CN107341179A - Generation method, device and the storage device of standard movement database - Google Patents
Generation method, device and the storage device of standard movement database Download PDFInfo
- Publication number
- CN107341179A CN107341179A CN201710386848.6A CN201710386848A CN107341179A CN 107341179 A CN107341179 A CN 107341179A CN 201710386848 A CN201710386848 A CN 201710386848A CN 107341179 A CN107341179 A CN 107341179A
- Authority
- CN
- China
- Prior art keywords
- standard
- human body
- parts
- image sequence
- range image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/58—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually
- G06F16/5866—Retrieval characterised by using metadata, e.g. metadata not derived from the content or metadata generated manually using information manually generated, e.g. tags, keywords, comments, manually generated location and time information
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Library & Information Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a kind of generation method, device and the storage device of standard movement database.This method includes obtaining the range image sequence for including the action of human body standard movement；The parts of body of human body is marked according to range image sequence；The standard trajectory information of parts of body in being moved according to range image sequence record standard；Standard trajectory information is preserved to form standard movement database.The device includes depth camera, processor and memory.The storage device is had program stored therein data, and the routine data can be performed to realize the above method.The present invention can obtain the comprehensive information of human body during standard movement, and it can accurately differentiate hiding relation of trunk and limbs etc. etc., thus can obtain more comprehensively and accurately data to form standard movement database, and then can make the human motion in later stage assessment and analyze it is also more accurate.
Description
Technical field
The present invention relates to technical field of image processing, more particularly to a kind of generation method of standard movement database, dress
Put and storage device.
Background technology
The depth information that each pixel has in the depth image of depth camera capturing scenes is scene surface to depth phase
The distance of machine, so as to which the positional information of scene objects can be obtained according to depth image.
In the training of the motions such as ball game, track and field sports, overall human body attitude, action and parts of body are caught
Movement locus formed electronic data, the analysis to these data, to improve training effect it is significant.Existing skill
Art obtains data by wearing the electronic equipment of traceable movement locus, or using 2D image sequences to movement locus and people
Body posture carries out analysis and evaluation.To in the research and practice process of prior art, the inventors found that using traceable
The wearable electronic gathered data of movement locus is not comprehensive, is only limitted to wear the motion number in the region of the electronic equipment
According to, and when parts of body wears the electronic equipment, then motion-affecting will necessarily act, and in 2D image sequences for
With such as limbs in front of trunk the identification of posture of hiding relation can not accurately differentiate, can cause analysis result
It is inaccurate, it is difficult to improve training effect.
The content of the invention
The present invention provides a kind of generation method, device and the storage device of standard movement database, can solve the problem that existing skill
The problem of analysis result inaccuracy be present in art.
In order to solve the above technical problems, one aspect of the present invention is：A kind of standard movement database is provided
Generation method, this method comprises the following steps：Obtain the range image sequence for including the action of human body standard movement；According to described
The parts of body of the human body is marked range image sequence；In being moved according to the range image sequence record standard
The standard trajectory information of the parts of body；The standard trajectory information is preserved to form standard movement database.
In order to solve the above technical problems, another technical solution used in the present invention is：A kind of generation quasi-moving number is provided
According to the device in storehouse, the device includes depth camera, processor and memory, the depth camera and the memory with it is described
Processor connects；Wherein, the depth camera is used to obtain the range image sequence for including the action of human body standard movement；The place
Reason device is used to the parts of body of the human body be marked according to the range image sequence；According to the depth image sequence
The standard trajectory information of parts of body described in the motion of row record standard；The memory is used to preserve the standard trajectory letter
Cease to form standard movement database.
In order to solve the above technical problems, another technical scheme that the present invention uses is：A kind of storage device is provided, this is deposited
Storage device is had program stored therein data, and described program data can be performed to realize the above method.
The beneficial effects of the invention are as follows：Be different from the situation of prior art, the present invention by range image sequence at
Reason, the movement locus of parts of body during tracking human body standard movement action is to obtain standard trajectory information, and by the standard
Trace information is preserved to form standard movement database, using the normative reference of the athletic performance as later stage human body to be assessed.
The present invention can obtain the comprehensive information of human body during standard movement, and can accurately differentiate trunk and limbs etc.
Hiding relation etc., thus more comprehensively and accurately data can be obtained, to form standard movement database, and then the people in later stage can be made
The assessment and analysis of body motion are also more accurate.
Brief description of the drawings
Technical scheme in order to illustrate the embodiments of the present invention more clearly, make required in being described below to embodiment
Accompanying drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for
For those of ordinary skill in the art, on the premise of not paying creative work, other can also be obtained according to these accompanying drawings
Accompanying drawing.
