Computer vision-based treatment cartridge body surface contour image enhancement and site identification method

Abstract

The present application relates to a computer vision-based treatment cavity surface contour image enhancement and site recognition method, comprising: acquiring two or more frames of surface images continuously collected when a user is in a treatment cavity; performing edge extraction on each frame of surface image to obtain a candidate contour segment and its two adjacent image regions; taking a fixed background region in the treatment cavity as a reference, calculating the displacement amount, displacement direction and displacement matching degree of the candidate contour segment in the adjacent frame and the suspected human image region, and determining the surface contour segment or the non-surface contour segment accordingly; enhancing the surface contour segment and suppressing the non-surface contour segment to generate a surface contour enhancement image, extracting a candidate treatment site in the human region defined by the surface contour enhancement image, and outputting an effective site mark or a non-effective site mark according to the distance relationship between the candidate treatment site and the surface contour segment and the non-surface contour segment exclusion result. The present application can reduce background false edge interference and improve the reliability of treatment site recognition.

Description

Technical Field

[0001] This invention relates to the field of image processing and visual positioning technology for intelligent treatment equipment, specifically a method for enhancing the contour image of the treatment chamber surface and identifying locations based on computer vision. Background Technology

[0002] Existing technologies in intelligent therapeutic devices typically employ image acquisition, 3D scanning, or deep learning models to identify the human body surface, acupoints, or treatment-related sites. For example, Chinese patent publication CN119722940A discloses a collision avoidance method and system for radiotherapy based on 3D scanning and real-time simulation. This method generates an initial 3D model based on patient surface data and updates the model during radiotherapy using real-time scanning data to perform collision avoidance simulation and path risk assessment. While this approach utilizes surface data for safety control during treatment, its technical focus is on 3D model creation and collision avoidance planning for the movement path of the therapeutic device. It does not address the issue of the blurring between the real human body contours in two-dimensional or continuous frame images within the treatment chamber and the fixed background, support edges, and lighting / shadow edges of the chamber, nor does it provide a method for determining the authenticity of the contours based on the displacement features of adjacent frames.

[0003] For example, Chinese patent publication CN117636446A discloses a method for locating facial acupoints, an acupuncture method, an acupuncture robot, and a storage medium. This method utilizes a depth camera to acquire RGB and depth images of the face and generate 3D point cloud data. It then uses a facial segmentation model to obtain feature regions such as eyebrows, eyes, nose, and mouth, and determines the acupoint locations based on the correlation between these feature regions and key acupoints. While this approach improves the efficiency and real-time performance of acupoint location in specific areas, it relies on relatively clear target region segmentation and pre-defined acupoint correlations. It is suitable for relatively stable and feature-rich local areas such as the face. However, in scenarios like treatment chambers where user posture changes, the chamber background is relatively fixed, and both body and non-body contours appear in the image simultaneously, problems may still arise such as enhanced false edges in the background, treatment sites falling into non-human areas, or incorrect exclusion of effective sites.

[0004] Therefore, when performing body contour enhancement and treatment site identification in a treatment chamber, relying solely on single-frame edge extraction, region segmentation, or pre-trained site models is insufficient to stably distinguish between real body contours that change with human movement and non-body contours that remain fixed relative to the chamber. Furthermore, it is difficult to prevent newly added false edges from misleading candidate treatment sites after contour enhancement. Especially when the chamber space is limited, the background structure is fixed, the acquisition angle is restricted, and the distance between the user's body surface and the chamber boundary is close, the strength of edge responses in the image does not necessarily correspond to the real body boundary. A technical solution is needed that can use the fixed background of the treatment chamber as a reference, combined with the displacement amount, displacement direction, and matching relationship of suspected human image regions, to filter, enhance, and eliminate candidate contour segments, and further output a validity identifier for candidate treatment sites based on this. Summary of the Invention

[0005] The purpose of this invention is to provide a method for enhancing the contour image of the treatment chamber and identifying the location based on computer vision, thereby solving some of the drawbacks and shortcomings pointed out in the background art.

[0006] The present invention adopts the following technical solution to solve the above-mentioned technical problems:

[0007] Acquire two or more consecutive body surface images of the user while they are in the treatment chamber, and perform edge extraction on each body surface image to obtain candidate contour segments and adjacent image regions on both sides of the candidate contour segments;

[0008] Using the fixed background area inside the treatment chamber as a reference, the displacement amount and displacement direction of the candidate contour segment in adjacent frames are calculated, and the displacement matching degree between the candidate contour segment and the suspected human image area is calculated. When the displacement amount of the candidate contour segment relative to the fixed background area is greater than the first displacement threshold and the displacement matching degree meets the preset matching conditions, it is determined to be a body surface contour segment; otherwise, it is determined to be a non-body surface contour segment.

[0009] Enhance the body surface contour segments and suppress non-body surface contour segments to generate a body surface contour enhanced image. Extract candidate treatment sites within the human body region defined by the body surface contour enhanced image. Based on the distance relationship between the candidate treatment sites and the body surface contour segments and the exclusion results of non-body surface contour segments, output the effective site identifier or ineffective site identifier of the candidate treatment sites.

[0010] Further, when determining the candidate contour segment, a first side image region and a second side image region are obtained on both sides of the candidate contour segment respectively; when the displacement matching degree between the first side image region and the suspected human body image region meets the preset matching condition, and the displacement of the second side image region relative to the fixed background region is less than the second displacement threshold, the candidate contour segment is determined to be the body surface contour segment; when the displacement of both the first side image region and the second side image region relative to the fixed background region is less than the second displacement threshold, the candidate contour segment is determined to be the non-body surface contour segment.

[0011] Furthermore, after enhancing the body surface contour segment, the newly appearing enhanced edge segment in the enhanced body surface contour image is subjected to original evidence verification; when the enhanced edge segment does not have a corresponding candidate contour segment that meets the preset matching conditions in two or more body surface images before enhancement, or when the displacement relative to the fixed background area is less than the second displacement threshold, the enhanced edge segment is marked as the non-body surface contour segment and removed from the enhanced body surface contour image.

[0012] Furthermore, when outputting the valid site identifier or invalid site identifier of the candidate treatment site, a preset neighborhood is determined with the candidate treatment site as the center, and a contour support path formed by continuous connection of the body surface contour segments is determined within the preset neighborhood; when the connection area between the candidate treatment site and the contour support path is not interrupted by the non-body surface contour segments, the valid site identifier is output; when the connection area is interrupted by the non-body surface contour segments, the invalid site identifier is output.

