Gait sequence evaluation method and system based on quality dimension

Abstract

The application discloses a gait sequence evaluation method and system based on a quality dimension, and the method comprises the following steps: acquiring a single gait sequence; acquiring a first characteristic parameter of the single gait sequence; evaluating the quality of the single gait sequence according to the first characteristic parameter to obtain a quality score of the single gait sequence; acquiring a multi-segment gait sequence; acquiring a second characteristic parameter of the multi-segment gait sequence; evaluating the quality of the multi-segment gait sequence according to the quality score of the single gait sequence and the second characteristic parameter to obtain a quality score of the multi-segment gait sequence; comparing the quality score of the single gait sequence or the quality score of the multi-segment gait sequence with a corresponding preset score threshold; and storing the single gait sequence or the multi-segment gait sequence in a gait database according to a comparison result. The application can quickly and accurately acquire the quality of a gait sample, facilitate hierarchical storage, and use gait samples with different qualities for different application scenarios.

Description

TECHNICAL FIELD

[0001] The application discloses a gait sequence evaluation method and system based on a quality dimension, and belongs to the technical field of gait recognition. BACKGROUND

[0002] Gait is a biological and behavioral characteristic of a natural person, and reflects the change law of the body shape in space and time. Since gait information has characteristics such as high anti-counterfeiting, cross-wearing and cross-view angle, the identity of a person can be recognized by using gait, and the accuracy is relatively high.

[0003] In the process of recognizing the identity of a person by using gait, the collected gait sample needs to be compared with a pre-stored gait sample in a gait database. Therefore, the quality of the pre-stored gait sample in the gait database will directly affect the accuracy of the recognition result.

[0004] The prior art mainly improves the accuracy of the gait recognition result by increasing the samples of a person in various external environments in the gait database. However, this method will result in a large amount of data in the gait database, and some of the data quality is poor, which causes low accuracy and slow recognition speed of the gait recognition result. SUMMARY

[0005] The purpose of the present application is to provide a gait sequence evaluation method and system based on a quality dimension, so as to solve the technical problems of low accuracy and slow recognition speed of the gait recognition result caused by a large number of pre-stored gait samples and poor quality of the gait samples in the gait database.

[0006] The first aspect of the present application provides a gait sequence evaluation method based on a quality dimension, comprising:

[0007] obtaining a single gait sequence;

[0008] obtaining a first characteristic parameter of the single gait sequence;

[0009] evaluating the quality of the single gait sequence according to the first characteristic parameter to obtain a quality score of the single gait sequence.

[0010] Preferably, the first characteristic parameter comprises at least one of gait cycle quality, clarity of human shape contour, lighting intensity and viewing angle.

[0011] Preferably, when the first characteristic parameter is gait cycle quality, obtaining the first characteristic parameter of the single gait sequence specifically comprises:

[0012] obtaining a gait silhouette sequence corresponding to the single gait sequence;

[0013] determining the gait cycle quality of the gait silhouette sequence, specifically comprising:

[0014] calculating the coincidence degree between the gait silhouette images in the gait silhouette sequence;

[0015] counting the number of the coincidence degrees greater than a preset coincidence degree threshold, and determining the gait cycle quality of the gait silhouette sequence according to the number.

[0016] Preferably, the calculating the coincidence degree between the gait silhouette images in the gait silhouette sequence specifically comprises:

[0017] acquiring a plurality of gait silhouette images in a set frame number interval in the gait silhouette sequence;

[0018] calculating the coincidence degree between each two adjacent gait silhouette images in the plurality of gait silhouette images, specifically comprising:

[0019] sequentially acquiring each two adjacent gait silhouette images from the plurality of gait silhouette images;

[0020] calculating the coincidence degree between each two adjacent gait silhouette images according to the pixel values at the same pixel positions of the two adjacent gait silhouette images.