Fig. 1 is a kind of schematic flow sheet of the generation method embodiment of standard movement database provided by the invention；
Fig. 2 is a kind of schematic flow sheet of another embodiment of generation method of standard movement database provided by the invention；
Fig. 3 is the schematic flow sheet of step S22 in Fig. 2；
Fig. 4 is the schematic flow sheet of record standard attitude information in step S23 in Fig. 2；
Fig. 5 be a kind of standard movement database provided by the invention the kneed barycenter of another embodiment of generation method with
The schematic diagram of the spatial relation at human body center；
Fig. 6 is a kind of structural representation for the device embodiment for generating standard movement database of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only the part of the embodiment of the present invention, rather than whole embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained all other under the premise of creative work is not made
Embodiment, belong to the scope of protection of the invention.
Referring to Fig. 1, the flow that Fig. 1 is a kind of generation method embodiment of standard movement database provided by the invention is shown
It is intended to.The generation method of standard movement database shown in Fig. 1 includes step：
S11, obtain the range image sequence for including the action of human body standard movement.
Wherein, the action of human body standard movement can be the action completed by professional athlete, coach etc..Wherein, mark
Quasi-moving action can be action of the action of deep-knee-bend motion, the action of push-up, sit-ups etc..Depth image not only wraps
The Pixel Information of space object is included, includes the depth information of each Pixel Information, i.e., object is between depth camera in space
Range information.Range image sequence can be obtained by depth camera, and range image sequence then referred in a period
Interior continuous depth image, i.e., the process for shooting the human body and entirely being acted is followed the trail of using depth camera.
S12, according to range image sequence the parts of body of human body is marked.
Specifically, the parts of body of human body can be the positions such as head, shoulder neck, trunk, four limbs, hand, foot, and
The joints such as knee, elbow, wrist, ankle, hip joint.By the analysis to range image sequence, human body is identified in range image sequence
Parts of body, and the parts of body is marked.
S13, the standard trajectory information according to parts of body in the motion of range image sequence record standard.
In step S13, the good parts of body of trace labelling in range image sequence, so as to obtain the parts of body
Standard trajectory information, and record the standard trajectory information.
S14, standard trajectory information is preserved to form standard movement database.
In step S14, the standard trajectory information of partes corporis humani position during human body standard movement is preserved to form data
Storehouse, with cause the later stage carry out human motion assessment when, its exercise data can be with the standard movement of standard movement lane database
Data are compared, and whether the motion done so as to analyze the human body to be assessed meets the requirement of standard movement action.For example, will
, can after the trace information of the parts of body of human body to be assessed is compared with the standard trajectory information of the standard movement lane database
To show whether the action of the human body to be assessed reaches the requirement of standard movement action, also, can be with it is further proposed that adjustment
It is recommended that including position and the adjustment direction of adjustment, for example, left hand needs to move still further below, biped portion spacing is turned down, double knee positions
No more than toe etc., in further embodiments, mobile distance further can also be clearly proposed, for example, left hand is downward
Mobile 5cm etc..
Prior art is different from, the present invention is handled by range image sequence, when tracking human body standard movement acts
Parts of body movement locus to obtain standard trajectory information, and the standard trajectory information is preserved to form standard movement
Database, using the normative reference of the athletic performance as later stage human body to be assessed.The present invention can obtain standard movement process
The comprehensive information of middle human body, and hiding relation of trunk and limbs etc. etc. can be accurately differentiated, thus can obtain more
Comprehensively and accurately data are to form standard movement database, and then can make the assessment of the human motion in later stage and analyze also more
Accurately.
Referring to Fig. 2, Fig. 2 is a kind of stream of another embodiment of generation method of standard movement database provided by the invention
Journey schematic diagram.
S21, obtain the range image sequence for including the action of human body standard movement.
S22, according to range image sequence the parts of body of human body is marked.
Specifically, as shown in figure 3, Fig. 3 is the schematic flow sheet of step S22 in Fig. 2.Step S22 includes：
Background in S221, removal series of depth images.
For example, a patch (blob, i.e. there is the connection group of the pixel of similar value) can be primarily determined that in depth map
As the body of object, other patches with significantly different depth value are then removed from the patch.It is preliminary by this way
The patch of determination generally has to have some minimum dimension.It is however, therefore, simple between the pixel coordinate of patch edge
Euclidean distance does not provide the accurate measurement of the size.The reason for this is inaccurate is, with the object with given actual size
The size (in units of pixel) of corresponding patch increaseds or decreases with the change of the object and the distance of equipment.
Therefore, in order to determine the actual size of object, first by following formula by (x, y, depth) coordinate of object
It is transformed to " real world " coordinate (xr, yr, depth)：
Xr=(x-fovx/2) * Pixel Dimensions * depth/reference depth
Yr=(y-fovy/2) * Pixel Dimensions * depth/reference depth
Here, fovx and fovy is the visual field of the depth map on x and y directions (in units of pixel).Pixel Dimensions are,
The length opposite to set a distance (reference depth) place pixel from plotting equipment.Then, the size of patch can be by seeking the spot
Euclidean distance between the real-world coordinates of block edge actually determines.
