Pipeline defect video inspection method based on a tracked pipeline robot

Abstract

The present application belongs to the technical field of image recognition, and relates to a pipeline defect video inspection method based on a tracked pipeline robot. The present application collects inspection videos and obtains inspection image frames and pipeline wall features, identifies a low-confidence region based on image quality states and generates a pollution candidate region; in combination with the image coordinate fixation of the pollution candidate region in consecutive frames, the corresponding relationship with the pipeline wall features and the region diffusion state, a water droplet interference region is judged. If it belongs to the water droplet interference region, the water droplet interference region in each inspection image frame in the pollution video segment is determined as the corresponding pollution frozen region, the pipeline defect confirmation output in each pollution frozen region is suspended, and the re-collection starting point is determined after cleaning confirmation. The present application can distinguish between lens water droplet diffusion interference and real pipeline wall defects, reduce false positives and missed detections caused by water droplets, water film bright spots or blurred regions, and improve the accuracy and reviewability of the pipeline defect video inspection results.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a video inspection method for pipeline defects based on a tracked pipeline robot. Background Technology

[0002] In the inspection of municipal drainage pipes, combined sewer systems, and underground pipe networks, tracked pipeline robots typically travel along the bottom of the pipe and collect video footage of the pipe's interior using cameras, lighting, and video transmission units. Existing video inspection methods for pipeline defects mostly identify defects such as cracks, leaks, corrosion, deposits, and misaligned interfaces in the pipe wall through image framing, brightness enhancement, edge extraction, texture analysis, and target detection.

[0003] However, in actual inspections, pipes often contain accumulated water, silt, water mist, and oil. When the robot passes through waterlogged or silt-disturbed areas, water droplets can easily splash onto the camera lens surface. Under the influence of robot vibration, airflow within the pipe, or continued splashing, these droplets can spread into a water film, causing fixed blurred areas, fixed bright spots, locally blurred areas, or overall fogging in the video image. These anomalies originate from contamination on the lens surface, not from actual defects in the pipe wall. Existing video recognition methods typically assume that abnormal edges, brightness, and textures in the image originate from the pipe wall itself, lacking a mechanism to distinguish between lens contamination and actual defects. Therefore, after water droplets spread through the lens, water droplet edges may be misidentified as cracks, bright spots in the water film may be misidentified as leaks, and blurred dark areas may be misidentified as corrosion. Simultaneously, actual cracks, leaks, or corrosion may be missed due to obscuring the water film.

[0004] This issue directly impacts the accuracy and reliability of video inspection results for pipeline defects using tracked pipeline robots. On one hand, misjudging lens droplets, bright spots from water film, or fogged dark areas as cracks, leaks, or corrosion will lead to an inflated number of defects in the inspection report, increasing the costs of manual verification, dredging, repair, and excavation. On the other hand, if genuine cracks, leaks, or corrosion are covered by a water film and fail to be identified, high-risk pipe sections may be incorrectly classified as normal, thus delaying pipeline maintenance and risk management. For aging drainage networks, densely underground areas, and pipelines beneath roads, these missed inspections could further lead to serious consequences such as pipeline leaks, road collapses, sewage overflows, and inaccurate operation and maintenance decisions. Therefore, there is an urgent need for a method that can identify lens droplet diffusion interference and distinguish between contamination at the acquisition end and genuine defects in the pipe wall during video inspection of pipeline defects, in order to improve the accuracy and verifiability of video inspection results from tracked pipeline robots. Summary of the Invention

[0005] The purpose of this invention is to provide a video inspection method for pipeline defects based on a tracked pipeline robot, so as to solve the problem of false detection and missed detection of pipeline defects caused by the diffusion of water droplets in the lens when the tracked pipeline robot inspects in environments with water accumulation, silt, and water mist.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a video inspection method for pipeline defects based on a tracked pipeline robot, comprising the following steps:

[0007] The system collects inspection videos generated by a tracked pipeline robot as it moves through the pipeline, and obtains inspection image frames and pipe wall features based on the inspection videos.

[0008] Based on the image quality status of the inspected image frames, low-confidence regions with confidence levels below a preset stable threshold are identified, and contamination candidate regions are generated based on these low-confidence regions.

[0009] Based on the changes in the image coordinates of the contamination candidate area in continuous inspection image frames, fixed values ​​are generated; based on the correspondence between the contamination candidate area and pipe wall features, corresponding values ​​are generated; based on the regional change state of the contamination candidate area, a diffusion trajectory is generated.

[0010] Based on fixed values, corresponding values, and diffusion trajectories, determine whether a pollution candidate area belongs to a water droplet interference area;

[0011] If the contamination candidate area belongs to the water droplet interference area, then the contamination video segment is determined. For each inspection image frame in the contamination video segment where the water droplet interference area is identified, the water droplet interference area in the corresponding inspection image frame is determined as the contamination freezing area corresponding to that inspection image frame, the freezing result is generated, and the pipeline defect confirmation output in each contamination freezing area is paused.

