory of the rod bounding box, under the preset constraint of continuous empty frames, the frame data sequence is segmented to obtain several rod trajectories, including: Based on the frame bounding box of each frame in the frame data sequence, calculate the area of the frame bounding box in chronological order; Valid frames are selected based on the area change trajectory of the rod bounding box: If the current frame is missing a pole bounding box, or if the area of the pole bounding box in the current frame is less than a preset effective area threshold, then the current frame is marked as a poleless frame. The frame data sequence is segmented according to the area change trajectory of the rod boundary box and the preset continuous empty frame constraint condition, and the sequence is traversed from the starting frame of the frame data sequence. When the length of consecutive frames marked as poleless exceeds the preset maximum empty frame threshold, or when the ratio of the area of the pole bounding box of the current frame to the historical maximum area of the current pole trajectory is lower than the preset area decrease ratio threshold, the current pole trajectory is determined to end. In response to the determination that the current pole trajectory has ended, the current pole trajectory ends and the recording of the next pole trajectory begins; Traverse all frames to generate several pole trajectory segments containing consecutive valid frames, thus obtaining several pole trajectories.
4. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 3, characterized in that, The preset continuous empty frame constraint condition is: allowing continuous poleless frames not exceeding a first threshold and non-continuous poleless frames not exceeding a second threshold within a single pole trajectory.
5. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 4, characterized in that, The pole trajectory is compressed into candidate pole nodes, and representative frames, kilometer marker area characters, license plate area characters, and pole type labels are extracted to obtain a time-ordered sequence of candidate pole nodes, including: The representative frame is determined based on the area of the bounding box of the pole in the pole trajectory. The frame with the largest bounding box area in the pole trajectory is selected as the representative frame of the current pole trajectory. The kilometer marker string is determined based on the kilometer marker region characters of each frame within the pole trajectory. From the frame data sequence belonging to the pole trajectory, all frames that are included in the kilometer marker set and their corresponding kilometer marker region characters are extracted. The frequency of occurrence of each kilometer marker region character is counted, and the character with the highest frequency is selected as the kilometer marker string of the current pole trajectory. If there are no frames included in the kilometer marker set within the current pole trajectory, the kilometer marker string of the current pole trajectory is marked as empty. The pole type label of the current pole trajectory is determined based on the pole type label of each frame within the pole trajectory. The pole type labels identified in each frame within the current pole trajectory are traversed to construct a set of type labels. If the type label set contains a preset special type label, then the special type label is used as the pole type label of the current pole trajectory; If the set of type labels does not contain preset special type labels but contains regular type labels, then the regular type labels will be used as the pole type labels for the current pole trajectory; If the set of type labels is empty, then the pole type label of the current pole trajectory will be marked as unknown; The license plate area characters are determined based on the license plate area characters of each frame within the pole trajectory. From the frame data sequence belonging to the current pole trajectory, all frames that are included in the candidate pole node set and their corresponding license plate area characters are extracted to construct a pole number candidate set. The character with the highest confidence is selected as the license plate area character of the current pole trajectory. If there are no frames in the current pole trajectory that are included in the candidate pole node set, the license plate area characters of the current pole trajectory are marked as empty. Based on the representative frame, kilometer marker string, pole type label, and license plate area characters, candidate pole nodes are constructed and arranged in chronological order to obtain a candidate pole node sequence.
6. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 5, characterized in that, Retrieve pre-stored trackside real structured data. Based on the uniqueness of kilometer markers, construct a keypoint dictionary indexed by pole type label, interval distance, row, and pole number. Generate a sequence of real pole numbers arranged in line order, including: Obtain pre-stored trackside real structured data, which includes at least the section segment, row, pole number, pole type label, and physical pole information of the kilometer marker; Based on the uniqueness of kilometer markers, a key point dictionary is constructed using interval segments, row types, and pole numbers as combined indexes. Each combined index uniquely identifies a physical pole, forming a set of real pole number primary keys. Based on the primary key set of real pole numbers, extract the kilometer marker and pole type label corresponding to each physical pole, and generate a real pole number sequence in the order of the route. Each real pole number contains a pole number, the corresponding kilometer marker, and a pole type label.
7. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 6, characterized in that, The matching score between the actual pole number and the candidate pole node is represented as follows: ; In the formula, Indicates the first The first real pole number and the first Matching scores between candidate pole nodes Indicates the first The first real pole number and the first Kilometer marker character matching degree among candidate pole nodes Indicates the first The first real pole number and the first The degree of matching of pole type labels among candidate pole nodes. Indicates the first The first real pole number and the first The degree of character matching of pole number plates among the candidate pole nodes, where... Indicates the first The first real pole number and the first Step index during the matching process of candidate pole nodes; The kilometer marker character matching degree is represented as follows: ; In the formula, Indicates the first The first real pole number and the first Matching score of kilometer markers between candidate pole nodes Preset weights, , This represents the matching function. Indicates the first A real pole number, kilometer marker character, Indicates the first The kilometer marker characters of each candidate pole node, This indicates the score when both the kilometer marker and the actual pole number match simultaneously. The preset weight for the score when both the kilometer marker and the actual pole number match simultaneously; The matching degree of the rod type label is expressed as follows: ; In the formula, Indicates the first The first real pole number and the first Link type label matching score among candidate link nodes Preset weights, , No. A real pole number and pole type label. Indicates the first The pole type label of each candidate pole node. The score indicates that both the pole type label and the actual pole number match. The character matching degree of the pole number plate is represented as follows: ; In the formula, Indicates the first The first real pole number and the first Pocket number matching score between candidate pole nodes Preset weights, , Indicates the first The pole number characters of each candidate pole node Indicates the first A real pole number plate character.
8. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 7, characterized in that, Based on the matching score, the Viterbi algorithm is used to solve for the globally optimal alignment path from the candidate pole node sequence to the actual pole number sequence, including: Based on the length of the actual pole number sequence and the length of the candidate pole node sequence, a cumulative score matrix and a corresponding state transition precursor matrix are constructed. The cumulative score at the starting position is initialized to a preset value, and the cumulative score at the remaining positions in the state transition precursor matrix is initialized to negative infinity. The frame span covered by the current candidate pole node in the video frame sequence is obtained. Based on the preset base tolerance, span tolerance coefficient, and the distance between the center frame of the previously matched candidate pole node and the center frame of the current candidate pole node, an adaptive jump smoothing penalty term is determined. Wherein, when the distance exceeds the sum of the base tolerance and the span tolerance coefficient multiplied by the frame span, the adaptive jump smoothing penalty term is calculated based on the excess part of the distance and the preset penalty intensity coefficient. If there is no previously matched candidate pole node or the current is the first match of the path, the adaptive jump smoothing penalty term is set to zero. Set a maximum allowed number of consecutive candidate pole nodes to be crossed. Iterate through each position in the cumulative score matrix. Under the condition that the difference between the current candidate pole node index and the previous matching node index does not exceed the maximum allowed number of consecutive candidate pole nodes to be crossed, calculate the cumulative score for three states: matching the real pole number with a candidate pole node, skipping a candidate pole node, and missing a match on the real pole number. Record the state action that maximizes the cumulative score and the corresponding predecessor node index. The cumulative score for matching the real pole number with a candidate pole node is the previous cumulative score plus the matching score minus the adaptive jump smoothing penalty. The cumulative score for skipping a candidate pole node is the previous cumulative score minus a preset skip penalty. The cumulative score for missing a match on the real pole number is the previous cumulative score minus a preset missing match penalty. The globally optimal cumulative score is obtained at the end of the cumulative score matrix. Based on the path actions recorded in the state transition predecessor matrix, the process is reversed from the end to the starting position to generate a globally optimal alignment path that includes the mapping relationship between the real pole number and the candidate pole node.
9. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 8, characterized in that, The recursive relationship of the state transition predecessor matrix is expressed as follows: ; In the formula, Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This indicates the operation of retrieving the maximum value. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. This represents a preset constant penalty term for skipping candidate pole nodes. This represents a preset constant penalty term for missing matching of the actual pole number. Indicates the first The first real pole number and the first The cumulative score of each candidate pole node. Indicates the first The candidate pole node is matched with the first... When the actual pole number is 1, an adaptive jump smoothing penalty term is applied relative to the previous matched node to constrain the jump distance of adjacent matched nodes on the time axis.
10. The method for automatically constructing contact wire support equipment archives based on multi-frame visual sequence alignment according to claim 9, characterized in that, The adaptive jump smoothing penalty term is expressed as follows: ; In the formula, Indicates the first The candidate pole node is matched with the first... When the actual pole number is [number], an adaptive jump smoothing penalty term is applied relative to the previous matched node. This represents the preset penalty intensity coefficient. This represents the distance between the center frame of the current candidate pole node and the center frame of the previously successfully matched candidate pole node. , This indicates the center frame number of the current candidate pole node within the frame sequence it covers. This indicates the center frame number of the previously matched candidate pole node. This indicates the permissible inter-frame spacing tolerance. , Indicates the basic tolerance level. This is the span tolerance coefficient. Indicates the first The frame span covered by each candidate pole node in the video frame sequence. Indicates other situations.