Therefore, the background in depth image can be removed by identifying the patch with required minimum dimension, its
In, there is smallest average depth value among each patch of the patch in the scene.It assume that, the spot nearest apart from depth camera
Block is human body, and all pixels of big at least some threshold value of the depth ratio average depth value, which are all assumed to be, belongs to background object, and
The depth value of these pixels is arranged to null value.Wherein, above-mentioned threshold value can determine according to being actually needed.In addition, at some
, can also be by each pixel zero setting of the depth value with the average depth value for being significantly less than patch in embodiment.Furthermore it is also possible to
A depth capacity is preset, so as to ignore the object for exceeding the depth capacity.
In certain embodiments, depth value can also be dynamically determined, if the depth value, object is just from depth map
Middle removal.Therefore, assume that the object in scene moves.Therefore, depth does not have times changed in certain minimal amount frame
What pixel is assumed to be background object.The pixel that depth value is more than the static depth value is considered as to belong to background object,
Therefore all it is zeroed out.Start, all pixels in scene can all be defined as all pixels in static state, or scene can be with
All it is defined as non-static.In both situations, once object setting in motion, it is possible to the actual depth mistake of dynamic generation
Filter.
It is, of course, also possible to remove the background in depth image by other methods well known in the prior art.
S222, obtain range image sequence in human body profile.
After background is removed, the exterior contour of body can be found out in depth map by edge detection method.This reality
Apply in example, the profile of human body is found out using two step thresholding mechanism：
First, travel through with all pixels in humanoid corresponding patch in depth image, also, if any give fixation
Element has effective depth value, and if the pixel and at least one in its four neighborhood pixels being connected (right, left, upper and lower)
The difference of depth value between individual pixel is more than first threshold, then is marked as outline position.(wherein, effective depth value and zero
Difference between value is considered as infinitely great).
Then, previous step and then secondary traversal patch are being completed, and if (pixel is also in any pixel
Be not flagged as outline position) eight connected neighborhood pixels among have a contour pixel, and if current pixel and surplus
Under connected close position at least one pixel between the difference of depth value be more than Second Threshold (being less than first threshold), then
It is marked as outline position.
S223, the trunk according to outline identification human body.
Finding out the outline of human body and then identifying each position of body, for example, head, trunk and four limbs.
First rotate depth image so that body contour is in vertical position.The purpose of the rotation is in order to by by body
The longitudinal axis alignd with Y-coordinate (vertical) axle to simplify the calculating in following step.Selectively, following calculating can be relative to body
The longitudinal axis of body performs, without carrying out the rotation, as understood by those skilled in the art.
Before each position of identification body, the 3D axles of body can be first found out.Specifically, the 3D axles for finding out body can
With using following methods：
It is node grid by original depth image down-sampling (down-sample), wherein, in the x-direction and the z-direction every n
Individual pixel takes a node.The depth value of each node is calculated based on the depth value in n × n squares centered on node.
If there is null value more than half pixel in square, respective nodes are arranged to null value.Otherwise, the node is arranged to n × n
The average value of effective depth value in square.
It is then possible to based on the value of adjacent node come the depth image of further " cleaning " down-sampling：If given section
Most of adjacent node of point has null value, then the node being also configured as into null value, (it has after abovementioned steps
The depth value of effect).
When above-mentioned steps are completed, the longitudinal axis of remaining node in the figure of down-sampling is found out.Therefore, linear minimum can be carried out
Two multiply fitting is most fitted the line of each node to find out.Selectively, an ellipse around each node can be fitted and find out it
Main shaft.
After the 3D axles of body are found out, by the thickness that body contour is measured on the direction for be parallel and perpendicular to the longitudinal axis
To identify the trunk of body.Therefore, constraint frame can be limited around body contour, then can be to the pixel value in the frame
Carry out binaryzation：Pixel with depth zero value is set to 0, and the pixel with non-zero depth value is set to 1.
Then, by being added along corresponding vertical line to binary pixel values, each X values calculating to inframe is vertical
To thickness value, and by being summed up along corresponding horizontal line to binary pixel values, transverse gage is calculated to each Y value
Value.To resulting value threshold application, to identify along which bar vertical line and horizontal line profile relative thick.
When the transverse gage in profile certain level region is more than X threshold values, the longitudinal thickness of a certain vertical area is more than Y threshold values
When, the common factor of the horizontal zone and vertical area can be defined as trunk.
S224, the parts of body according to trunk identification human body.
After trunk is determined, the head of body and four limbs can be identified based on geometry consideration.Hand arm is to connect
It is connected to the left side of torso area and the region on right side；Head is the join domain above torso area；Leg is under torso area
The join domain of side.The upper left corner of torso area and the upper right corner can also be tentatively identified as shoulder.
S225, parts of body is marked.
Parts of body is marked in order to be tracked to the movement locus of the parts of body.