[0021] Preferably, when the first feature parameter is the definition of the human contour, the acquiring the first feature parameter of the single gait sequence specifically comprises:

[0022] converting each gait image in the single gait sequence into a gray-scale image;

[0023] filtering the gray-scale image to obtain a filtering sequence;

[0024] calculating the variance of each filtering image in the filtering sequence, and determining the definition of the human contour according to the variance.

[0025] Preferably, after acquiring the single gait sequence, the method further comprises:

[0026] judging whether the single gait sequence meets a preset pre-evaluation condition; if yes, acquiring the first feature parameter of the single gait sequence;

[0027] if not, deleting the single gait sequence;

[0028] wherein, the pre-evaluation condition comprises a complete human contour, a natural walking state and an image quality.

[0029] the image quality comprises at least one of resolution, size and frame number of the gait image.

[0030] Correspondingly, the judging whether the single gait sequence meets a preset pre-evaluation condition specifically comprises:

[0031] determining whether a complete human figure is contained in the gait image in the single gait sequence;

[0032] determining whether the pedestrian in the single gait sequence is in a natural walking state;

[0033] determining whether the image quality of the gait image in the single gait sequence is greater than a preset quality threshold.

[0034] Preferably, determining whether the pedestrian in the single gait sequence is in a natural walking state specifically comprises:

[0035] respectively acquiring the mean value, standard deviation and covariance of adjacent two gait images in the single gait sequence;

[0036] According to the mean value, standard deviation and covariance, the first similarity of adjacent two gait images is calculated, and a plurality of first similarities are obtained.

[0037] The mean value of the plurality of first similarities is calculated, and it is determined whether the mean value is greater than a preset similarity threshold. If yes, the pedestrian in the single gait sequence is in a stationary state.

[0038] Preferably, it further comprises:

[0039] obtaining a plurality of gait sequences, the plurality of gait sequences comprising a plurality of single gait sequences;

[0040] obtaining a second feature parameter of the plurality of gait sequences;

[0041] According to the quality score of the single gait sequence and the second feature parameter, the quality of the plurality of gait sequences is evaluated, and a quality score of the plurality of gait sequences is obtained.

[0042] The second feature parameter comprises sequence number and view angle richness.

[0043] Preferably, it further comprises:

[0044] The quality score of the single gait sequence or the quality score of the plurality of gait sequences is compared with a corresponding preset score threshold, and the single gait sequence or the plurality of gait sequences is stored in a gait library according to the comparison result.

[0045] The second aspect of the present application provides a quality-dimension-based gait sequence evaluation system using the above-mentioned quality-dimension-based gait sequence evaluation method, comprising:

[0046] The sequence acquisition module is used to obtain a single gait sequence.

[0047] The feature acquisition module is used to obtain a first feature parameter of the single gait sequence.

[0048] an evaluation module configured to evaluate the quality of the single-stage gait sequence according to the first feature parameter, and obtain a quality score of the single-stage gait sequence.

[0049] Compared with the prior art, the gait sequence evaluation method and system based on the quality dimension have the following beneficial effects:

[0050] The method of the present application includes a single-stage gait sequence evaluation method and a multi-stage gait sequence evaluation method combined with the single-stage gait sequence evaluation method. The evaluation method of the present application can quickly and accurately obtain the quality of the gait sequence, facilitate hierarchical storage, and use gait sequences of different qualities for different application scenarios. For example, gait sequences with a quality greater than or equal to a preset quality threshold can be used for gait recognition, and gait sequences with a quality less than the preset quality threshold can be used for gait retrieval. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A flowchart of the gait sequence evaluation method based on the quality dimension provided for an embodiment of the present application is shown in the figure;

[0052] Figure 2 A gait silhouette sequence graph that is not normalized in the embodiment of the present application is shown in the figure;

[0053] Figure 3 A gait silhouette sequence graph that is normalized in the embodiment of the present application is shown in the figure;

[0054] Figure 4 A flowchart of the gait sequence evaluation method based on the quality dimension provided for another embodiment of the present application is shown in the figure;

[0055] Figure 5 A structure diagram of the gait sequence evaluation system based on the quality dimension provided for the embodiment of the present application is shown in the figure.