[0012] The starting point for re-mining is determined based on the freezing results, and the inspection video of the re-mining section is re-acquired based on the starting point to obtain the re-mining image;

[0013] Based on the re-sampling images, defect candidate areas are re-identified, and based on the correspondence between the defect candidate areas and the re-extracted pipe wall features, real defect results or pollution pseudo-anomaly results are generated.

[0014] In summary, due to the adoption of the above-mentioned pipeline defect video inspection method based on a tracked pipeline robot, the beneficial effects of this invention are:

[0015] This invention, through step-by-step screening of low-confidence regions and contamination candidate regions, can locate abnormal areas in pipeline inspection videos that are affected by lens water droplets, bright spots of water film, or local fogging, thus avoiding directly outputting low-confidence image regions as pipe wall defects.

[0016] This invention uses fixed values, corresponding values, and diffusion trajectories to jointly determine whether a contamination candidate area belongs to a water droplet interference area. It can distinguish between contamination at the acquisition end fixed in the image coordinates and real defects corresponding to fixed pipe wall features, reducing the situation where water droplet edges are misjudged as cracks, bright spots in water films are misjudged as leaks, and blurry dark areas are misjudged as corrosion.

[0017] This invention generates freezing results after confirming the water droplet interference area, and pauses the pipeline defect confirmation output in each contaminated frozen area. Then, it re-collects the inspection video of the re-collected pipeline section based on the re-collection starting point. This can reduce false detections and missed detections caused by contaminated video segments and improve the accuracy and verifiability of pipeline defect inspection results. Attached Figure Description

[0018] Figure 1 A schematic diagram of the overall process of the present invention is shown;

[0019] Figure 2 A schematic diagram of the water droplet interference zone determination process of the present invention is shown;

[0020] Figure 3 A schematic diagram of the low-confidence region identification process of the present invention is shown. Detailed Implementation

[0021] The pipeline defect video inspection method based on a tracked pipeline robot according to the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention.

[0022] The pipeline defect video inspection method based on a tracked pipeline robot provided by the present invention will be described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are used to illustrate the technical solution of the present invention and are not intended to limit the scope of protection of the present invention;

[0023] Example 1:

[0024] In this embodiment, the preset stability threshold, preset image location range, and preset region overlap condition can all be predetermined based on historical inspection videos, manually labeled samples, or on-site calibration results. The preset stability threshold is used to determine whether an image region is suitable for defect confirmation; the preset image location range is used to determine whether the change in the center position of the contamination candidate area is within an allowable range; the preset region overlap condition is used to determine whether contamination candidate areas in adjacent inspection image frames remain within the same image region range; wherein, the preset stability threshold is used for comparison with the recognition confidence level, and both the recognition confidence level and the preset stability threshold are values ​​between 0 and 1; the preset stability threshold can be determined based on the correct recognition performance of the pipeline defect recognition model in historical inspection images. In a specific embodiment, the preset stability threshold can be set to a value between 0.6 and 0.8. When the recognition confidence level is lower than this preset stability threshold, it indicates that the pipe wall feature boundary, bright spot occlusion state, or texture continuity recognition ratio in the corresponding image region is insufficient to support stable defect recognition.

[0025] Furthermore, the preset image position range and preset region overlap condition are used to determine whether the contamination candidate area remains stably within the same image coordinate range. Specifically, the preset image position range can be determined based on the image jitter amplitude, image region division size, and contamination candidate area boundary recognition error of the tracked pipeline robot in a stable traveling state;

[0026] In one specific embodiment, historical inspection videos or on-site calibration videos without lens contamination are selected. The center position offset of the same image region caused by track vibration, lighting fluctuations, and image segmentation errors is statistically analyzed. The upper limit of this center position offset, or the upper limit plus a preset margin, is used as the preset image position range. For example, the preset image position range can be set to 0.5% to 3% of the inspection image frame width, or it can be set to 10 to 30 pixels. Subsequently, based on the regional boundaries of contamination candidate areas in adjacent inspection image frames, the overlapping area between two contamination candidate areas is calculated, and this overlapping area is compared with the smaller coverage area of ​​the two contamination candidate areas to obtain the regional overlap ratio. When the regional overlap ratio is not lower than the preset overlap ratio, it is determined that the contamination candidate areas in adjacent inspection image frames meet the preset regional overlap condition. In one specific embodiment, the preset overlap ratio can be set to 0.6 to 0.8.