In another embodiment, identifying the parts of body of human body can also be realized by following three steps：
(1) human body segmentation.The method that the present embodiment is combined using inter-frame difference and background subtraction split-phase is moved to split
Human body, the frame in RGBD images is chosen in advance as background frames, establishes the Gauss model of each pixel, then with inter-frame difference
Method carries out difference processing to adjacent two field pictures, distinguishes background dot and region (the region bag changed in the current frame of change
Include and appear area and moving object), then the corresponding region of region of variation and background frames progress models fitting is distinguished and appears area
And moving object, shade is finally removed in moving object, so comes out the moving meshes without shade.Context update
When inter-frame difference is determined as to the point of background, then be updated with certain rule；Background subtraction timesharing is determined as appearing area
Point, then background frames are updated with larger turnover rate, region corresponding to moving object is without renewal.The method can be managed relatively
The segmentation object thought.
(2) contours extract and analysis.After the image after obtaining binaryzation, calculated using some classical rim detections
Method obtains profile.For example with Canny algorithms, Canny edge detection operators fully reflect the number of optimal edge detector
Characteristic is learned, for different types of edge, good signal to noise ratio is respectively provided with, excellent positioning performance, single edge is produced more
The low probability of individual response and the maximum suppression ability to false edge response, after obtaining light stream segmentation field using partitioning algorithm,
Our all moving targets of concern are included in these cut zone.Therefore, Canny will be utilized in these cut zone
Operator extraction edge, ambient interferences on the one hand can be limited significantly, on the other hand can effectively improve the speed of operation.
(3) artis marks automatically.Obtain being partitioned into moving target by calculus of finite differences, Canny edge detection operators carry
After contouring, by MaylorK.LeungandYee-HongYang 2D belt patterns (RibbonModel) to human body target
Further analysis.Human body front is divided into different regions by the model, for example, human body is constructed with 5 U-shaped regions, this 5
U-shaped region represents head and the four limbs of human body respectively.
So, the body end points of 5 U-shapes of searching is passed through, so that it may the approximate location of body is determined, in the profile extracted
On the basis of, compressed by vector outline, to extract the information of needs, retain the feature of most important human limb, by human body
Profile is compressed into a fixed shape, such as so that profile is inverted U-shaped with fixed 8 end points and 5 U-shaped points and 3
Point, so obvious feature can conveniently calculate profile.The distance algorithm that adjacent end points on profile may be used herein carrys out compression wheel
Exterior feature, pass through 8 end points of iterative processing profile boil down to.
Automatic marking is carried out to parts of body using following algorithm can after compression profile is obtained：
(1) the body end points of U-shape is determined.Some reference length M is set, the vector more than M can consider that it is body
A part for body profile, then ignore less than it.Begun look for according to the profile after vector quantization from certain point, find one and be more than M
Vector be designated as Mi, find next vector and be designated as Mj, compare Mi to Mj angle, if angle within a certain range (0~
90 °) (noticing that angle is just, it is convex to represent it here), then it is assumed that they are U end points, record the two vectors, find one
U end points.So until finding out 5 U end points.
(2) end points of three inverted u-shaped is determined.Same step (1), as long as angle condition is just being changed to negative.
(3) head, hand, the position of pin are readily available according to U and U end points.According to the physiology shape of body, so that it may
To determine each artis, using arm and body angle portions, head and leg angle portions, trunk can be determined respectively
Width and length；Then trunk ratio 0.75,0.3 is accounted for respectively using neck, waist position, ancon is located at the midpoint of shoulder and hand,
Knee is located at the midpoint of waist and pin.So parts of body approximate location, which can define, comes.
S23, the standard trajectory information according to parts of body in the motion of range image sequence record standard, and record standard
In motion repetitive operation start or at the end of human body standard attitude information.
Record parts of body standard movement track method have it is a variety of, for example, OGHMs (Orthogonal
Gaussian-Hermite Moments) detection method, its general principle is：By comparing between continuous picture frame in time
The intensity of variation of corresponding pixel value judges whether the pixel belongs to foreground moving region.
One group of image sequence of input is represented with { f (x, y, t) | t=0,1,2 ... }, f (x, y, t) represents the figure of t
Picture, x, y represent the coordinate of pixel on image, if Gaussian functions are g (x, σ), Bn (t) is g (x, σ) and Hermite
Polynomial product, then n ranks OGHMs be represented by：
Wherein aiDetermined by the standard deviation of Gaussian functions.According to the property of convolution algorithm, n ranks OGHMs can be regarded as
It is the convolution of the all-order derivative sum of image sequence function in time and Gaussian functions.Certain point derivative value is bigger, then table
Show that the pixel value changes changed over time on the position are also bigger, illustrate that the point should belong to moving region block, this is OGHMs
Method can detect that moving object provides theoretical foundation.In addition, can be seen that from formula (1), OGHMs basic function isThis is formed by the different order derivative linear combinations of Gaussian functions.Because Gaussian function has in itself
There is the ability of smooth noise, so OGHMs equally effectively filters out the performance of various noises.