[0013] Furthermore, when performing original evidence verification on the enhanced edge segment, grayscale sampling sequences are taken on both sides of the normal of the enhanced edge segment; when the gradient direction of the grayscale sampling sequence does not match the grayscale change order on both sides of the corresponding candidate contour segment in two or more frames of body surface images before enhancement, the enhanced edge segment is marked as the non-body surface contour segment.

[0014] Furthermore, when verifying the original evidence for the enhanced edge segment, the relative positional change between the enhanced edge segment and the fixed background region is compared; when the edge response value of the enhanced edge segment in the enhanced body contour image is higher than the edge response value before enhancement, and the positional change of the enhanced edge segment relative to the fixed background region in two or more frames of body surface images before enhancement is lower than a preset change, the enhanced edge segment is determined to be a false background edge generated by enhancement.

[0015] Furthermore, when verifying the original evidence for the enhanced edge segment, it is determined whether the enhanced edge segment changes the connectivity of the human body region defined by the body surface contour segment; when the enhanced edge segment divides the same human body region into two unconnected regions, and there are no corresponding candidate contour segments that meet the preset matching conditions in the two or more frames of body surface images before enhancement, the enhanced edge segment is removed from the enhanced body surface contour image.

[0016] Furthermore, when determining whether the candidate treatment site and the contour support path are separated by the non-body surface contour segment, a connection region is generated along the direction from the candidate treatment site to the contour support path; when the displacement matching degree between the non-body surface contour segment in the connection region and the suspected human image region does not meet the preset matching condition, and both ends of the non-body surface contour segment are respectively connected to the fixed background region, it is determined that the candidate treatment site and the contour support path are separated.

[0017] Furthermore, when multiple contour support paths exist within the preset neighborhood, a contour support path whose distance to the candidate treatment site is less than a preset distance threshold and which does not pass through the non-body surface contour segment is selected as the target support path; when the displacement directions of the body surface contour segments corresponding to both ends of the target support path match in adjacent frames, the effective site identifier is output.

[0018] Furthermore, after determining that the candidate treatment site and the contour support path are separated, bypass contour segments that meet the preset matching conditions with the displacement matching degree of the suspected human image region are retrieved along both sides of the connecting region; when there is a bypass contour segment on either side of the two sides, and the bypass contour segment is connected to the contour support path, the ineffective site identifier of the candidate treatment site is adjusted to the site identifier to be reviewed.

[0019] This invention acquires two or more consecutive frames of body surface images while the user is inside the treatment chamber. It then performs edge extraction on each frame of the body surface image to obtain candidate contour segments. Using a fixed background area inside the treatment chamber as a reference, it calculates the displacement amount, displacement direction, and displacement matching degree of the candidate contour segments in adjacent frames. This allows the invention to distinguish body surface contour segments from non-body surface contour segments by utilizing the dynamic difference between the human body and the background, thereby reducing the probability that background factors such as chamber edges, fixed support structures, and reflective shadows are mistaken for human body contours.

[0020] This invention, after determining the body surface contour segments, enhances these segments and suppresses non-body surface contour segments to generate an enhanced body surface contour image. This makes the boundaries of the human body region clearer, and candidate treatment sites are extracted within this region. Since the validity of candidate treatment sites is output based on their distance from the body surface contour segments and the exclusion of non-body surface contour segments, it reduces misjudgments caused by false edges, background interruptions, or interference from fixed structures within the treatment site identification process, thus improving the reliability and applicability of the treatment site identification results. Attached Figure Description

[0021] Figure 1 This is a flowchart of the treatment chamber surface contour image enhancement and site recognition process of the present invention.

[0022] Figure 2 This is a relative displacement determination diagram of candidate contour segments in Embodiment 1 of the present invention.

[0023] Figure 3 This is a graph showing the displacement matching degree and threshold determination in Embodiment 1 of the present invention.

[0024] Figure 4 This is the original evidence verification image for the enhanced edge segment in Embodiment 1 of the present invention.

[0025] Figure 5 This is the preset neighborhood and contour support path selection diagram in Embodiment 2 of the present invention.

[0026] Figure 6 This is a verification diagram of the connection area partition in Embodiment 2 of the present invention.

[0027] Figure 7 This is an adjustment diagram of the verification mark for the bypass contour segment in Embodiment 2 of the present invention. Detailed Implementation

[0028] As attached Figure 1 As shown, in one specific embodiment, an image acquisition device is installed inside the treatment chamber, continuously acquiring images of the user's body surface area. The image acquisition device can be installed on the inner wall, top, side wall, or a support structure fixedly connected to the treatment chamber, and its acquisition direction and range cover the user's body surface area to be identified. When the user is inside the treatment chamber, the system acquires two or more consecutively acquired body surface images and performs edge extraction processing on each frame. Edge extraction can be achieved using gradient detection, edge response calculation, or contour tracking to obtain multiple candidate contour segments. For each candidate contour segment, the system determines adjacent image regions on both sides of the candidate contour segment for subsequent determination of whether the candidate contour segment belongs to the human body surface boundary or the background edge formed by the fixed structure inside the treatment chamber.