[0056] In the figure, 101 is a sequence acquisition module; 102 is a feature acquisition module; and 103 is an evaluation module. DETAILED DESCRIPTION

[0057] In the following description, specific details are set forth in order to provide a thorough understanding of embodiments of the application. However, persons having ordinary skill in the art will appreciate that embodiments of the application can be practiced without the specific details, and that the present application is not limited to the specific details or the particular order described herein. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0058] Before the technical solutions of the present application are described in detail, the terms used in the present application are first described.

[0059] Gait: the biological and behavioral characteristics of a natural person, reflecting the change rule of body shape in space-time.

[0060] Gait sample: the gait video or gait image sequence (referred to as gait sequence) of a natural person obtained through collection, pretreatment and the like. In the present application, the sample refers to the original (cut) video or continuous image sequence containing the gait cycle of a natural person, containing clothing, hat and the like.

[0061] Gait silhouette sequence: the solid shape representation of the human body image presented by the gait sample after removing the background useless information.

[0062] Gait cycle: the walking process of two feet respectively from the heel off the ground to the heel landing when walking continuously.

[0063] Gait recognition: automatic recognition of a natural person based on the biological and behavioral characteristics contained in the gait. The gait recognition can be used for identity recognition, and can also be used for non-identity recognition scenes such as behavior analysis, posture analysis or abnormal analysis.

[0064] Gait database: a database storing gait data, which can be classified according to the quality score.

[0065] The technical solutions of the present application will be described in detail below.

[0066] The first aspect of the embodiment of the present application provides a gait sequence evaluation method based on quality dimension, including a single-segment gait sequence evaluation method and a multi-segment gait sequence evaluation method combined with the single-segment gait sequence evaluation method.

[0067] The flow chart of the gait sequence evaluation method based on quality dimension involving the single-segment gait sequence evaluation method in the embodiment of the present application is shown in Figure 1 , including:

[0068] Step S1, acquiring a single-segment gait sequence.

[0069] The single-segment gait sequence in the embodiment of the present application is composed of a plurality of continuous single-frame gait images.

[0070] The gait sequence in the embodiment of the present application requires human shape features. For example, if non-human input data is sent into the segmentation model, the quality of human input data is too poor, the segmentation model accuracy is insufficient, and the human segmentation result does not have human shape features due to the above reasons, it is considered unqualified and needs to be re-collected.

[0071] The gait sequence only retains the walking human shape. For example, the non-walking human shape such as squatting, riding a bike, lying horizontally and the like is considered unqualified and needs to be re-collected.

[0072] The gait sequence requires human integrity, such as the presence of upper body loss, lower body loss, left hemiplegia, right hemiplegia, and the like, and more than 1 / 3 of the loss is considered unqualified, and needs to be re-collected.

[0073] To ensure that the obtained single gait sequence meets the subsequent evaluation requirements, the embodiment of the application further comprises, after obtaining the single gait sequence:

[0074] determining whether the single gait sequence meets the preset pre-evaluation condition;

[0075] If yes, go to step S2;

[0076] If no, delete the single gait sequence.

[0077] The purpose of the embodiment of the application for determining whether the single gait sequence meets the preset pre-evaluation condition is to control the collection quality from the data collection side, so as to avoid the influence of the collected data on the subsequent gait recognition or gait retrieval.

[0078] The pre-evaluation condition in the embodiment of the application includes complete human form, natural walking state and image quality. The reason for using the above three pre-evaluation conditions is that the gait image collected by the camera on the collection side may have defects, and the causes of the defects include three categories of subject features, subject behaviors and collection processes. Among them, the defects caused by the subject features include upper body loss, lower body loss, left hemiplegia, right hemiplegia, backpack, crutch, clothes color consistent with background color and shielding of other people / objects, etc.; the defects caused by the subject behaviors include stillness, cycling, running, high leg lifting, nose picking, playing mobile phone, carrying instruments, etc.; the defects caused by the collection process include complex scene background, extremely strong or weak lighting, unfocusing (low definition, motion blur caused by fast motion, etc.) and small image size or resolution.