And for example, time differencing method, time differencing method (Temporal Difference) are to utilize image continuous in time
Several consecutive frames before and after sequence, the time difference based on pixel, the moving region in image is extracted by thresholding.The side of early stage
Method is to obtain moving object using adjacent two frame difference, such as sets FkIt is kth frame gradation of image Value Data, F in image sequencek+1Table
Show the two field picture of kth+1 gray value data in image sequence, then the difference image of temporally adjacent two field pictures is defined as：
Wherein T is threshold value.If difference is more than T, illustrate that the grey scale change in the region is larger, that is, need the fortune detected
Moving-target region.
And for example, optical flow method (Optical Flow), optical flow method be based on it is assumed hereinafter that：The change of gradation of image entirely due to
Caused by the motion of target or background.That is, the gray scale of target and background does not change over time.Motion inspection based on optical flow approach
Survey, the characteristic for showing as velocity field in the picture is exactly changed over time using moving object, is estimated according to certain constraints
The corresponding light stream of motion is calculated, its advantage is the interframe movement less-restrictive to target, can handle larger interframe displacement.
For another example, background subtraction method (Background Subtraction), its general principle are to build a background first
Model image, then made the difference point with current frame image and background two field picture, moving target is detected by thresholding difference result.
Assuming that t background two field picture is F0, corresponding current frame image is Ft, then the difference of present frame and background frames be represented by：
Assuming that the gray value difference of current frame image and background two field picture respective pixel is more than threshold value, then resulting two-value
Corresponding value is 1 in image, that is, assert that the region belongs to moving target.
In some other embodiment, motion rail of the son to the parts of body of human body can also be described by HOG-HOF
Mark is described and obtains trace information.
Wherein, referring to Fig. 4, Fig. 4 is the schematic flow sheet of record standard attitude information in step S23 in Fig. 2.Record mark
In quasi-moving repetitive operation start or at the end of the standard attitude information of human body can include：
S231, the parts of body of human body is identified according to range image sequence and determines the human body reference point of the human body.
Specifically, human body reference point can be mass center of human body or human body center, and the present embodiment is used as people using human body center
The present invention will be described for body reference point.Certainly, in some other embodiment, it is also an option that other specified points of human body are made
For human body reference point.
Human body center is the geometric center of human body in depth image, can be identified in the parts of body when trunk and human body
After out, you can by the profile of the whole human body of depth image, i.e. the intermediate value of the outside edge value at 3 D human body edge is come true
Determine human body center.
S232, obtain parts of body and the standard relative position relation of human body reference point.
The standard relative position relation of the present embodiment is the parts of body of the human body of the standard movement of posture to be assessed
Barycenter and the relative position relation at human body center.In one embodiment, relative position relation can be the matter of parts of body
Euclidean distance and COS distance between the heart and ginseng human body center, and standard can be formed according to the Euclidean distance and COS distance
The standard vector of body posture.For example, Euclidean distance and COS distance of the barycenter on head to human body center, hand barycenter to people
The Euclidean distance and COS distance at body center etc..
The calculation of Euclidean distance and COS distance can be as follows：
First, first coordinate value at the human body center of standard posture is obtained.In the present embodiment, first coordinate at human body center
It is worth the coordinate value in the camera coordinates system of depth camera for human body center.
For example, in the end posture of deep-knee-bend action, human body central point A the first coordinate value is (x1,y1,z1)。
Then the second coordinate value of the barycenter of parts of body and the barycenter of parts of body is obtained.
Specifically, after parts of body identifies, it may be determined that the barycenter in each region of body.Wherein,
The barycenter in region refers to representative depth or the position in the region.Therefore, for example, can with the histogram of depth value in formation zone, and
Depth value (or average value of two or more depth values with highest frequency) with highest frequency is set to the region
Barycenter.After the barycenter that parts of body is determined, you can determine coordinate of the barycenter of parts of body in camera coordinates system.
It is noted that the barycenter in the present invention refers to handle acquired barycenter by depth image, and not thing
Manage barycenter.The barycenter of the present invention can be obtained by centroid method, can also be obtained by other methods, and the present invention does not limit.
For example, in the end posture of deep-knee-bend action, the barycenter B of knee endoprosthesis the second coordinate value is (x2,y2,z2)。
Finally, according to the first coordinate value and the second coordinate value calculate the barycenter of parts of body and the European of human body center away from
From and COS distance, and form the standard vector of the standard posture of the human body.
COS distance, also referred to as cosine similarity, it is to be used as measurement by the use of two vectorial angle cosine values in vector space
The measurement of the size of two interindividual variations.This concept is borrowed in machine learning to weigh the difference between sample vector.Its
In, two vectorial COS distances can be represented with the cosine value of angle between them.
For example, as shown in figure 5, Fig. 5 be a kind of standard movement database provided by the invention generation method it is another
The schematic diagram of the kneed barycenter of embodiment and the spatial relation at human body center.Obtaining the first coordinate value and the second coordinate
After value, it can be deduced that the vector at human body centerAnd the vector of the barycenter of knee endoprosthesisSpecifically, the Euclidean distance between the barycenter of knee endoprosthesis and human body center passes through below equation meter
Calculate and obtain：
WithBetween COS distance can by below equation calculate and：
Wherein, what Euclidean distance was weighed is the absolute distance of spatial points, for example, dABIt is absolute between measurement point A and point B
Distance, it is directly related with the position coordinates where each point；And COS distance weigh be space vector angle, more embody
Difference on direction, rather than position.