[0029] After extracting candidate contour segments, the system selects a fixed background region within the treatment chamber as a reference region. This fixed background region can be the inner wall of the treatment chamber, the positioning support, the edge of the chamber, or other regions with stable positions in continuous images. The system can determine the corresponding reference position in adjacent frame body surface images based on the texture features, edge features, or pre-calibrated position features of the fixed background region, and then perform position calibration on the adjacent frame body surface images based on the corresponding reference position. The system performs position calibration on the adjacent frame body surface images based on the fixed background region and calculates the displacement amount and direction of the candidate contour segment in the adjacent frames. Simultaneously, the system determines suspected human image regions based on the image texture, motion region, or initial human body region recognition results surrounding the candidate contour segment, and calculates the displacement matching degree between the candidate contour segment and the suspected human image region. Suspected human image regions can be determined through continuous frame difference regions, skin color or body surface texture regions, regions located within the preset human activity range of the treatment chamber, or regions enclosed by the effective body surface contour segments from the previous moment. The displacement matching degree can be determined based on the difference in displacement direction, the difference in displacement amount, or the similarity of motion vectors. The first displacement threshold, second displacement threshold, preset matching conditions, preset change amount, and preset distance threshold can be preset during system initialization or treatment chamber calibration, or adjusted based on image resolution, image acquisition frequency, stability of the fixed background area within the treatment chamber, and the normal movement amplitude of the user's body surface in consecutive frames. When the displacement of the candidate contour segment relative to the fixed background area is greater than the first displacement threshold, and the displacement matching degree between the candidate contour segment and the suspected human image area meets the preset matching conditions, it indicates that the candidate contour segment moves synchronously with the human body surface, and the system determines the candidate contour segment as a body surface contour segment. Conversely, if the candidate contour segment does not reach the first displacement threshold relative to the fixed background area, or the displacement matching degree with the suspected human image area does not meet the preset matching conditions, it indicates that the candidate contour segment is more likely to originate from the treatment chamber background, image noise, or non-human body edges, and the system determines the candidate contour segment as a non-body surface contour segment.

[0030] After identifying the body surface contour segments and non-body surface contour segments, the system enhances the body surface contour segments and suppresses the non-body surface contour segments. Enhancement processing may include improving the edge response of the body surface contour segments, extending broken adjacent body surface contour segments, smoothing the contour segment direction, or enhancing the grayscale difference between the body surface contour segments and surrounding areas. Suppression processing may include reducing the edge response of the non-body surface contour segments, deleting isolated noisy edges, weakening edge segments connected to a fixed background region, or excluding the region containing the non-body surface contour segments during subsequent site identification. Enhancement processing improves the edge response, continuity, and recognizability of the body surface contour segments, while suppression processing reduces the interference of background edges or false edges on site identification. After the above processing, the system generates a body surface contour enhanced image. The system extracts candidate treatment sites within the human body region defined by the enhanced body surface contour image and calculates the distance relationship between the candidate treatment sites and the body surface contour segments. Candidate treatment sites can be determined based on a preset treatment site template, the calibration relationship between the treatment chamber coordinate system and the image coordinate system, local image features within the human body region, or externally input information about the location to be treated. Subsequently, the system combines the exclusion results of non-surface contour segments to determine whether the candidate treatment site is located within a reliable human body surface area. If the positional relationship between the candidate treatment site and the surface contour segment meets the treatment site identification conditions and is not interfered with by non-surface contour segments, a valid site identifier for the candidate treatment site is output. If the positional relationship between the candidate treatment site and the surface contour segment does not meet the treatment site identification conditions, or is interfered with by non-surface contour segments, an invalid site identifier for the candidate treatment site is output.

[0031] Based on the above specific implementation, when determining a candidate contour segment, the system can obtain a first-side image region and a second-side image region on both sides of the candidate contour segment along the normal direction of the candidate contour segment. The first-side image region and the second-side image region can be image blocks located within a preset width range on both sides of the candidate contour segment, or they can be sampling bands continuously distributed along the extension direction of the candidate contour segment. The system calculates the relative displacement between the first-side image region, the second-side image region and the fixed background region, and calculates the displacement matching degree between the first-side image region and the suspected human image region. The direction of obtaining the first-side image region and the second-side image region can be determined according to the local extension direction of the candidate contour segment. When the candidate contour segment is curved, the system can obtain the image regions on both sides along multiple sampling positions of the candidate contour segment, and combine the judgment results of multiple sampling positions to obtain the judgment result of the candidate contour segment.

[0032] When the displacement matching degree between the first-side image region and the suspected human image region meets the preset matching conditions, and the displacement of the second-side image region relative to the fixed background region is less than the second displacement threshold, it indicates that one side of the candidate contour segment has image features that move with the human body, while the other side has relatively fixed background features. At this time, the candidate contour segment is located at the boundary between the human body surface region and the treatment chamber background region, so the system determines the candidate contour segment as a body surface contour segment. When the displacement of both the first-side and second-side image regions relative to the fixed background region is less than the second displacement threshold, it indicates that the image regions on both sides of the candidate contour segment remain relatively stable with the fixed background, and the candidate contour segment does not have the feature of moving with the human body surface, so the system determines the candidate contour segment as a non-body surface contour segment. When both the first-side and second-side image regions show movement with the suspected human image region, the system can combine the connectivity between the candidate contour segment and the confirmed body surface contour segment, whether the candidate contour segment is located near the boundary of the human body region, and whether it passes through the interior of the human body region to further determine whether the candidate contour segment belongs to the outer boundary of the body surface, the inner texture edge of the body surface, or the contour segment to be verified.

[0033] Based on the above specific implementation, after enhancing the body surface contour segment, the system performs original evidence verification on newly appearing enhanced edge segments in the enhanced body surface contour image. Original evidence verification is used to determine whether the enhanced edge segment originates from genuine contour evidence in the pre-enhancement body surface image, avoiding the enhancement process from mistakenly identifying background textures, noise edges, or algorithm-compensated edges as human body surface contours. The system performs position matching, direction matching, and motion matching between the enhanced edge segment and two or more pre-enhancement body surface images. Position matching can be used to determine whether a candidate contour segment exists at the corresponding position of the enhanced edge segment in the pre-enhancement image; direction matching can be used to determine whether the extension direction of the enhanced edge segment is consistent with the corresponding candidate contour segment; and motion matching can be used to determine whether the motion characteristics of the enhanced edge segment in consecutive frames are consistent with the suspected human image region. When there is no corresponding candidate contour segment that meets the preset matching conditions in two or more pre-enhancement body surface images for the enhanced edge segment, or when the displacement of the enhanced edge segment relative to the fixed background region is less than a second displacement threshold, it indicates that the enhanced edge segment lacks original image evidence or has fixed background edge features. Therefore, the system marks the enhanced edge segment as a non-body surface contour segment and removes it from the enhanced body surface contour image.

