CN-121982024-A - Household oral health monitoring system and method based on image recognition

CN121982024ACN 121982024 ACN121982024 ACN 121982024ACN-121982024-A

Abstract

The invention discloses a home oral health monitoring system and method based on image recognition, which are suitable for monitoring adjacent surface ore removal of target adjacent surface contact point neighborhood by user handheld equipment under a light supplementing condition under a home non-diagnosis and treatment environment. Aiming at the problems that deep reflection strong bright spots are easy to form in a tooth gap through area and overlap with suspected ore removal bright spots of an adjacent surface shallow layer to induce false focus, the risk score is unstable and fails to lack a re-acquisition basis, a blind highlight area index sequence is constructed in the tooth gap through area, step blanking indexes are calculated, a measuring window in-point and a measuring window out-point are determined by combining single-side shielding shadow boundary propulsion and boundary quality threshold, ore removal candidate responses are extracted in the window only, a normalized difference result is generated based on posture baseline calibration, the suspected ore removal area of the adjacent surface and ore removal risk score are output, a degradation mark and re-acquisition guide are output when the window is not closed or the window is too short, and confidence degree punishment is applied to the risk score.

Inventors

DONG XIAOMENG
WANG SHAOWEI
ZHAO LICHAO
JIA QINRONG

Assignees

北京亿家老小科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260403

Claims (10)