[0027] Since the water droplet interference area may spread outward in continuous inspection image frames, the region overlap ratio is determined by comparing the overlapping area with the smaller coverage area, rather than judging solely by the intersection-union ratio of two contaminated candidate areas. This avoids incorrectly judging the image coordinates as unstable when the coverage of the contaminated candidate area increases.

[0028] See Figure 1A video inspection method for pipeline defects based on a tracked pipeline robot includes the following steps:

[0029] S1, collect inspection video, and obtain inspection image frames and pipe wall features.

[0030] Since pipeline defect identification requires determining whether the abnormal area originates from the pipe wall itself, and lens water droplets, water film bright spots, and localized fogging areas are usually fixed in the image coordinates and do not move synchronously with the pipe wall texture, pipe interface loops, or deposition boundaries, it is necessary to extract pipe wall features that can characterize the changes in pipe wall position from the inspection video before confirming the defect. These pipe wall features will then serve as the reference basis for distinguishing between lens contamination and real pipe wall defects.

[0031] In one specific embodiment, firstly, the inspection video is segmented into frames according to a preset video analysis cycle to obtain inspection image frames. Then, pipe region identification is performed on the inspection image frames, excluding non-pipe regions at the edges of the image to obtain the pipe image region. Further, within the pipe image region, candidate pipe wall features present in the current inspection image frame are identified from pipe interface loops, pipe wall textures, waterline boundaries, and sedimentation boundaries. Specifically, when a candidate pipe wall feature can repeatedly appear in consecutive inspection image frames, and its positional changes match the travel direction of the tracked pipe robot, the candidate pipe wall feature is determined as a pipe wall feature. Finally, the image coordinates of the pipe wall feature in consecutive inspection image frames are recorded for subsequent determination of whether a contaminated candidate area consistently corresponds to the same pipe wall feature.

[0032] S2 identifies low-confidence regions based on the image quality status of the inspected image frames and generates contamination candidate regions from the low-confidence regions.

[0033] Considering that the diffusion of water droplets on the lens does not necessarily render the entire inspection image unrecognizable, but usually manifests as local bright spots, local blurring, local boundary blurring, or local texture interruption, it is not appropriate to directly determine the entire inspection image as unusable. Instead, the image frame should be divided into image regions, and recognition confidence should be generated for each image region separately.

[0034] See Figure 3 The method for identifying low-confidence regions includes steps S201 to S203.

[0035] S201, divide each inspection image frame into at least two image regions.

[0036] S202 generates the recognition credibility of each image region based on the clarity of the pipe wall feature boundary, the coverage of the bright area, and the proportion of continuous texture recognition.

[0037] Furthermore, the texture continuity recognition ratio is used to characterize the degree of preservation of the continuously identifiable pipe wall texture within the image region of the current inspection image frame, relative to the pipe wall texture that should be identifiable under the corresponding clear state. To clarify the source of the length of the pipe wall texture to be identified, firstly, based on the correspondence of pipe wall features in different inspection image frames, in the inspection image frames before the current inspection image frame, image regions corresponding to the image region to be judged and whose recognition confidence is not lower than a preset stable threshold are found, and the image region closest to the current inspection image frame is determined as the clear reference image region.

[0038] When no clear reference image region meets the conditions in the current inspection video, an image region corresponding to the image region to be judged is selected from historical clear inspection image frames or on-site calibration images as the clear reference image region. Subsequently, continuously identifiable pipe wall texture segments are extracted from the clear reference image region, and the pixel lengths of each pipe wall texture segment are accumulated. The total length of the accumulated pipe wall texture is determined as the length of the pipe wall texture to be identified. In the image region to be judged in the current inspection image frame, pipe wall texture segments corresponding to the pipe wall texture segments in the clear reference image region are searched. The pixel lengths of the pipe wall texture segments that still correspond continuously are accumulated to obtain the continuously identifiable pipe wall texture length.

[0039] Texture continuity recognition ratio = Length of continuously recognizable pipe wall texture ÷ Length of pipe wall texture to be recognized;

[0040] When there is no pre-determined continuous pipe wall texture within the clearly referenced image area, or when the length of the pipe wall texture to be identified is 0, the corresponding image area is determined as a textureless reference area, and the continuous texture recognition ratio is not calculated for this image area. For textureless reference areas, the third weight is set to 0, and the first and second weights are normalized according to their original ratios, so that the sum of the normalized first weight and the normalized second weight is 1. Then, the recognition confidence is generated based on the sharpness result and the bright spot result.

[0041] When the image region does not belong to a textureless reference region, the recognition confidence score is generated based on the sharpness result, bright spot result, and blur result. The recognition confidence score is generated in the following manner:

[0042] In one specific embodiment, firstly, each inspection image frame is divided according to a preset number of rows and a preset number of columns to obtain at least two image regions; then, image quality judgment is performed on each image region to avoid local water droplets, water film bright spots or fogging areas causing the entire inspection image frame to be judged as unusable.