[0079] Since the number and severity of defects will affect the performance of the gait recognition and gait retrieval system, the embodiment of the application selects the above-mentioned indicators of the causes of the defects, such as the subject features, the subject behaviors and the collection process, to judge the quality of the single gait sequence after obtaining the single gait sequence, and when the quality of the collected single gait sequence does not meet the above-mentioned pre-evaluation condition, it is deleted and not used.

[0080] In the embodiment of the application, whether the single gait sequence meets the preset pre-evaluation condition is determined, specifically:

[0081] determining whether the gait image in the single gait sequence contains a complete human form;

[0082] determining whether the pedestrian in the single gait sequence is in a natural walking state;

[0083] Determine whether the image quality of the gait image in the single gait sequence is greater than a preset threshold. The quality of the gait image includes at least one of the resolution of the gait image, the size of the gait image, and the frame number of the gait image. That is, it is necessary to determine whether the resolution of the gait image is greater than a preset resolution threshold, whether the size of the gait image is greater than a preset size threshold, and whether the frame number of the gait image is greater than a preset frame number threshold. Only when the corresponding threshold is exceeded can the gait image be entered into the gait database. The image resolution is nominally marked by the number of image rows and columns of pixels. The human height and width in pixels can be used to measure the pixel range relative to the human features. The frame number is expressed by counting. A too small value affects the quality of the gait image. When the single gait sequence has a complete human form, the pedestrian is in a natural walking state, and the quality of the gait image is greater than the preset threshold, it is considered that the single gait sequence meets the preset pre-evaluation condition.

[0084] Further, the embodiment of the present application determines whether the pedestrian in the single gait sequence is in a natural walking state, specifically comprising:

[0085] The mean, standard deviation, and covariance of the adjacent two gait images in the single gait sequence are obtained respectively.

[0086] According to the mean, standard deviation, and covariance, the first similarity of the adjacent two gait images is calculated, and a plurality of first similarities are obtained.

[0087] The mean of the plurality of first similarities is calculated, and it is determined whether the mean is greater than a preset similarity threshold. If yes, the pedestrian in the single gait sequence is in a stationary state, and the sequence in a stationary state is unqualified and can be deleted.

[0088] The first similarity of the adjacent two gait images is calculated according to the mean, standard deviation, and covariance, specifically comprising:

[0089] The first similarity of the adjacent two pedestrian images is calculated using formula (1):

[0090]

[0091] In formula (1), SSIM(K, L) is the first similarity of the adjacent two gait images K and gait image L, μ K and σ K are the mean and standard deviation of the gait image K respectively, μ L and σ L are the mean and standard deviation of the gait image L respectively, σ xy is the covariance of the gait image K and the gait image L, c1 and c2 are constants, and the setting of c1 and c2 avoids the instability caused by the denominator approaching 0.

[0092] A plurality of first similarities can be obtained by traversing all gait images in the single gait sequence using the above formula (1).

[0093] The SSIM in the embodiment of the application is an index for measuring the similarity of pictures, and is a number between 0 and 1. The greater the SSIM, the smaller the difference between two images. In the embodiment of the application, the same region of two adjacent gait images is obtained, and then the similarity of the two regions is calculated by using the SSIM (Structural Similarity) algorithm, thereby obtaining a plurality of first similarities.

[0094] In step S2, a first feature parameter of the single gait sequence is obtained.

[0095] The first feature parameter in the embodiment of the application includes, but is not limited to, at least one of gait cycle quality, human contour definition, illumination intensity and viewing angle.