[0034] During the verification of original evidence, the system can also obtain grayscale sampling sequences on both sides of the normal of the enhanced edge segment. The grayscale sampling sequences are used to characterize the direction and order of grayscale changes on both sides of the enhanced edge segment. The system searches for candidate contour segments corresponding to the enhanced edge segment in two or more frames of body surface images before enhancement, and compares the gradient direction of the grayscale sampling sequence with the order of grayscale changes on both sides of the corresponding candidate contour segment. When the gradient direction of the grayscale sampling sequence does not match the order of grayscale changes on both sides of the corresponding candidate contour segment, it indicates that the grayscale transition features on both sides of the enhanced edge segment are inconsistent with the original candidate contour segment. The enhanced edge segment may be formed by background texture enhancement or local noise amplification. Therefore, the system marks the enhanced edge segment as a non-body surface contour segment. If the grayscale sampling sequence is affected by local shadows, reflections, or changes in treatment chamber lighting, the system can combine the motion matching results in adjacent frames and the positional relationship between the enhanced edge segment and the suspected human body image area to make a joint judgment, so as to avoid the body surface contour segment being mistakenly deleted due to local brightness changes alone.

[0035] During the verification of original evidence, the system can also compare the relative positional changes between the enhanced edge segment and the fixed background region. The system acquires the edge response value of the enhanced edge segment in the enhanced body contour image, as well as the edge response value and positional change of the enhanced edge segment in at least two frames of body surface images before enhancement. When the edge response value of the enhanced edge segment in the enhanced body contour image is higher than the edge response value before enhancement, and the positional change of the enhanced edge segment relative to the fixed background region in at least two frames of body surface images before enhancement is lower than a preset change, it indicates that the enhancement process increased the response intensity of the edge segment. However, the edge segment did not undergo effective displacement with the human body surface in the original image sequence. Therefore, the system classifies the enhanced edge segment as a false background edge generated by enhancement. For enhanced edge segments identified as false background edges, the system can delete them from the enhanced body contour image or reduce their edge response value to minimize the impact of false background edges on the identification of candidate treatment sites.

[0036] During the verification of original evidence, the system can also determine whether the enhanced edge segment alters the connectivity of the human body region defined by the body surface contour segment. The system determines the connectivity of the human body region based on the boundary formed by the body surface contour segment and determines whether the addition of the enhanced edge segment divides the same human body region into two unconnected regions. When the enhanced edge segment divides the same human body region into two unconnected regions, and there are no corresponding candidate contour segments that meet the preset matching conditions in two or more frames of body surface images before enhancement, it indicates that the enhanced edge segment lacks original contour basis and causes incorrect segmentation of the human body region's topology. Therefore, the system removes the enhanced edge segment from the enhanced body surface contour image. The connectivity of the human body region can be determined based on the connection relationship between the area enclosed by the body surface contour segment, the human body region mask, or the continuous body surface boundary. After removing the enhanced edge segment, the system can update the human body region boundary so that the updated enhanced body surface contour image can be used for subsequent candidate treatment site identification.

[0037] Based on the above specific implementation, when outputting valid or invalid site identifiers for candidate treatment sites, the system determines a preset neighborhood centered on the candidate treatment site. The preset neighborhood can be determined based on the size of the treatment site, the distribution density of surface contour segments, and the image resolution. The system searches for surface contour segments within the preset neighborhood and combines continuously connected surface contour segments to form a contour support path. The contour support path is used to characterize a credible human body surface boundary near the candidate treatment site. Continuous connection can include the distance between the endpoints of adjacent surface contour segments satisfying a preset connection condition, the change in extension direction satisfying a preset smoothing condition, or the displacement direction of adjacent surface contour segments being consistent in consecutive frames. The system determines a connection region between the candidate treatment site and the contour support path and determines whether the connection region is interrupted by non-surface contour segments. When the connection region is not interrupted by non-surface contour segments, it indicates that there is a continuous spatial support relationship between the candidate treatment site and the credible human body surface boundary; therefore, the system outputs a valid site identifier. When the connected region is separated by non-body surface contour segments, it indicates that there is background edge, pseudo-edge or non-human structure interference between the candidate treatment site and the credible human body surface boundary, so the system outputs an invalid site identifier.

[0038] When determining whether a non-surface contour segment separates the candidate treatment site from the contour support path, the system generates a connection region along the direction from the candidate treatment site to the contour support path. The connection region can be a strip-shaped area covering the image channel between the candidate treatment site and the contour support path, or a local detection region extending along the line connecting the two. The system detects whether a non-surface contour segment exists within the connection region and calculates the displacement matching degree between the non-surface contour segment and the suspected human image region within the connection region. When the displacement matching degree between the non-surface contour segment and the suspected human image region within the connection region does not meet the preset matching conditions, and both ends of the non-surface contour segment are connected to fixed background regions, it indicates that the non-surface contour segment has background structural boundary features and forms a continuous barrier between the candidate treatment site and the contour support path. Therefore, the system determines that the candidate treatment site and the contour support path are separated. If multiple non-surface contour segments exist within the connection region, the system can determine the connection relationship between each non-surface contour segment and the fixed background region separately. When multiple non-surface contour segments together form a continuous barrier across the connection region, the system determines that the candidate treatment site and the contour support path are separated.

[0039] When multiple contour support paths exist within a preset neighborhood, the system can calculate the distance between each contour support path and the candidate treatment site, and determine whether each contour support path crosses a non-surface contour segment. The system selects the contour support path whose distance to the candidate treatment site is less than a preset distance threshold and which does not cross a non-surface contour segment as the target support path. Subsequently, the system detects the displacement direction of the corresponding surface contour segments at both ends of the target support path in adjacent frames. When the displacement directions of the corresponding surface contour segments at both ends of the target support path match in adjacent frames, it indicates that the target support path consists of continuous contours near the same human body surface region, which can provide reliable support for the candidate treatment site. Therefore, the system outputs a valid site identifier. If multiple target support paths meet the conditions, the system can select the target support path with higher continuity, closer distance to the candidate treatment site, or higher displacement matching degree with the suspected human image region as the final support path.

[0040] After determining that the candidate treatment site is separated from the contour support path, the system can also retrieve bypass contour segments along both sides of the connecting region. A bypass contour segment is a contour segment located laterally within the connecting region and whose displacement matching degree with the suspected human image region meets preset matching conditions. The system determines whether the bypass contour segment can bypass the non-surface contour segments within the connecting region and form a connection with the contour support path. When a bypass contour segment exists on either side of the connecting region's boundary and is connected to the contour support path, it indicates that although the candidate treatment site is separated by non-surface contour segments in the direct connection direction, it may still obtain spatial support through the lateral human contour. Therefore, the system adjusts the invalid site identifier of the candidate treatment site to a site identifier to be reviewed, so that it can be further confirmed by manual review or additional recognition rules. The site identifier to be reviewed can be output along with the candidate treatment site's location, corresponding contour support path, bypass contour segment, and the non-surface contour segment forming the separation, so that subsequent review processes can be confirmed based on the same image evidence.