[0043] Specifically, the sharpness result is used to characterize whether the pipe wall feature boundaries within the image area are clear. The edge response values ​​of the pipe wall feature boundaries within the image area are obtained and compared with a sharpness reference edge response value to obtain the sharpness result. The sharpness result ranges from 0 to 1; the closer the edge response value is to or higher than the sharpness reference edge response value, the closer the sharpness result is to 1; the lower the edge response value, the closer the sharpness result is to 0. The sharpness reference edge response value can be determined from uncontaminated adjacent inspection image frames in the same inspection video, historical clear inspection image frames, or on-site calibration images.

[0044] Subsequently, the bright spot results are used to characterize the degree to which the bright areas obscure the pipe wall features. Bright pixels with brightness values ​​higher than a preset brightness upper limit are acquired within the image region. The proportion of the area of ​​the pipe wall features or texture covered by these bright pixels to the total image region area is calculated to obtain the bright spot occlusion ratio. Based on this ratio, bright spot results are generated in reverse. The value of the bright spot result ranges from 0 to 1; the larger the bright spot occlusion ratio, the closer the bright spot result is to 0, and the smaller the ratio, the closer the bright spot result is to 1.

[0045] The texture continuity recognition ratio ranges from 0 to 1. A ratio closer to 1 indicates more complete preservation of the pipe wall texture in the current image area; a ratio closer to 0 indicates a greater degree of interruption in the pipe wall texture due to water droplets, water film, or fogging. A blurred result is generated based on the texture continuity recognition ratio, with a value ranging from 0 to 1. A higher texture continuity recognition ratio results in a blurred result closer to 1, while a lower ratio results in a blurred result closer to 0.

[0046] Recognition credibility = First weight × Sharpness result + Second weight × Bright spot result + Third weight × Blur result;

[0047] Recognition confidence = Normalized first weight × Sharpness result + Normalized second weight × Bright spot result.

[0048] S203, when the recognition confidence is lower than the preset stable threshold, the corresponding image region is determined as a low confidence region.

[0049] The sum of the first, second, and third weights is 1, and all are greater than 0. The first, second, and third weights can be determined based on historical inspection videos, manually labeled samples, or on-site calibration results. For textureless reference regions, the recognition confidence level is generated as follows:

[0050] When the recognition confidence level is not lower than the preset stability threshold, the corresponding image region is retained for subsequent pipeline defect recognition; when the recognition confidence level is lower than the preset stability threshold, the corresponding image region is determined as a low confidence region.

[0051] It should be noted that low-confidence areas are not necessarily lens contamination areas; they may also be caused by actual cracks, leaks, corrosion, or deposits. Therefore, this embodiment only identifies low-confidence areas that meet the continuity condition as contamination candidate areas, and does not directly output a lens contamination conclusion.

[0052] In one specific embodiment, firstly, the region boundary and coverage area of ​​the low-confidence area are obtained. Then, adjacent low-confidence areas with continuous boundaries are merged to obtain candidate coverage areas. Further, based on the sharpness results, bright spot results, and texture continuity recognition ratio, it is determined whether the candidate coverage area contains low-confidence sources such as unclear pipe wall feature boundaries, bright area occlusion, or discontinuous pipe wall feature textures. Finally, when a candidate coverage area contains a low-confidence source and appears repeatedly in consecutive inspection image frames, the candidate coverage area is identified as a contamination candidate area. This contamination candidate area is only used as input for subsequent fixed value, corresponding value, and diffusion trajectory determination, and is not directly used as the final contamination result output.

[0053] S3 generates fixed values, corresponding values, and diffusion trajectories based on the changes in the image coordinates of the pollution candidate area, their correspondence with pipe wall features, and the state of regional changes.

[0054] To distinguish between lens water droplet diffusion interference and actual pipe wall defects, this embodiment judges the contamination candidate area from three directions: image coordinates, pipe wall reference, and regional diffusion. Fixed values ​​characterize whether the contamination candidate area is fixed in the image coordinates; corresponding values ​​characterize whether the contamination candidate area consistently corresponds to the same pipe wall feature; and diffusion trajectories characterize whether the contamination candidate area undergoes a process of expansion from a local area to a continuously covered area.

[0055] In one specific embodiment, firstly, the center position and region boundary of the contaminated candidate region in each inspection image frame are obtained; a center position change result is generated based on the center position of the contaminated candidate region in adjacent inspection image frames; and a region overlap result is generated based on the region boundary of the contaminated candidate region in adjacent inspection image frames. When the center position change result indicates that the center position change of the contaminated candidate region is within a preset image position range, and the region overlap result indicates that the region overlap state of the contaminated candidate region meets the preset region overlap condition, a fixed value indicating that the contaminated candidate region remains fixed in the image coordinates is generated.