[0096] The illumination intensity is expressed by a quality score, thereby evaluating the strength of the illumination intensity. The analysis object is a histogram corresponding to the un-normalized pixel values of the entire image. When the illumination is normal, the histogram span is larger. When the illumination is too weak or too strong, the distribution of the gray value is concentrated at both ends of the histogram. Figure 1

[0097] Suppose that H0 is the histogram when the illumination is standard, and H is the histogram of the gait image to be evaluated. The quality score can be defined according to the difference between H0 and H in the distribution of the gray value.

[0098] Further, when the first feature parameter is the gait cycle quality, obtaining the first feature parameter of the single gait sequence specifically includes:

[0099] In step A, a gait silhouette sequence corresponding to the single gait sequence is obtained.

[0100] In the embodiment of the application, each gait image in the single gait sequence is segmented to obtain a gait silhouette image corresponding to each gait image, and then the gait silhouette sequence corresponding to the single gait sequence can be further obtained.

[0101] After obtaining the gait silhouette sequence corresponding to the single gait sequence, the following steps can be further included:

[0102] The gait silhouette sequence is normalized and binarized. Specifically, the gait silhouette sequence is normalized, cropped around the human center, and then the normalized gait silhouette sequence is binarized.

[0103] The gait silhouette sequence before normalization is shown in FIG. 1, and the normalized gait silhouette sequence is shown in FIG. 2. Figure 2 Figure 3

[0104] ​​​Step B, determining gait cycle quality of the gait silhouette sequence, specifically comprising:

[0105] Step B1, calculating coincidence degree between gait silhouette images in the gait silhouette sequence, specifically comprising:

[0106] (b1), acquiring multiple gait silhouette images with a set frame number interval in the gait silhouette sequence, wherein the set frame number interval is determined by the video snapshot frame rate.

[0107] (b2), respectively calculating coincidence degree between adjacent two gait silhouette images in the multiple gait silhouette images, specifically comprising:

[0108] sequentially acquiring adjacent two gait silhouette images from the multiple gait silhouette images;

[0109] According to pixel values at the same pixel position of the adjacent two gait silhouette images, calculating the coincidence degree between the adjacent two gait silhouette images, specifically calculating the coincidence degree between the adjacent two gait silhouette images by using formula (2).

[0110]

[0111] In formula (2), C is the coincidence degree, p and q are the adjacent two gait silhouette images, g(m, n, p) is a pixel value of the gait silhouette image p at its corresponding pixel point (m, n), g(m, n, q) is a pixel value of the gait silhouette image q at its corresponding pixel point (m, n), and the coincidence degree of the embodiment is obtained by traversing pixel values at all pixel points of the gait silhouette image.

[0112] Step B2, counting the number of the coincidence degrees greater than the preset coincidence degree threshold, and determining the gait cycle quality of the gait silhouette sequence according to the number.

[0113] The method for determining the gait cycle quality according to the number in the embodiment of the application can determine the gait cycle quality according to a preset mapping relationship between the number and the gait cycle quality, for example, when the number of the coincidence degrees greater than the preset coincidence degree threshold is 1, the gait cycle quality is 0.2, when the number of the coincidence degrees greater than the preset coincidence degree threshold is 2, the gait cycle quality is 0.4, when the number of the coincidence degrees greater than the preset coincidence degree threshold is 3, the gait cycle quality is 0.6, when the number of the coincidence degrees greater than the preset coincidence degree threshold is 4, the gait cycle quality is 0.8, and when the number of the coincidence degrees greater than the preset coincidence degree threshold is 5, the gait cycle quality is 1. The application does not specifically limit the mapping relationship between the number and the gait cycle quality, and the mapping relationship can be set according to actual conditions.

[0114] In the embodiment, the gait with the coincidence degree greater than the preset coincidence degree threshold is considered as a repeatedly appearing gait, the number of the repeatedly appearing gaits and the number of the repeatedly appearing frames are counted, the gait cycle quality is evaluated, and the quality of the gait sequence is further obtained. The high-quality gait sequence needs to ensure the number of the repeatedly appearing gaits. That is, when the number of the repeatedly appearing gaits is greater than a preset number threshold, the gait sequence quality is considered to be better.