[0041] Example 1:

[0042] In one specific implementation, Company A's treatment chamber is equipped with an image acquisition device that continuously acquires images of the user's body surface area. The image resolution is 1280×720, and the acquisition time interval between two adjacent frames of body surface images is 0.04s. When the user is in the treatment chamber, the system acquires two or more consecutively acquired body surface images and performs edge extraction processing on each frame to obtain multiple candidate contour segments and adjacent image regions on both sides of each candidate contour segment. Edge extraction can be achieved using gradient detection, edge response calculation, or contour tracking. The candidate contour segments can be line segments, curve segments, or locally closed boundary segments composed of continuous edge pixels.

[0043] After extracting candidate contour segments, the system selects a fixed background region within the treatment chamber as a reference region. This fixed background region can be the inner wall of the treatment chamber, the positioning support, the edge of the chamber, or other regions with stable positions in continuous images. The system determines the corresponding reference position based on the texture or edge features of the fixed background region in adjacent frame images and uses this corresponding reference position to calibrate the adjacent frame images, eliminating the effects of slight jitter in the image acquisition device or overall image shift. Figure 2 As shown, the fixed background region has only a residual offset of 0.6 pixels in adjacent frames, while the candidate contour segment and the suspected human body image region produce more obvious synchronous displacement. This allows us to distinguish the small offset caused by the background of the warehouse from the real displacement caused by the movement of the human body surface.

[0044] After calibration, the system calculates the displacement of the candidate contour segment relative to the fixed background region. This displacement can be determined by the following formula:

[0045]

[0046] in, Indicates the first The displacement of each candidate contour segment relative to a fixed background region. Indicates the first The length of the displacement vector of each candidate contour segment in two adjacent body surface images. This represents the length of the residual displacement vector of a fixed background region in two adjacent body surface images. This represents the actual displacement difference of the candidate contour segment relative to the fixed background region after background calibration. Since the fixed background region remains stable within the treatment chamber, it is determined first... Then, by subtracting the residual displacement of the fixed background area from the displacement of the candidate contour segment, a displacement that is closer to the actual movement of the human body surface can be obtained.

[0047] The system also calculates the displacement matching degree between candidate contour segments and suspected human image regions. Suspected human image regions can be determined by consecutive frame difference regions, surface texture regions, regions located within the preset human activity range of the treatment chamber, or regions enclosed by valid surface contour segments from the previous moment. The displacement matching degree can be determined by the following formula:

[0048]

[0049] in, Indicates the first The degree of displacement matching between candidate contour segments and suspected human image regions. Indicates the first The displacement of each candidate contour segment relative to a fixed background region. This indicates the displacement of a region in a suspected human image relative to a fixed background region. Indicates the first The directional difference between the displacement direction of each candidate contour segment and the displacement direction of the suspected human body image region. This indicates the allowable displacement difference reference value. This indicates the allowable directional difference reference value. and This indicates the weights of displacement difference and direction difference in the matching judgment. The system first compares the displacement difference between the candidate contour segment and the suspected human image region, and then compares the displacement direction difference between the two to determine whether the candidate contour segment moves synchronously with the human body surface.

[0050] In one example, the residual displacement of the fixed background region after calibration is 0.6 pixels, and the displacement of a candidate contour segment in an adjacent frame is 5.8 pixels. Therefore, the displacement of the candidate contour segment relative to the fixed background region is 5.2 pixels. The displacement of the suspected human image region is 6.1 pixels, the first displacement threshold is 3.0 pixels, the directional difference between the candidate contour segment and the suspected human image region is 4°, and the preset matching threshold is 0.75. Take 6.0 pixels, Take 30° Take 0.5, When the value is 0.5, The score is 0.86. Since 5.2 pixels is greater than the first displacement threshold of 3.0 pixels, and the matching degree of 0.86 meets the preset matching conditions, the system determines the candidate contour segment as a body surface contour segment.

[0051] like Figure 3 As shown, the system can simultaneously draw the relative displacement and displacement matching degree of multiple candidate contour segments, and compare them with the first displacement threshold and the matching threshold respectively. Figure 3 In the analysis, candidate contour segment C3 has a relative displacement of 5.2 pixels and a displacement matching degree of 0.86, both of which meet the judgment criteria, and is therefore confirmed as a body surface contour segment. Candidate contour segments C1, C2, and C4 do not meet the body surface contour judgment criteria due to insufficient relative displacement or displacement matching degree. Although candidate contour segment C5 also meets the threshold condition, it can be further confirmed by combining the verification results of the image regions on both sides and the enhanced edge. In this way, the system can avoid making misjudgments based solely on a single displacement amount or a single matching degree.

[0052] When determining a candidate contour segment, the system can also obtain a first-side image region and a second-side image region on both sides of the candidate contour segment along its normal direction. The first-side and second-side image regions can be image blocks located within a preset width range on both sides of the candidate contour segment, or they can be sampling bands continuously distributed along the extension direction of the candidate contour segment. When the candidate contour segment is curved, the system obtains the image regions on both sides along multiple sampling positions of the candidate contour segment and combines the judgment results from multiple sampling positions to obtain the determination result of the candidate contour segment.

[0053] When the displacement matching degree between the first-side image region and the suspected human image region meets the preset matching condition, and the displacement of the second-side image region relative to the fixed background region is less than the second displacement threshold, it indicates that one side of the candidate contour segment has image features that move with the human body, and the other side has relatively fixed background features. This candidate contour segment is located at the boundary between the human body surface area and the treatment chamber background area, and the system determines this candidate contour segment as a body surface contour segment. In the above example, the second displacement threshold is 1.5 pixels. If the displacement matching degree between the first-side image region and the suspected human image region is 0.84, and the displacement of the second-side image region relative to the fixed background region is 0.9 pixels, then the system confirms that the candidate contour segment is a body surface contour segment. When the displacement of both the first-side and second-side image regions relative to the fixed background region is less than the second displacement threshold, it indicates that both image regions on both sides of the candidate contour segment remain relatively stable with the fixed background, and the system determines this candidate contour segment as a non-body surface contour segment.