[0056] Subsequently, based on the correspondence of pipe wall features in different inspection image frames, the pipe wall features covered by the contamination candidate area in each inspection image frame are determined, and it is judged whether the pipe wall features covered by the contamination candidate area in different inspection image frames belong to the same pipe wall feature, thus obtaining the pipe wall feature preservation result. When the pipe wall feature preservation result indicates that the pipe wall features covered by the contamination candidate area in different inspection image frames belong to the same pipe wall feature, a corresponding value is generated to indicate that the contamination candidate area is fixed to the same pipe wall feature, and this corresponding value is used as the basis for subsequent judgment of defect candidate areas. When the pipe wall feature preservation result indicates that the pipe wall features covered by the contamination candidate area in different inspection image frames do not belong to the same pipe wall feature, a corresponding value is generated to indicate that the contamination candidate area is not fixed to the same pipe wall feature, and this corresponding value is used as the basis for subsequent judgment of water droplet interference areas.

[0057] Furthermore, the region boundaries and coverage of the contaminated candidate region in consecutive inspection image frames are obtained. Coverage change results are generated based on the coverage changes of the contaminated candidate region in adjacent inspection image frames. Boundary expansion results are generated based on the region boundaries of the contaminated candidate region in the previous inspection image frame and the region boundaries in the next inspection image frame. Continuous expansion results are obtained based on the boundary expansion results in consecutive inspection image frames. When the coverage change results indicate that the coverage of the contaminated candidate region is increasing, and the continuous expansion results indicate that the region boundaries of the contaminated candidate region are continuously expanding outward, a diffusion trajectory indicating that the contaminated candidate region is in a diffusion state is generated.

[0058] S4. Determine whether the pollution candidate area belongs to the water droplet interference area based on the fixed value, corresponding value and diffusion trajectory. If the pollution candidate area belongs to the water droplet interference area, determine the pollution video segment, identify the water droplet interference area in each inspection image frame within the pollution video segment as the corresponding pollution freezing area, generate the freezing result, and pause the pipeline defect confirmation output within each pollution freezing area.

[0059] See Figure 2 The method for determining whether a pollution candidate area belongs to a water droplet interference area includes steps S101 to S105.

[0060] S101, Based on the fixed value, determine whether the pollution candidate area remains within the same image coordinate range in the continuous inspection image frames, and obtain the image fixation result.

[0061] Specifically, when the change in the center position of the contamination candidate area is within the preset image position range, and the overlap state of the regions in adjacent inspection image frames meets the preset region overlap condition, the image fixation result indicates that the contamination candidate area remains fixed in the image coordinates.

[0062] S102, Based on the corresponding value, determine whether the pipe wall features corresponding to the pollution candidate area in different inspection image frames belong to the same pipe wall feature, and obtain the pipe wall correspondence result.

[0063] Specifically, when the pipe wall features covered by the contamination candidate area in different inspection image frames belong to the same pipe wall feature, the pipe wall correspondence result indicates that the contamination candidate area is fixedly corresponding to the same pipe wall feature; when the covered pipe wall features do not belong to the same pipe wall feature, the pipe wall correspondence result indicates that the contamination candidate area is not fixedly corresponding to the same pipe wall feature.

[0064] S103, based on the diffusion trajectory, determine whether the coverage area of ​​the pollution candidate area has increased and whether the regional boundary has continuously expanded outward, and obtain the diffusion confirmation result.

[0065] Specifically, when the coverage area of ​​the contamination candidate area increases and the area boundary continues to expand outward in consecutive inspection image frames, the diffusion confirmation result indicates that the contamination candidate area is in a diffusion state.

[0066] S105, when the pipe wall corresponding result characterizes the contamination candidate area as the same pipe wall feature, the contamination candidate area is retained as the defect candidate area.

[0067] S104, when the image fixation result indicates that the pollution candidate area remains fixed in the image coordinates, the pipe wall correspondence result indicates that the pollution candidate area does not correspond to the same pipe wall feature, and the diffusion confirmation result indicates that the pollution candidate area is in a diffusion state, the pollution candidate area is determined as the water droplet interference area.

[0068] The image fixation results, pipe wall correspondence results, and diffusion confirmation results are used together as inputs to determine whether the contamination candidate area belongs to the water droplet interference area.

[0069] If the contamination candidate area belongs to the water droplet interference area, then the contamination video segment is determined. For each inspection image frame in the contamination video segment where the water droplet interference area is identified, the water droplet interference area in the corresponding inspection image frame is determined as the contamination freezing area corresponding to that inspection image frame, the freezing result is generated, and the pipeline defect confirmation output in each contamination freezing area is paused.