[0115] Further, when the first feature parameter is the definition of the human contour, the first feature parameter of the single-segment gait sequence is obtained, and specifically includes:

[0116] Step A, each gait image in the single-segment gait sequence is converted into a gray image.

[0117] Step B, the gray images in the single-segment gait sequence are filtered to obtain a filtering sequence.

[0118] In the embodiment of the application, the Laplace operator is used to filter the gray images in the single-segment gait sequence, and the convolution kernel L of the Laplace operator used is as follows:

[0119]

[0120] Step C, the variance of each filtering image in the filtering sequence is calculated, and the definition of the human contour is determined according to the variance. Specifically, the variance of each filtering image in the filtering sequence is calculated by using formula (3):

[0121]

[0122] In formula (3), LAP_VAR(I) is the variance of the filtering image I, M and N are the width and height of the filtering image respectively, m represents the mth column of the image, n represents the nth row of the image, and L(m, n) represents the result of convolution of the filtering image I(m, n) by the convolution kernel L, is the average absolute value, and the calculation formula is as follows:

[0123]

[0124] After obtaining the variance, the definition of the human contour is determined according to the variance. Specifically, the variance is compared with a preset variance threshold. If the variance is greater than the preset variance threshold, the human contour is clear, and the quality of the gait sequence is high. Otherwise, the quality is low.

[0125] The variance used above can be the variance of a single filtering image or the mean value of the variances of all filtering images in the filtering sequence.

[0126] Step S3, the quality of the single-segment gait sequence is evaluated according to the first feature parameter, and a quality score of the single-segment gait sequence is obtained.

[0127] When the first characteristic parameter is only one of the gait cycle quality, the clarity of the human profile, the lighting intensity, and the viewing angle, the quality score of the single gait sequence is determined according to a pre-established mapping relationship between the first characteristic parameter and the quality score of the single gait sequence.

[0128] When the first characteristic parameter is at least two of the gait cycle quality, the clarity of the human profile, the lighting intensity, and the viewing angle, each first characteristic parameter is regularized, which can be specifically regularized according to a pre-established corresponding relationship between each first characteristic parameter and a value or a value interval in the interval [0, 1]. Then, the multiple regularized first characteristic parameters are weighted and averaged to obtain the quality score of the single gait sequence. The selection of the weight used in the weighted averaging needs to be able to reflect the performance of the sample in the recognition environment.

[0129] The embodiment of the present application also includes a multi-segment gait sequence evaluation method, which combines the quality score of the single gait sequence to make the result more accurate.

[0130] The flow of the embodiment of the present application is shown in Figure 4 as shown, after step S3, it further includes:

[0131] Step S4, obtaining a multi-segment gait sequence, the multi-segment gait sequence including multiple single-segment gait sequences, the premise of the multi-segment gait sequence being multiple single-segment gait sequences of the same target person.

[0132] Step S5, obtaining a second characteristic parameter of the multi-segment gait sequence, wherein the second characteristic parameter includes but is not limited to the sequence number and the viewing angle richness. The sequence number refers to the number of single-segment sequences. The viewing angle richness is the number of different viewing angles in the multi-segment gait sequence, which mainly represents the diversity of the viewing angle.

[0133] In the embodiment of the present application, when the second characteristic parameter is the viewing angle richness, the second characteristic parameter of the multi-segment gait sequence is obtained, which specifically includes:

[0134] Step A, obtaining a walking trajectory of a pedestrian in the multi-segment gait sequence, and determining multiple feature points corresponding to the walking trajectory by using a trajectory segmentation algorithm based on the minimum description length principle;

[0135] Step B, connecting two adjacent feature points using a vector respectively to obtain multiple approximate trajectories of the pedestrian;

[0136] Step C, determining the viewing angle of each approximate trajectory in combination with a pre-determined reference direction vector, which specifically includes: taking the reference direction vector as the starting point, and determining the included angle between the reference direction vector and each trajectory vector in the clockwise direction, which is the viewing angle;

[0137] Step D, count the number of different views, and the number of different views is denoted as the view richness of the gait sequence.