1. The home oral health monitoring method based on image recognition is characterized by comprising the following steps of: s101, acquiring an oral multi-frame image sequence, positioning a target adjacent surface contact point neighborhood, determining a tooth gap through region for statistical highlighting in the contact point neighborhood, taking the edge of an adjacent dental crown as a single-side shading screen to extract a single-side shielding shadow boundary, calculating a blinding highlight area index sequence in the tooth gap through region, and calculating a step blanking index based on the blinding highlight area index sequence of the adjacent frame; S102, determining a skip interval by a blinding highlight area index sequence, and searching forward frame by frame along a multi-frame image sequence from the frame next to the frame at the tail of the skip interval to generate an in-point and an out-point of a measurement window; in the forward searching process, judging whether the candidate frames meet an event condition frame by frame, wherein the event condition is that a step blanking index crosses a preset step threshold value and a single-side shielding shadow boundary advances to a tooth gap penetrating direction relative to the previous frame; when a candidate frame which meets the event condition for the first time appears, further checking whether all the blind highlight area index sequences in a continuous preset number of frames from the candidate frame are not higher than an entering safety margin, and determining the candidate frame as an in-point when the blind highlight area index sequences meet the entering safety margin; And S103, extracting a demineralization candidate response in a contact point neighborhood only in a time interval defined by the access point and the access point, obtaining a gesture baseline based on cross-frame displacement of a single-side shielding shadow boundary in the time interval, performing calibration on the demineralization candidate response based on the gesture baseline, generating a normalized differential result, delineating an adjacent surface demineralization suspected region according to the normalized differential result, calculating a demineralization risk score, and outputting the adjacent surface demineralization suspected region and the demineralization risk score.
2. The method of claim 1, wherein locating the target adjacent contact point neighborhood comprises performing a segmentation of a crown region of adjacent dental pairs on at least one frame in the sequence of multi-frame images of the mouth to obtain two side crown edges, determining a contact point based on a minimum distance point pair of the two side crown edges, determining a contact point neighborhood of a preset scale centered on the contact point, and determining a through area of the dental gap in the contact point neighborhood, wherein the determining of the through area of the dental gap comprises determining a set of gap pixels between the two side crown edges as a base gap area, and extending to a preset length in a direction from the crown edge toward the interior of the dental gap.
3. The method of claim 2, wherein extracting the single-sided shadow mask boundary comprises constructing a luminance profile within the contact point neighborhood in a direction from the single-sided shading screen toward the interproximal areas, and determining a location of a maximum gradient of luminance from dark to light as the single-sided shadow mask boundary.
4. The method according to claim 2, wherein calculating the sequence of blinding highlight area indicators comprises counting the number of pixels with luminance not lower than a saturation threshold in the interdental penetration area, and dividing the total number of pixels in the interdental penetration area by the total number of pixels in the interdental penetration area to obtain the blinding highlight area indicator per frame, thereby forming the sequence of blinding highlight area indicators, wherein the saturation threshold is determined by a saturation value output by the imaging device or by a high percentile of an image luminance histogram.
5. The method according to claim 4, wherein the calculating of the step blanking index includes calculating a decrease amount of the blinding highlight area index of two adjacent frames, characterizing the step blanking index by a ratio of the decrease amount to the blinding highlight area index of the previous frame, using a preset ratio threshold as a step threshold, and performing sliding window mean processing or sliding window median processing on the blinding highlight area index sequence before calculating the step blanking index, wherein the step blanking index crosses the step threshold, which means that the step blanking index is not less than the step threshold.
6. The method of claim 1, wherein determining a skip interval from a sequence of blinding highlighting area indicators comprises determining the continuous frame segment as a skip interval when none of a predetermined number of consecutive frames of the sequence of blinding highlighting area indicators is less than a blinding threshold, wherein a last frame of the continuous frame segment is a last frame of the continuous frame segment, wherein a temporally last skip interval is selected when there are a plurality of skip intervals, and wherein a frame-by-frame forward search is performed along the sequence of multi-frame images from a frame subsequent to the last frame of the skip interval, wherein the predetermined number of consecutive frames is determined from a frame rate of the sequence of multi-frame images, and wherein the ingress safety margin is an upper limit of the blinding threshold minus a predetermined margin value.
7. The method according to claim 6, wherein the step of determining that the single-side shadow mask boundary is advanced in the interdental space penetration direction with respect to the previous frame by using the direction in which the interdental space is penetrated by using the crown edge as the direction in which the interdental space is penetrated by using the interdental space, comprises extracting the single-side shadow mask boundary from the previous frame in the forward search process, calculating a displacement component of the single-side shadow mask boundary of the adjacent frame in the interdental space penetration direction, determining that the single-side shadow mask boundary is advanced in the interdental space penetration direction with respect to the previous frame when the displacement component is positive, determining that a boundary quality threshold comprises a continuous boundary segment length of the single-side shadow mask boundary not smaller than a length threshold, an extraction confidence of the single-side shadow mask boundary not lower than a confidence threshold, a brightness gradient strength corresponding to the single-side shadow mask boundary not lower than a gradient threshold, and determining that the single-side shadow mask boundary does not satisfy a boundary quality threshold when any condition is not satisfied.
8. The method for home oral health monitoring based on image recognition according to claim 1, wherein extracting the ore removal candidate response comprises, for a target pixel in a contact point neighborhood of a target adjacent surface of each frame within a time interval defined by an in-point and an out-point, taking a brightness average value of a same-frame non-highlight mask pixel set as a background brightness reference, taking a saturation average value of the same-frame non-highlight mask pixel set as a background saturation reference, respectively calculating a brightness increment of the target pixel relative to the background brightness reference and a saturation decrement of the target pixel relative to the background saturation reference, adding a positive part of the brightness increment and a positive part of the saturation decrement to obtain an white spot enhancement response, and taking the white spot enhancement response as the ore removal candidate response, wherein the highlight mask is a pixel set with brightness not lower than a saturation threshold, setting the ore removal candidate response of the corresponding pixel to zero when performing the elimination on the highlight mask coverage pixel, and multiplying the ore removal candidate response of the corresponding pixel by a preset coefficient not greater than 1 when performing the weight reduction on the highlight mask coverage pixel.
9. The method for home oral health monitoring based on image recognition according to claim 8, wherein alignment and multi-frame fusion are performed on the ore removal candidate responses within a time interval defined by an in-point and an out-point, the alignment is performed on the basis of displacement of a single-side shielding shadow boundary in an adjacent frame, the multi-frame fusion is performed to obtain a median value or an average value, a posture baseline is obtained on the basis of cross-frame displacement of the single-side shielding shadow boundary in the time interval so as to calibrate the fused ore removal candidate responses to generate a normalized differential result, an adjacent ore removal suspected region is obtained by aggregation of connected domains meeting a normalized differential threshold in the normalized differential result, the ore removal risk score is determined on the basis of a weighted combination of the proportion of the area of the adjacent ore removal suspected region to the neighborhood area of the contact point and the normalized differential value average value in the adjacent ore removal suspected region, and confidence penalty is applied to the ore removal risk score when the length of the time interval is smaller than a preset minimum window length, and the heavy mining guide information in a measurement window result set is output or updated.
10. The home oral health monitoring system based on image recognition is used for realizing the home oral health monitoring method based on image recognition according to any one of claims 1-9, and is characterized by comprising a multi-frame input and frame level criterion generating unit, a measuring window searching and window result set generating unit, an in-window ore removal evaluation output unit and a preset parameter storage area, wherein the multi-frame input and frame level criterion generating unit, the measuring window searching and window result set generating unit and the in-window ore removal evaluation output unit are respectively used for executing S101, S102 and S103, and the preset parameter storage area is used for storing and calling saturation threshold values, blinding threshold values, continuous preset number frames, step threshold values, preset margin values, length threshold values, confidence threshold values, gradient threshold values, normalized difference threshold values, preset minimum window lengths and risk scoring weight parameters.