[0070] Since the image content within the water droplet interference zone originates from contamination on the lens surface rather than defects in the pipe wall itself, continuing to output conclusions of cracks, leaks, or corrosion in this area could easily lead to false positives. Conversely, directly stopping defect identification for the entire inspection video would waste information from uncontaminated areas. Therefore, this embodiment only suspends the output of pipeline defect confirmation within each contaminated / frozen area.

[0071] In one specific embodiment, firstly, the inspection image frame in which the water droplet interference area first appears is determined as the pollution start frame. Subsequently, when no water droplet interference area is identified in at least two consecutive inspection image frames, the first inspection image frame in the sequence of consecutive inspection image frames in which no water droplet interference area is identified is determined as the pollution end frame; when no pollution end frame is obtained, the last frame of the inspection video is determined as the pollution end frame. Further, a pollution video segment is determined based on the pollution start frame and the pollution end frame. For each inspection image frame in the pollution video segment in which a water droplet interference area is identified, the water droplet interference area in that inspection image frame is determined as the pollution frozen area corresponding to that inspection image frame; the boundary of each pollution frozen area is consistent with the boundary of the corresponding water droplet interference area. Finally, a freezing result is generated based on the pollution video segment and the pollution frozen areas corresponding to each inspection image frame, and the pipeline defect confirmation output within each pollution frozen area is paused.

[0072] Before performing a re-collection, it is necessary to confirm whether the water droplets or water film on the lens have been removed. If the lens contamination has not been removed, even if the tracked pipeline robot returns to the re-collection starting point to re-collect video, the re-collected images may still be in a low-confidence state and cannot be used to confirm the actual defects;

[0073] In one specific embodiment, firstly, a lens cleaning prompt is generated, instructing the operator to clean the lens of the tracked pipeline robot. Then, a cleaning confirmation image is acquired using the camera on the tracked pipeline robot, and the range of each contaminated frozen area within the contaminated video segment in the image coordinates is used as a judgment range to determine whether all low-confidence areas within each judgment range have disappeared. Specifically, if at least one judgment range still has unclear pipe wall feature boundaries, highlighted area occlusion, or discontinuous pipe wall feature texture, it is determined that the low-confidence areas within the image coordinate range corresponding to each contaminated frozen area have not all disappeared; if the pipe wall feature boundaries within each judgment range are all restored to clarity, and the highlighted area occlusion and discontinuous pipe wall feature texture are eliminated, it is determined that all low-confidence areas within the image coordinate range corresponding to each contaminated frozen area have disappeared. When all low-confidence areas have disappeared, re-sampling is allowed; when the low-confidence areas have not all disappeared, the freezing result is maintained, and a new lens cleaning prompt is generated.

[0074] During the cleaning and confirmation process to determine if low-confidence areas have disappeared, the recognition confidence level within the image coordinate range corresponding to each contaminated frozen area is recalculated in the cleaning confirmation image. If at least one low-confidence source with unclear boundaries, highlight occlusion, or texture discontinuity still exists within at least one image coordinate range, it is determined that the low-confidence areas within the image coordinate range corresponding to each contaminated frozen area have not all disappeared; if the pipe wall feature boundaries within each image coordinate range are all restored to clarity, and the highlight occlusion and texture interruption states are eliminated, it is determined that the low-confidence areas within the image coordinate range corresponding to each contaminated frozen area have all disappeared.

[0075] S5. Based on the freezing results, determine the starting point for re-mining, re-acquire inspection videos of the re-mining section, and obtain re-mining images.

[0076] In one specific embodiment, firstly, based on the pollution start frame, pollution end frame, and pollution-frozen regions corresponding to each inspection image frame in the freezing results, inspection image frames located before the pollution start frame and within the image coordinate range corresponding to each pollution-frozen region with an identification confidence level not lower than a preset stability threshold are identified and determined as re-sampling reference frames. Subsequently, the pipe position corresponding to the re-sampling reference frame is determined based on the pipe wall features in the re-sampling reference frame, and this pipe position is determined as the re-sampling start point. Further, the pipe position corresponding to the pollution end frame is determined based on the pipe wall features in the pollution end frame;

[0077] In one specific embodiment, the pipe positions corresponding to the re-mining reference frame and the pollution end frame can be corrected by combining the frame order of the inspection video and the travel distance of the tracked pipeline robot. The travel distance of the tracked pipeline robot can be obtained from track encoder records, cumulative results of travel control commands, or mileage records during the inspection process. When the forward movement corresponds to the acquisition time, the estimated travel distance of the corresponding inspection image frame can be obtained. Then, by combining the pipe wall features in the re-mining reference frame and the pipe wall features in the pollution end frame, the pipe positions corresponding to the re-mining start point and the pollution end frame are corrected respectively. Finally, the pipe segment between the re-mining start point and the pipe position corresponding to the pollution end frame is determined as the re-mining segment, and the inspection video of the re-mining segment is re-acquired from the re-mining start point to obtain the re-mining image.