[0138] Step S6, according to the quality score of the single-segment gait sequence and the second characteristic parameter, the quality of the multi-segment gait sequence is evaluated, and the quality score of the multi-segment gait sequence is obtained, and specifically includes:

[0139] According to the quality score of the single-segment gait sequence and the second characteristic parameter, the quality of the multi-segment gait sequence is evaluated by using a weighted evaluation model, and the quality score of the multi-segment gait sequence is obtained.

[0140] The weighted average can be an arithmetic average or an exponential average.

[0141] When it is an arithmetic average, it is specifically:

[0142] QS=α1(σ1M1+σ2M2+…+σ Q M Q )+α2(β1N1+…+β S N S ) (7)

[0143] In formula (7), QS is the quality score of the multi-segment gait sequence, M Q is the quality score of the Qth single-segment gait sequence, σ Q is the weight of the quality score of the Qth single-segment gait sequence, and σ1+σ2+…+σ Q =1; N S is the Sth second characteristic parameter, and N S is the normalized second characteristic parameter, when the second characteristic parameter only has the sequence number and the view richness, then S=2, β S is the weight of the Sth second characteristic parameter, and β1+β2+…+β Q =1; α1 and α2 are weights, and α1+α2=1. The selection of all the weights needs to consider that the subject gait image quality score can reflect the performance of the sample in the identification environment.

[0144] After obtaining the quality score, the embodiment of the application can store and use the gait sequence according to the quality score, and specifically:

[0145] Step S7, the quality score of the single-segment gait sequence or the quality score of the multi-segment gait sequence is compared with a corresponding preset score threshold, and the single-segment gait sequence or the multi-segment gait sequence is stored in the gait base according to the comparison result.

[0146] According to the preset score threshold, the embodiment of the application can divide the gait sequence into two levels, the low level can be used for gait retrieval, and the high level can be used for gait identification.

[0147] The second aspect of the present application provides a quality dimension-based gait sequence evaluation system, which uses the quality dimension-based gait sequence evaluation method described above.

[0148] The structure of the quality dimension-based gait sequence evaluation system of the present application is shown in Figure 5 The system comprises a sequence acquisition module 101, a feature acquisition module 102 and an evaluation module 103.

[0149] The sequence acquisition module 101 is configured to acquire a single-segment gait sequence.

[0150] The feature acquisition module 102 is configured to acquire a first feature parameter of the single-segment gait sequence.

[0151] The evaluation module 103 is configured to evaluate the quality of the single-segment gait sequence according to the first feature parameter, and obtain a quality score of the single-segment gait sequence.

[0152] The present application comprehensively considers the quality scores of the local (single-segment gait sequence in a multi-segment gait sequence) and the whole (multi-segment gait sequence), and the obtained result has high precision.

[0153] The above is only a few embodiments of the present application, and does not limit the present application in any form. Although the above is disclosed in the preferred embodiments, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical contents without departing from the scope of the technical solution of the present application, and such changes or modifications are equivalent to equivalent embodiments, and are within the scope of the technical solution.