[0054] After identifying the body surface contour segments and non-body surface contour segments, the system enhances the body surface contour segments and suppresses the non-body surface contour segments. Enhancement processing may include improving the edge response of the body surface contour segments, lengthening broken adjacent body surface contour segments, smoothing the contour segment direction, or enhancing the grayscale difference between the body surface contour segments and surrounding areas. Suppression processing may include reducing the edge response of the non-body surface contour segments, removing isolated noisy edges, weakening edge segments connected to a fixed background region, or excluding the region containing the non-body surface contour segments during subsequent site identification. After these processes, the system generates an enhanced body surface contour image.

[0055] After enhancing the body surface contour segment, the system performs original evidence verification on newly appearing enhanced edge segments in the enhanced body surface contour image. The system performs position matching, orientation matching, and motion matching on the enhanced edge segment and two or more frames of body surface images before enhancement. Position matching is used to determine whether there is a candidate contour segment at the corresponding position of the enhanced edge segment in the image before enhancement; orientation matching is used to determine whether the extension direction of the enhanced edge segment is consistent with the corresponding candidate contour segment; and motion matching is used to determine whether the motion characteristics of the enhanced edge segment in consecutive frames are consistent with the suspected human body image region. When there is no corresponding candidate contour segment that meets the preset matching conditions in two or more frames of body surface images before enhancement, or when the displacement of the enhanced edge segment relative to the fixed background area is less than the second displacement threshold, it indicates that the enhanced edge segment lacks original image evidence or has fixed background edge features. The system marks the enhanced edge segment as a non-body surface contour segment and removes it from the enhanced body surface contour image.

[0056] like Figure 4As shown, the system can jointly verify the pre-enhancement response value, post-enhancement response value, and the positional change of the enhanced edge segment relative to the fixed background area. Enhanced edge segments E2 and E3 both have high edge response values ​​after enhancement, but their relative positional changes are 0.8 pixels and 0.6 pixels respectively, both less than the second displacement threshold of 1.5 pixels. This indicates that these edge segments mainly represent the enhancement of fixed background edges, rather than edges formed by the actual contours of the human body surface as the frame moves. Therefore, even though the response values ​​of E2 and E3 increase after enhancement, the system still marks them as non-contour segments and removes them to reduce the impact of false background edges on subsequent site recognition.

[0057] During the verification of original evidence, the system also acquires grayscale sampling sequences on both sides of the enhanced edge segment's normal direction. These grayscale sampling sequences characterize the direction and sequence of grayscale changes on both sides of the enhanced edge segment. The system searches for candidate contour segments corresponding to the enhanced edge segment in at least two frames of body surface images before enhancement and compares the gradient direction of the grayscale sampling sequence with the grayscale change sequence on both sides of the corresponding candidate contour segment. When the gradient direction of the grayscale sampling sequence does not match the grayscale change sequence on both sides of the corresponding candidate contour segment, it indicates that the grayscale transition features on both sides of the enhanced edge segment are inconsistent with the original candidate contour segment. This enhanced edge segment may be formed by background texture enhancement or local noise amplification, and the system marks this enhanced edge segment as a non-body surface contour segment.

[0058] During the verification of original evidence, the system also compares the relative positional changes between the enhanced edge segment and the fixed background region. The system acquires the edge response value of the enhanced edge segment in the enhanced body contour image, as well as the edge response value and positional change of the enhanced edge segment in two or more frames of body surface images before enhancement. When the edge response value of the enhanced edge segment in the enhanced body contour image is higher than the edge response value before enhancement, and the positional change of the enhanced edge segment relative to the fixed background region in two or more frames of body surface images before enhancement is lower than a preset change, it indicates that the enhancement process has increased the response intensity of the edge segment, but the edge segment has not undergone effective displacement with the human body surface in the original image sequence. The system determines the enhanced edge segment as a false background edge generated by enhancement and deletes it from the enhanced body contour image or reduces its edge response value.

[0059] During the verification of original evidence, the system also determines whether the enhanced edge segment alters the connectivity of the human body region defined by the body surface contour segment. The system determines the connectivity of the human body region based on the boundary formed by the body surface contour segment and determines whether the addition of the enhanced edge segment divides the same human body region into two unconnected regions. When the enhanced edge segment divides the same human body region into two unconnected regions, and there are no corresponding candidate contour segments that meet the preset matching conditions in two or more frames of the body surface image before enhancement, it indicates that the enhanced edge segment lacks original contour basis and causes incorrect segmentation of the human body region's topology. The system then removes the enhanced edge segment from the enhanced body surface contour image.

[0060] Through the above processing, the system can identify credible body surface contour segments and suppress non-body surface contour segments even when there are interferences such as chamber edges, positioning supports, or local textures in the fixed background area of ​​the treatment chamber. This is achieved by utilizing the displacement of candidate contour segments relative to the fixed background area, the motion consistency with suspected human image areas, the relative motion differences between the two image areas, and the original evidence verification results of the enhanced edge segments. The system updates the human region boundary based on the enhanced body surface contour image after removing false background edges, which is used for subsequent candidate treatment site extraction and effective site identification.

[0061] Example 2:

[0062] In another specific embodiment, after obtaining an enhanced body contour image, Company A's treatment chamber image processing system extracts candidate treatment sites within the human body region defined by the enhanced body contour image. Candidate treatment sites can be determined based on a preset treatment site template, the calibration relationship between the treatment chamber coordinate system and the image coordinate system, local image features within the human body region, or externally input information about the location to be treated. The system maps the candidate treatment sites to pixel coordinates in the enhanced body contour image and determines a preset neighborhood centered on the candidate treatment site.

[0063] The preset neighborhood can be determined based on the treatment site size, image resolution, distribution density of body surface contour segments, and allowable positioning error. In one example, the resolution of the enhanced body surface contour image is 1280×720, the x-coordinate of the candidate treatment site P is 642 pixels, and the y-coordinate is 358 pixels. The system determines a preset neighborhood with a radius of 80 pixels centered on the candidate treatment site P, and searches for continuous edge segments that have been identified as body surface contour segments within this preset neighborhood. When the endpoint distance of adjacent body surface contour segments meets a preset connection condition, and the change in extension direction meets a preset smoothing condition, the system combines them to form a contour support path. Figure 5The relative positional relationship of candidate treatment sites, preset neighborhood, contour support path A and contour support path B in image coordinates is shown. The preset neighborhood covers the local area around the candidate treatment site that can be used to find spatial support contours. Path A is located near the candidate treatment site. Although path B is near the neighborhood, it is relatively far from the candidate treatment site.