Description

Household oral health monitoring system and method based on image recognition Technical Field The invention relates to the technical field of oral health monitoring and image recognition, in particular to a home oral health monitoring system and method based on image recognition. Background Oral health monitoring and image recognition techniques are commonly used for home oral health monitoring. In such applications, the neighboring face demineralization monitoring belongs to a class of scenes requiring important attention, and a user can acquire an oral image under the light supplementing condition through a handheld device so as to obtain a neighboring face demineralization suspected region and a demineralization risk score thereof for subsequent follow-up comparison. The prior proposal generally extracts suspected white spots or demineralized enhancement response from single-frame or multi-frame images, and marks a suspected area and outputs a score by combining a threshold value or a classifier, and for multi-frame acquisition, the common practice is to execute cross-frame alignment or simple fusion or perform multi-frame optimization according to brightness and stability so as to select a measurement frame segment. The prior art has the following defects: On the one hand, when the neighborhood of the contact point of the adjacent surface is shot, the light supplementing light rays easily penetrate through the dental clearance through region to irradiate saliva or soft tissues at a deeper part and form strong reverse bright spots, and the strong reverse bright spots are often overlapped with the suspected descaled bright spots of the shallow layer of the adjacent surface on a two-dimensional image to induce false focus, and the suspected bright spots are shown to be plausible in position but not positioned on the target dental surface. In the existing scheme, if single frame judgment is mainly relied on or the brightness and stability are selected, strong reflection light is easily carried into a measurement frame section by mistake, so that suspected region delineation and risk scoring are unstable. The difficulty is that it is difficult to form a verifiable basis to distinguish deep reflection from shallow suspected demineralization appearance only by two-dimensional brightness and stability. On the other hand, in the process of hand-held angle sweeping, the edge of the adjacent dental crown gradually shields the through area of the dental gap, deep reflection can obviously fade in a short time when being shielded and cut off, and the suspected descaled bright spots of the adjacent surface shallow layer usually do not show the same rapid fading characteristic. If the existing scheme still stays at the brightness level or stability screening, the shielding advancing process is difficult to be converted into a measurement frame section start-stop judging basis, the false focus is easy to be brought into a measurement window, and an explicit re-acquisition triggering basis and an executable re-acquisition guide are absent when the window cannot be reliably determined. The difficulty is that the gesture change caused by the hand-held sweeping angle can cause reflection drift and the adjacent surface texture is weak, so that the cross-frame alignment or fusion is more easily disturbed, and the instability of frame segment selection and output is further amplified. Disclosure of Invention In order to overcome the above-mentioned drawbacks of the prior art, embodiments of the present invention provide a home oral health monitoring system and method based on image recognition, so as to solve the problems set forth in the above-mentioned background art. In order to achieve the above purpose, the present invention provides the following technical solutions: a home oral health monitoring method based on image recognition comprises the following steps: s101, acquiring an oral multi-frame image sequence, positioning a target adjacent surface contact point neighborhood, determining a tooth gap through region for statistical highlighting in the contact point neighborhood, taking the edge of an adjacent dental crown as a single-side shading screen to extract a single-side shielding shadow boundary, calculating a blinding highlight area index sequence in the tooth gap through region, and calculating a step blanking index based on the blinding highlight area index sequence of the adjacent frame; S102, determining a skip interval by a blinding highlight area index sequence, and searching forward frame by frame along a multi-frame image sequence from the frame next to the frame at the tail of the skip interval to generate an in-point and an out-point of a measurement window; in the forward searching process, judging whether the candidate frames meet an event condition frame by frame, wherein the event condition is that a step blanking index crosses a preset step threshold value and a single-side shielding shadow boundary