Claims

1. A gait sequence evaluation method based on quality dimension, characterized in that: include: Get a single gait sequence; Acquiring a first characteristic parameter of the single-segment gait sequence; evaluating the quality of the single-segment gait sequence according to the first characteristic parameter to obtain a quality score of the single-segment gait sequence; The first characteristic parameter includes the gait cycle quality, and obtaining the gait cycle quality of the single gait sequence specifically includes: Obtaining a gait silhouette sequence corresponding to the single gait sequence; Determining the gait cycle quality of the gait silhouette sequence specifically includes: calculating the degree of overlap between gait silhouette images in the gait silhouette sequence; Counting the number of overlaps greater than a preset overlap threshold, and determining the gait cycle quality of the gait silhouette sequence according to the number; After obtaining a single gait sequence, it also includes: Determining whether the single-segment gait sequence satisfies a preset pre-evaluation condition; if so, obtaining a first characteristic parameter of the single-segment gait sequence; If not satisfied, the single gait sequence is deleted; The pre-evaluation conditions include complete human form, natural walking state and image quality; The image quality includes at least one of a resolution of the gait image, a size of the gait image, and a number of frames of the gait image; Accordingly, it is determined whether the single-segment gait sequence meets the preset pre-evaluation conditions, specifically: determining whether the gait image in the single gait sequence contains a complete human figure; Determining whether the pedestrian in the single gait sequence is in a natural walking state; determining whether the image quality of the gait image in the single gait sequence is greater than a preset quality threshold; Determining whether the pedestrian in the single gait sequence is in a natural walking state specifically includes: respectively obtaining the mean, standard deviation, and covariance of two adjacent gait images in the single gait sequence; Calculating a first similarity between two adjacent gait images according to the mean, standard deviation, and covariance to obtain a plurality of first similarities; Calculating an average of multiple first similarities, and determining whether the average is greater than a preset similarity threshold; if so, the pedestrian in the single gait sequence is in a stationary state; Also includes: Acquire a plurality of gait sequences, wherein the plurality of gait sequences include a plurality of the single-segment gait sequences; Acquiring second characteristic parameters of the multiple gait sequences; evaluating the quality of the multiple gait sequences according to the quality scores of the single gait sequences and the second characteristic parameter to obtain quality scores of the multiple gait sequences; The second characteristic parameters include the number of sequences and the richness of perspectives.

2. The gait sequence evaluation method based on quality dimension according to claim 1 is characterized in that: The first characteristic parameter also includes at least one of the clarity of the human figure outline, the lighting intensity and the viewing angle.

3. The gait sequence evaluation method based on quality dimension according to claim 1, characterized in that: Calculating the overlap between the gait silhouette images in the gait silhouette sequence specifically includes: Acquire multiple gait silhouette images at set frame intervals in a gait silhouette sequence; Calculate the overlap between two adjacent gait silhouette images in the multiple gait silhouette images respectively, specifically including: sequentially acquiring two adjacent gait silhouette images from a plurality of gait silhouette images; The overlap degree between the two adjacent gait silhouette images is calculated according to the pixel values ​​at the same pixel position of the two adjacent gait silhouette images.

4. The gait sequence evaluation method based on quality dimension according to claim 2, characterized in that: When the first characteristic parameter is the clarity of the human figure outline, obtaining the first characteristic parameter of the single gait sequence specifically includes: Converting each gait image in the single gait sequence into a grayscale image; filtering the grayscale image to obtain a filtering sequence; The variance of each filtered image in the filtering sequence is calculated, and the clarity of the human figure outline is determined based on the variance.

5. The gait sequence evaluation method based on quality dimension according to claim 1, characterized in that: Also includes: The quality score of the single-segment gait sequence or the quality scores of the multiple-segment gait sequences are compared with a corresponding preset score threshold, and the single-segment gait sequence or the multiple-segment gait sequences are stored in a gait base library in a hierarchical manner according to the comparison result.

6. A gait sequence evaluation system based on quality dimension, characterized in that: The method for evaluating a gait sequence based on a quality dimension according to any one of claims 1 to 5 comprises: A sequence acquisition module, wherein the sequence acquisition module is used to acquire a single gait sequence; A feature acquisition module, configured to acquire a first feature parameter of the single-segment gait sequence; An evaluation module is used to evaluate the quality of the single-segment gait sequence according to the first characteristic parameter to obtain a quality score of the single-segment gait sequence.

Images