Specifically, COS distance span is [- 1,1].Included angle cosine is bigger to represent that two vectorial angles are smaller, folder
The angle of the smaller vector of expression two of angle cosine is bigger.When two vectorial directions overlap, included angle cosine takes maximum 1, when two
The complete opposing angles cosine in direction of vector takes minimum value -1.
Certainly, in human body attitude evaluation process, the barycenter of hand, foot etc. other body parts generally can also be calculated
To the Euclidean distance and COS distance at human body center.Finally, the barycenter of all required body parts is to the European of human body center
Value obtained by distance and COS distance corresponds with each body part, and is integrally formed the standard vector of the standard posture.
When later stage is assessed human motion, vector to be assessed can be formed by method same as described above,
Assessed by the vectorial comparison with standard vector to be assessed.
It is to be appreciated that in some other embodiment, when human body attitude can also be some time point in motion process
Human body attitude, and be not limited to start or at the end of human body attitude.
S24, standard trajectory information and standard relative position relation are preserved to form standard movement database.
Specifically, in the present embodiment, step S24 can preserve standard relative position while standard trajectory information is preserved
Relation, that is, the barycenter and the Euclidean distance and COS distance of human body reference point of the parts of body of the human body of standard posture are preserved,
And standard vector, so that when assessing human action and posture, the trace information of human body to be assessed and standard trajectory information to be entered
Row compares, by the Euclidean distance and COS distance of the barycenter of the parts of body of human body to be assessed and human body reference point and standard appearance
The barycenter of the parts of body of the human body of state, i.e., will be to be assessed compared with the Euclidean distance and COS distance of human body reference point
Vector sum standard vector is compared.
The present embodiment is not only preserved standard trajectory information by the processing of range image sequence, and also extraction repeats dynamic
Make to start or at the end of standardized human body's posture in parts of body and human body reference point, and pass through the parts of body and people
Relative position relation between body reference point obtains and preserved standard attitude information, and it can differentiate partes corporis humani position exactly
Position simultaneously obtains accurate relative position relation, to improve the accuracy of posture assessment result, so as to improve every motion
Training effectiveness.
Referring to Fig. 6, Fig. 6 is a kind of structural representation for the device embodiment for generating standard movement database of the present invention.
Specifically, the device of the generation quasi-moving database includes depth camera 10, processor 11 and memory 12, depth
Camera 10 and memory 12 are connected with processor 11.
Wherein, depth camera 10 is used to obtain the range image sequence for including the action of human body standard movement.
Processor 11 is used to the parts of body of human body be marked according to range image sequence；According to depth image sequence
The standard trajectory information of parts of body in the motion of row record standard.
Memory 12 is used to preserve standard trajectory information to form standard movement database.
In the present embodiment, processor 11 is additionally operable to remove the background in range image sequence；Obtain range image sequence
In human body profile；According to the trunk of outline identification human body；The parts of body of human body is identified according to trunk；To each portion of body
Position is marked.
Alternatively, processor 11 be additionally operable to record standard motion in repetitive operation start or at the end of human body standard posture
Information；Preservation standard attitude information.
Alternatively, processor 11 is additionally operable to identify the parts of body of human body according to range image sequence and determines the human body
Human body reference point；Obtain parts of body and the standard relative position relation of human body reference point.
Memory 12 is additionally operable to preservation standard relative position relation.
In certain embodiments, standard relative position relation for standard posture human body parts of body barycenter to
The relative position relation of the human body reference point of the human body；Processor 11 is additionally operable to first seat at the human body center of acquisition standard posture
Scale value；Obtain the second coordinate value of the barycenter of parts of body and the barycenter of parts of body；According to the first coordinate value and second
Coordinate value calculates the barycenter of the parts of body of the human body of standard posture and the Euclidean distance and COS distance of human body reference point, with
Form the standard vector of the standard physical posture.
Present invention also offers a kind of storage device, the storage device is had program stored therein data, and the routine data can be by
Perform to realize the generation method of the standard movement database of any of the above-described embodiment.
For example, the storage device can be portable storage media, such as USB flash disk, mobile hard disk, read-only storage
(ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disc or CD
Deng.It is to be appreciated that storage device can also be that server etc. is various can be with the medium of store program codes.
In summary, the present invention can obtain the comprehensive information of human body during standard movement, and can accurately differentiate
Hiding relation of trunk and limbs etc. etc., thus can obtain more comprehensively and accurately data to form standard movement data
Storehouse, and then the assessment of the human motion in later stage and analysis can be made also more accurate.
Embodiments of the present invention are the foregoing is only, are not intended to limit the scope of the invention, it is every to utilize this
The equivalent structure or equivalent flow conversion that description of the invention and accompanying drawing content are made, or directly or indirectly it is used in other correlations
Technical field, it is included within the scope of the present invention.