[0064] The system calculates the distance between candidate treatment sites and each contour support path, and selects contour support paths with a distance less than a preset distance threshold as candidate support paths. This distance can be determined by the following formula:

[0065]

[0066] in, Indicates candidate therapeutic sites P to the 1st The outline supports the shortest distance of the path. Indicates the first The outline supports the path. Indicates the first The first contour support path One sampling point, and Let x and y represent the x and y coordinates of the candidate treatment site P in the image, respectively. and These represent the sampling points. The horizontal and vertical coordinates in the image, This indicates that the minimum distance is selected among all sampling points in the contour support path. The system first obtains multiple sampling points on the contour support path, and then calculates the planar pixel distance between the candidate treatment site and each sampling point, thereby determining the nearest spatial support relationship between the candidate treatment site and the contour support path.

[0067] In one example, the candidate treatment site P has an x-coordinate of 642 pixels and a y-coordinate of 358 pixels, with a preset neighborhood radius of 80 pixels. The nearest sampled point Q1 on contour support path A has an x-coordinate of 610 pixels and a y-coordinate of 346 pixels, and the nearest sampled point Q2 on contour support path B has an x-coordinate of 700 pixels and a y-coordinate of 390 pixels. The preset distance threshold is 55 pixels. Substituting these data into the formula, the distance from candidate treatment site P to contour support path A is 34.18 pixels, and the distance from candidate treatment site P to contour support path B is 66.24 pixels. Since 34.18 pixels is less than the preset distance threshold of 55 pixels, and 66.24 pixels is greater than the preset distance threshold of 55 pixels, the system selects contour support path A as the candidate support path and excludes contour support path B. Figure 5The above filtering process is presented through connecting lines and numerical labels, so that the relationship between the selection criteria of candidate support paths and the distance threshold can correspond to the implementation steps.

[0068] The system generates a connection region along the direction from the candidate treatment site to the candidate support path. The connection region can be a strip-shaped area covering the image channel between the candidate treatment site and the contour support path, or it can be a local detection region extending along the line connecting the two. The system detects whether there are non-surface contour segments within the connection region and determines whether these segments constitute a break. If the connection region is not broken by non-surface contour segments, it indicates that there is a continuous spatial support relationship between the candidate treatment site and the reliable human body surface boundary, and the system outputs a valid site identifier.

[0069] When determining whether a connected region is interrupted by non-body surface contour segments, the system calculates the displacement matching degree between the non-body surface contour segments within the connected region and the suspected human image region. This displacement matching degree can be determined by the following formula:

[0070]

[0071] in, Indicates the first [unit] within the connected region The degree of displacement matching between non-body surface contour segments and suspected human body image regions. Indicates the first The displacement of a non-body surface contour segment relative to a fixed background area. This indicates the displacement of a region in a suspected human image relative to a fixed background region. Indicates the first The directional difference between the displacement direction of a non-body surface contour segment and the displacement direction of the suspected human body image region. This indicates the allowable displacement difference reference value. This indicates the allowable directional difference reference value. and This indicates the weights of displacement difference and direction difference in the matching judgment. The system first compares the displacement difference between the non-body surface contour segment and the suspected human body image region, and then compares the displacement direction difference between the two to determine whether the non-body surface contour segment moves with the human body surface.

[0072] In the example above, the displacement matching degree between a non-body surface contour segment within the connected region and the suspected human image region is 0.42, while the preset matching threshold is 0.75. Furthermore, both ends of this non-body surface contour segment are connected to the fixed background region of the treatment chamber. Since 0.42 is less than the preset matching threshold of 0.75, and both ends of the non-body surface contour segment are connected to the fixed background region, it indicates that the non-body surface contour segment has background structure boundary features and forms a continuous barrier between the candidate treatment site P and the contour support path A. The system determines that the candidate treatment site P and the contour support path A are separated and outputs an invalid site identifier. Figure 6 The positional relationship between the connecting region, the fixed background region, the support path A, and the non-body surface contour segment is shown. The non-body surface contour segment passes through the connecting region and connects to the upper and lower fixed background regions respectively. Combined with the judgment result that the displacement matching degree is 0.42, which is lower than the matching threshold of 0.75, it can be shown that the non-body surface contour segment is not a reliable boundary that moves with the human body surface, but rather a background structure boundary that blocks the candidate treatment site from the body surface contour support path.

[0073] When multiple contour support paths exist within a preset neighborhood, the system calculates the distance between each contour support path and the candidate treatment site, and determines whether the connection region between each contour support path and the candidate treatment site crosses a non-body surface contour segment. The system selects the contour support path whose distance to the candidate treatment site is less than a preset distance threshold and whose connection region does not cross a non-body surface contour segment as the target support path. If multiple contour support paths meet the above conditions, the system may select the contour support path with a smaller distance, higher continuity, or higher matching degree with the displacement of the suspected human image region as the target support path.

[0074] After determining the target support path, the system detects the displacement directions of the corresponding body surface contour segments at both ends of the target support path in adjacent frames. If the displacement directions of the corresponding body surface contour segments at both ends of the target support path match in adjacent frames, it indicates that the target support path consists of continuous contours near the same human body surface region, which can provide reliable support for candidate treatment sites, and the system outputs a valid site identifier. If the displacement directions of the corresponding body surface contour segments at both ends of the target support path do not match, it indicates that the path may be formed by splicing contour segments from different regions, and the system does not use it as a valid spatial support basis.

[0075] After determining that the candidate treatment site is isolated from the contour support path, the system searches for bypass contour segments along both sides of the connecting region. A bypass contour segment is a contour segment located laterally within the connecting region and whose displacement matching degree with the suspected human image region meets preset matching conditions. The system determines whether the bypass contour segment can bypass the non-surface contour segments within the connecting region and form a connection with the contour support path. If a bypass contour segment exists on either side of the connecting region's boundary, and this bypass contour segment is connected to the contour support path, it indicates that although the candidate treatment site is isolated by non-surface contour segments in the direct connection direction, it may still obtain spatial support through the lateral human contour. The system then adjusts the ineffective site identifier of the candidate treatment site to a site identifier to be reviewed.