[0078] S6 re-identifies defect candidate regions based on re-sampled images, generating either true defect results or contamination pseudo-anomaly results.

[0079] In one specific embodiment, when the tracked pipeline robot cannot return to the re-sampling starting point, or when the re-acquired re-sampling image still contains low-confidence areas corresponding to each contaminated frozen area, the actual defect result is not directly generated. Instead, the frozen result is maintained, and the pipeline defect confirmation output within each contaminated frozen area in the contaminated video segment remains paused. Subsequently, a review prompt is generated based on the pipeline position corresponding to the re-sampling starting point, the end frame of the contamination, and each contaminated frozen area for subsequent manual review or re-sampling. This avoids using re-sampling images still in a low-confidence state as the basis for judging actual defects when the re-sampling conditions are not met.

[0080] To avoid including false anomalies caused by lens contamination in the inspection report, this embodiment re-identifies the defect candidate area in the re-acquired image and uses pipe wall features again to determine the source of the defect candidate area.

[0081] In one specific embodiment, firstly, pipe wall features are re-extracted based on the re-acquired images, and defect candidate areas are re-identified. Then, it is determined whether the defect candidate area consistently corresponds to the same pipe wall feature in consecutively re-acquired images. When the defect candidate area consistently corresponds to the same pipe wall feature, it indicates that the defect candidate area originates from the pipe wall itself, generating a true defect result. When the defect candidate area does not consistently correspond to the same pipe wall feature, and the defect candidate area and the water droplet interference area remain consistent in image coordinates, it indicates that the defect candidate area originates from lens contamination, generating a contamination pseudo-anomaly result. Here, the consistency between the defect candidate area and the water droplet interference area in image coordinates means that the center position changes of the defect candidate area and the water droplet interference area are within a preset image position range, and their overlapping regions meet a preset region overlap condition. When no defect candidate area is re-identified within the image coordinate range corresponding to the original water droplet interference area in the re-acquired image, and the corresponding pipe wall feature in the re-acquired image can be normally identified, the abnormal area in the original contaminated video segment is determined as a contamination pseudo-anomaly result caused by lens contamination. Therefore, this method can cover the situation where the anomaly completely disappears after re-acquired images, avoiding the need to make a judgment only when the defect candidate area reappears in the re-acquired image. Finally, the actual defect results were used in the pipeline defect inspection report, and the false pollution anomaly results were used to mark the source of the anomaly in the original pollution video segment.

[0082] The above description is only a preferred embodiment 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 video inspection method for pipeline defects based on a tracked pipeline robot and its inventive concept, should be covered within the scope of protection of the present invention.

Claims

1. A video inspection method for pipeline defects based on a tracked pipeline robot, characterized in that, Includes the following steps: Collect inspection videos and obtain inspection image frames and pipe wall features; Low-confidence regions are identified based on the image quality status of the inspected image frames, and contamination candidate regions are generated from the low-confidence regions. Based on the changes in the image coordinates of the pollution candidate area, the correspondence with the pipe wall features, and the regional change status, fixed values, corresponding values, and diffusion trajectories are generated respectively. Determine whether the pollution candidate area belongs to the water droplet interference area based on the fixed value, corresponding value, and diffusion trajectory; If the contamination candidate area belongs to the water droplet interference area, then the contamination video segment is determined. For each inspection image frame in the contamination video segment where the water droplet interference area is identified, the water droplet interference area in the corresponding inspection image frame is determined as the contamination freezing area corresponding to that inspection image frame, the freezing result is generated, and the pipeline defect confirmation output in each contamination freezing area is paused. Based on the freezing results, the starting point for re-mining was determined, and inspection videos of the re-mining section were re-acquired to obtain re-mining images; Based on the re-sampling images, defect candidate regions are re-identified to generate real defect results or contamination pseudo-anomaly results; Methods for generating fixed values ​​include: Obtain the center position and region boundary of the contamination candidate area in each inspection image frame; When the change in the center position is within the preset image position range, and the overlap of regions in adjacent inspection image frames meets the preset region overlap condition, a fixed value is generated to represent the pollution candidate region in the image coordinates. Methods for generating corresponding values ​​include: Obtain the image coordinates of the contamination candidate area in each inspection image frame; Based on the image coordinates of the contamination candidate area in each inspection image frame and the correspondence of pipe wall features in different inspection image frames, the pipe wall features covered by the contamination candidate area in each inspection image frame are determined. When the pipe wall features covered by the contamination candidate area in different inspection image frames belong to the same pipe wall feature, a corresponding value representing the contamination candidate area corresponding to the same pipe wall feature is generated. When the pipe wall features covered by the contamination candidate area in different inspection image frames do not belong to the same pipe wall feature, a corresponding value is generated to represent that the contamination candidate area does not correspond to the same pipe wall feature. Methods for generating diffusion trajectories include: Obtain the region boundaries and coverage of contaminated candidate areas in consecutive inspection image frames; As the coverage area increases and the regional boundary continues to expand outward, a diffusion trajectory is generated to characterize the diffusion state of the pollution candidate area. Methods for determining whether a contamination candidate area belongs to a water droplet interference area include: Based on the fixed value, it is determined whether the pollution candidate area remains within the same image coordinate range in consecutive inspection image frames, thus obtaining the image fixation result; Based on the corresponding values, determine whether the pipe wall features corresponding to the pollution candidate area in different inspection image frames belong to the same pipe wall feature, and obtain the pipe wall correspondence result; Based on the diffusion trajectory, it is determined whether the coverage area of ​​the pollution candidate area has increased and whether the regional boundary has continuously expanded outward, thus obtaining the diffusion confirmation result. When the image fixation result indicates that the contamination candidate area remains fixed in the image coordinates, the pipe wall correspondence result indicates that the contamination candidate area does not correspond to the same pipe wall feature, and the diffusion confirmation result indicates that the contamination candidate area is in a diffused state, the contamination candidate area is identified as the water droplet interference area. When the pipe wall corresponding result characterizes the contamination candidate area as the same pipe wall feature, the contamination candidate area is retained as the defect candidate area.