Claims (11)
1. a kind of generation method of standard movement database, it is characterised in that comprise the following steps：
Obtain the range image sequence for including the action of human body standard movement；
The parts of body of the human body is marked according to the range image sequence；
The standard trajectory information of parts of body according to range image sequence record standard motion；
The standard trajectory information is preserved to form standard movement database.
2. according to the method for claim 1, it is characterised in that it is described according to the range image sequence to the human body
The step of parts of body is marked includes：
Remove the background in the range image sequence；
Obtain the profile of the human body in the range image sequence；
According to the trunk of human body described in the outline identification；
The parts of body of the human body is identified according to the trunk；
The parts of body is marked.
3. according to the method for claim 2, it is characterised in that described to be moved according to the range image sequence record standard
Described in parts of body trace information the step of also include：In record standard motion repetitive operation start or at the end of it is described
The standard attitude information of human body；
The preservation standard trajectory information is also included with being formed the step of standard movement database：Preserve the standard posture
Information.
4. according to the method for claim 3, it is characterised in that repetitive operation starts or terminated in the record standard motion
The step of standard attitude information of Shi Suoshu human bodies, includes：
The parts of body of the human body is identified according to the range image sequence and determines the human body reference point of the human body；
Obtain the standard relative position relation of the parts of body and the human body reference point；
The step of preservation standard attitude information, includes：Preserve the standard relative position relation.
5. according to the method for claim 4, it is characterised in that the standard relative position relation is the human body of standard posture
Parts of body barycenter to the human body reference point with the human body relative position relation；
Described the step of obtaining the parts of body and the standard relative position relation of the human body reference point, includes：
First coordinate value at the human body center of acquisition standard posture；
Obtain the second coordinate value of the barycenter of the parts of body and the barycenter of parts of body；
The matter of the parts of body of the human body of the standard posture is calculated according to first coordinate value and second coordinate value
The heart and the Euclidean distance and COS distance of human body reference point, to form the standard vector of the standard physical posture.
6. a kind of device for generating quasi-moving database, it is characterised in that described including depth camera, processor and memory
Depth camera and the memory are connected with the processor；
Wherein, the depth camera is used to obtain the range image sequence for including the action of human body standard movement；
The processor is used to the parts of body of the human body be marked according to the range image sequence；According to described
The standard trajectory information of parts of body described in the motion of range image sequence record standard；
The memory is used to preserve the standard trajectory information to form standard movement database.
7. device according to claim 6, it is characterised in that the processor is additionally operable to remove the range image sequence
In background；Obtain the profile of the human body in the range image sequence；According to the body of human body described in the outline identification
It is dry；The parts of body of the human body is identified according to the trunk；The parts of body is marked.
8. device according to claim 7, it is characterised in that the processor is additionally operable to repeat to move in record standard motion
Make start or at the end of the human body standard attitude information；Preserve the standard attitude information.
9. device according to claim 8, it is characterised in that the processor is additionally operable to according to the range image sequence
Identify the parts of body of the human body and determine the human body reference point of the human body；Obtain the parts of body and the human body
The standard relative position relation of reference point；
The memory is additionally operable to preserve the standard relative position relation.
10. device according to claim 9, it is characterised in that the standard relative position relation is the people of standard posture
Relative position relation of the barycenter of the parts of body of body to the human body reference point with the human body；
The processor is additionally operable to first coordinate value at the human body center of acquisition standard posture；Obtain the matter of the parts of body
Second coordinate value of the barycenter of the heart and parts of body；The mark is calculated according to first coordinate value and second coordinate value
The barycenter of the parts of body of the human body of quasi- posture and the Euclidean distance and COS distance of human body reference point, to form the standard body
The standard vector of body posture.