[0076] In a further example, after the direct connection region from candidate treatment site P to contour support path A is interrupted by a non-surface contour segment, the system retrieves a bypass contour segment at the left boundary of the connection region. The displacement matching degree between this bypass contour segment and the suspected human image region is 0.81, which is greater than the preset matching threshold of 0.75, and this bypass contour segment is connected to contour support path A. At this time, the system does not directly maintain the invalid site identifier, but adjusts the candidate treatment site P to the site identifier to be reviewed, and outputs the coordinates of the candidate treatment site, contour support path A, the non-surface contour segment forming the interruption, and the bypass contour segment together for subsequent manual review or further judgment with additional recognition rules. Figure 7 The diagram illustrates the scenario where, after the direct connection direction is interrupted, the connection with the supporting path A is re-established through a lateral bypass contour segment in the connection region. The matching degree of the bypass contour segment is 0.81, which is higher than the matching threshold of 0.75. This indicates that the bypass contour segment has motion consistency with the human body surface. Therefore, the candidate treatment site is not directly determined as ineffective, but enters a state of pending review.

[0077] In this way, the system can use the distance relationship between candidate treatment sites and contour support paths, the separation relationship of non-body surface contour segments in the connected region, the motion consistency at both ends of the target support path, and the connectivity relationship of bypass contour segments in the enhanced body contour image to output candidate treatment sites as effective, ineffective, or pending verification, thereby reducing the interference of fixed background edges, pseudo edges, or non-human structures in the treatment chamber on the site recognition results.

Claims

1. A method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision, characterized in that, include: Acquire two or more consecutive body surface images of the user while they are in the treatment chamber, extract edges from each body surface image to obtain candidate contour segments and adjacent image regions on both sides of the candidate contour segments; Using the fixed background area inside the treatment chamber as a reference, the displacement amount and displacement direction of the candidate contour segment in adjacent frames are calculated, and the displacement matching degree between the candidate contour segment and the suspected human image area is calculated. When the displacement amount of the candidate contour segment relative to the fixed background area is greater than the first displacement threshold and the displacement matching degree meets the preset matching conditions, it is determined to be a body surface contour segment; otherwise, it is determined to be a non-body surface contour segment. Enhance the body surface contour segments and suppress non-body surface contour segments to generate a body surface contour enhanced image. Extract candidate treatment sites within the human body region defined by the body surface contour enhanced image. Based on the distance relationship between the candidate treatment sites and the body surface contour segments and the exclusion results of non-body surface contour segments, output the effective site identifier or ineffective site identifier of the candidate treatment sites.

2. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 1, characterized in that, A first side image region and a second side image region are obtained on both sides of the candidate contour segment respectively; when the displacement matching degree between the first side image region and the suspected human body image region meets the preset matching condition, and the displacement of the second side image region relative to the fixed background region is less than the second displacement threshold, the candidate contour segment is determined to be the body surface contour segment. When the displacement of both the first side image region and the second side image region relative to the fixed background region is less than the second displacement threshold, the candidate contour segment is determined to be the non-body surface contour segment.

3. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 1, characterized in that, The newly appearing enhanced edge segments in the enhanced body contour image are verified using original evidence. When the enhanced edge segment does not have a corresponding candidate contour segment that meets the preset matching conditions in two or more frames of body surface images before enhancement, or when the displacement relative to the fixed background area is less than the second displacement threshold, the enhanced edge segment is marked as the non-body surface contour segment and removed from the enhanced body contour image.

4. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 1, characterized in that, A preset neighborhood is determined centered on the candidate treatment site, and a contour support path formed by continuous connection of the body surface contour segments is determined within the preset neighborhood; when the connection area between the candidate treatment site and the contour support path is not interrupted by the non-body surface contour segments, the valid site identifier is output; when the connection area is interrupted by the non-body surface contour segments, the invalid site identifier is output.

5. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 3, characterized in that, Gray-scale sampling sequences are taken on both sides of the normal of the enhanced edge segment; when the gradient direction of the gray-scale sampling sequence does not match the gray-scale change order on both sides of the corresponding candidate contour segment in two or more frames of body surface images before enhancement, the enhanced edge segment is marked as the non-body surface contour segment.

6. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 3, characterized in that, The relative positional change between the enhanced edge segment and the fixed background region is compared; when the edge response value of the enhanced edge segment in the enhanced body surface contour image is higher than the edge response value before enhancement, and the positional change of the enhanced edge segment relative to the fixed background region in two or more frames of body surface images before enhancement is lower than a preset change, the enhanced edge segment is determined to be a false background edge generated by enhancement.

7. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 3, characterized in that, Determine whether the enhanced edge segment changes the connectivity of the human body region defined by the body surface contour segment; when the enhanced edge segment divides the same human body region into two unconnected regions, and there are no corresponding candidate contour segments that meet the preset matching conditions in the two or more frames of body surface images before enhancement, remove the enhanced edge segment from the enhanced body surface contour image.

8. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 4, characterized in that, When determining whether the candidate treatment site and the contour support path are separated by the non-body surface contour segment, a connection region is generated along the direction from the candidate treatment site to the contour support path; when the displacement matching degree between the non-body surface contour segment in the connection region and the suspected human image region does not meet the preset matching condition, and both ends of the non-body surface contour segment are respectively connected to the fixed background region, it is determined that the candidate treatment site and the contour support path are separated.

9. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 4, characterized in that, A contour support path that is less than a preset distance threshold from the candidate treatment site and does not cross the non-body surface contour segment is selected as the target support path; when the displacement directions of the body surface contour segments corresponding to both ends of the target support path match in adjacent frames, the effective site identifier is output.

10. The method for enhancing and identifying the surface contour image of a treatment chamber based on computer vision according to claim 8, characterized in that, Search along both sides of the connection area for bypass contour segments that meet the preset matching conditions for displacement matching with the suspected human image area; when there is a bypass contour segment on either side of the two sides and the bypass contour segment is connected to the contour support path, adjust the non-effective site identifier of the candidate treatment site to the site identifier to be reviewed.

Images