2. The pipeline defect video inspection method based on a tracked pipeline robot according to claim 1, characterized in that, Methods for obtaining pipe wall features include: Identify candidate pipe wall features from pipe interface loops, pipe wall textures, waterline boundaries, and sedimentation boundaries present in the current inspection image frame; When a candidate pipe wall feature appears repeatedly in consecutive inspection image frames and its positional changes match the robot's direction of travel, it is identified as a pipe wall feature, and the correspondence between the pipe wall features in different inspection image frames is recorded.

3. The pipeline defect video inspection method based on a tracked pipeline robot according to claim 1, characterized in that, Methods for identifying low-confidence regions include: Each inspection image frame is divided into at least two image regions; Based on the clarity of the pipe wall feature boundaries, the coverage of the highlighted area, and the proportion of continuous texture recognition, the recognition credibility of each image region is generated. When the recognition confidence level is lower than the preset stability threshold, the corresponding image region is identified as a low confidence region.

4. The pipeline defect video inspection method based on a tracked pipeline robot according to claim 3, characterized in that, Methods for generating contamination candidate regions include: Merge adjacent and continuous low-confidence regions to obtain candidate coverage areas; When a candidate coverage area has at least one low-confidence source among unclear boundaries, highlight occlusion, or texture discontinuity, and it appears repeatedly in consecutive inspected image frames, the candidate coverage area is identified as a contaminated candidate area.

5. The pipeline defect video inspection method based on a tracked pipeline robot according to claim 1, characterized in that, Methods for generating frozen results include: The first inspection image frame appearing in the water droplet interference area is identified as the pollution initiation frame. When no water droplet interference area is identified in at least two consecutive inspection image frames, the first inspection image frame in the consecutive inspection image frame sequence in which no water droplet interference area is identified is determined as the pollution end frame. If the pollution end frame is not obtained, the last frame of the inspection video will be determined as the pollution end frame; The contaminated video segment is determined based on the contamination start frame and contamination end frame; for each inspection image frame in the contaminated video segment where water droplet interference areas are identified, the water droplet interference area in the corresponding inspection image frame is determined as the contamination freezing area corresponding to that inspection image frame; a freezing result is generated based on the contaminated video segment and the contamination freezing area corresponding to each inspection image frame.

6. The pipeline defect video inspection method based on a tracked pipeline robot according to claim 5, characterized in that, Methods for obtaining resampled images include: Generate lens cleaning tips based on the freeze results; Acquire a confirmation image of the cleaning process after lens cleaning; If the low confidence areas corresponding to each contaminated frozen area in the clean confirmation image have not all disappeared, retain the freeze results and regenerate the lens cleaning prompt; When all low-confidence areas corresponding to each contaminated frozen area in the clean confirmation image have disappeared, find the inspection image frame before the contamination start frame and whose corresponding image area recognition confidence is not lower than the preset stable threshold, and determine it as the re-collection reference frame. The starting point of the re-sampling is determined based on the pipe wall features in the re-sampling reference frame, and the pipe position corresponding to the end of the pollution frame is determined based on the pipe wall features in the end of the pollution frame. The pipe segment between the starting point of the re-sampling and the pipe position corresponding to the end of the pollution frame is re-acquired to obtain the re-sampling image.

Images