11. a kind of storage device, it is characterised in that had program stored therein data, and described program data can be performed to realize such as
Method described in any one of claim 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710386848.6A CN107341179B (en) | 2017-05-26 | 2017-05-26 | Standard motion database generation method and device and storage device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710386848.6A CN107341179B (en) | 2017-05-26 | 2017-05-26 | Standard motion database generation method and device and storage device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107341179A true CN107341179A (en) | 2017-11-10 |
| CN107341179B CN107341179B (en) | 2020-09-18 |
Family
ID=60220186
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710386848.6A Active CN107341179B (en) | 2017-05-26 | 2017-05-26 | Standard motion database generation method and device and storage device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107341179B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109635925A (en) * | 2018-11-30 | 2019-04-16 | 北京首钢自动化信息技术有限公司 | A kind of sportsman's supplemental training data capture method, device and electronic equipment |
| CN111164376A (en) * | 2018-03-26 | 2020-05-15 | 株式会社爱考斯研究 | Body orientation estimation device and body orientation estimation program |
| CN112883969A (en) * | 2021-03-01 | 2021-06-01 | 河海大学 | Rainfall intensity detection method based on convolutional neural network |
| CN114599487A (en) * | 2019-12-10 | 2022-06-07 | 卡普西克斯公司 | Device for defining a sequence of movements on a generic model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101657825A (en) * | 2006-05-11 | 2010-02-24 | 普莱姆传感有限公司 | Modeling of humanoid forms from depth maps |
| KR20130110441A (en) * | 2012-03-29 | 2013-10-10 | 한국과학기술원 | Body gesture recognition method and apparatus |
| CN103390174A (en) * | 2012-05-07 | 2013-11-13 | 深圳泰山在线科技有限公司 | Physical education assisting system and method based on human body posture recognition |
| CN104036488A (en) * | 2014-05-04 | 2014-09-10 | 北方工业大学 | Binocular vision-based human body posture and action research method |
-
2017
- 2017-05-26 CN CN201710386848.6A patent/CN107341179B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101657825A (en) * | 2006-05-11 | 2010-02-24 | 普莱姆传感有限公司 | Modeling of humanoid forms from depth maps |
| KR20130110441A (en) * | 2012-03-29 | 2013-10-10 | 한국과학기술원 | Body gesture recognition method and apparatus |
| CN103390174A (en) * | 2012-05-07 | 2013-11-13 | 深圳泰山在线科技有限公司 | Physical education assisting system and method based on human body posture recognition |
| CN104036488A (en) * | 2014-05-04 | 2014-09-10 | 北方工业大学 | Binocular vision-based human body posture and action research method |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111164376A (en) * | 2018-03-26 | 2020-05-15 | 株式会社爱考斯研究 | Body orientation estimation device and body orientation estimation program |
| CN109635925A (en) * | 2018-11-30 | 2019-04-16 | 北京首钢自动化信息技术有限公司 | A kind of sportsman's supplemental training data capture method, device and electronic equipment |
| CN114599487A (en) * | 2019-12-10 | 2022-06-07 | 卡普西克斯公司 | Device for defining a sequence of movements on a generic model |
| CN112883969A (en) * | 2021-03-01 | 2021-06-01 | 河海大学 | Rainfall intensity detection method based on convolutional neural network |
| CN112883969B (en) * | 2021-03-01 | 2022-08-26 | 河海大学 | Rainfall intensity detection method based on convolutional neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107341179B (en) | 2020-09-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107335192A (en) | Move supplemental training method, apparatus and storage device | |
| CN107392086A (en) | Apparatus for evaluating, system and the storage device of human body attitude | |
| CN111144217B (en) | Motion evaluation method based on human body three-dimensional joint point detection | |
| Hesse et al. | Computer vision for medical infant motion analysis: State of the art and rgb-d data set | |
| CN101604447B (en) | No-mark human body motion capture method | |
| CN107174255A (en) | Three-dimensional gait information gathering and analysis method based on Kinect somatosensory technology | |
| US9235753B2 (en) | Extraction of skeletons from 3D maps | |
| CN102184541B (en) | Multi-objective optimized human body motion tracking method | |
| CN103679171B (en) | A gait feature extraction method based on human body gravity center track analysis | |
| CN101894278B (en) | Human motion tracing method based on variable structure multi-model | |
| CN102609683A (en) | Automatic labeling method for human joint based on monocular video | |
| CN109344694B (en) | Human body basic action real-time identification method based on three-dimensional human body skeleton | |
| CN108717531A (en) | Estimation method of human posture based on Faster R-CNN | |
| CN102622766A (en) | Multi-objective optimization multi-lens human motion tracking method | |
| CN114067358A (en) | Human body posture recognition method and system based on key point detection technology | |
| KR20110113152A (en) | Apparatus, method and computer-readable medium providing marker-less motion capture of human | |
| US9990857B2 (en) | Method and system for visual pedometry | |
| CN107341179A (en) | Generation method, device and the storage device of standard movement database | |
| CN107742306A (en) | Moving Target Tracking Algorithm in a kind of intelligent vision | |
| CN107564035A (en) | The video tracing method for being identified and being matched based on important area | |
| Senior | Real-time articulated human body tracking using silhouette information | |
| Beltrán et al. | Automated Human Movement Segmentation by Means of Human Pose Estimation in RGB-D Videos for Climbing Motion Analysis. | |
| CN110765925A (en) | Carrier detection and gait recognition method based on improved twin neural network | |
| CN115205744A (en) | Intelligent exercise assisting method and device for figure skating | |
| Lim et al. | Depth image based gait tracking and analysis via robotic walker |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder | ||
| CP01 | Change in the name or title of a patent holder |
Address after: The joint headquarters building in Nanshan District Guangdong streets in Shenzhen city of Guangdong province 518057 Xuefu Road No. 63 building high-tech zones 11-13 Patentee after: Obi Zhongguang Technology Group Co., Ltd Address before: The joint headquarters building in Nanshan District Guangdong streets in Shenzhen city of Guangdong province 518057 Xuefu Road No. 63 building high-tech zones 11-13 Patentee before: SHENZHEN ORBBEC Co.,Ltd. |