CN103517042A - Nursing home old man dangerous act monitoring method - Google Patents

Nursing home old man dangerous act monitoring method Download PDF

Info

Publication number
CN103517042A
CN103517042A CN201310489096.8A CN201310489096A CN103517042A CN 103517042 A CN103517042 A CN 103517042A CN 201310489096 A CN201310489096 A CN 201310489096A CN 103517042 A CN103517042 A CN 103517042A
Authority
CN
China
Prior art keywords
old man
frame
tracks
tracking
behavior
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
Application number
CN201310489096.8A
Other languages
Chinese (zh)
Other versions
CN103517042B (en
Inventor
于哲舟
刘小华
李斌
刘昱昊
逄淑超
郑恒
刘继健
吴朝霞
章杰
于祥春
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Jilin University
Original Assignee
Jilin University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Jilin University filed Critical Jilin University
Priority to CN201310489096.8A priority Critical patent/CN103517042B/en
Publication of CN103517042A publication Critical patent/CN103517042A/en
Application granted granted Critical
Publication of CN103517042B publication Critical patent/CN103517042B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Abstract

The invention discloses a nursing home old man dangerous act monitoring method. The problems of low old man tracking accuracy, poor behavior analysis effect, frequent behavior recognition algorithm calling, large system load, excessive dependence on device performance of a monitoring effect and the like in the prior art are overcome. The method is characterized in that a monitoring camera, a video capture card, a personal PC machine, a display and an alarm sound box are used to be combined with a contour detection program module, a tracking program module and a behavior recognition program module in the personal PC machine; and the method based on the combination of target tracking and behavior recognition monitors the movement states of old men in a nursing home in daily life and gives an alarm due to the risks and the abnormal behaviors of the old men. The method has the advantages of high monitoring efficiency, high old man behavior recognition degree, low operation cost and good practicability and application prospect.

Description

A kind of home for destitute old man's hazardous act monitoring method
Technical field
The present invention relates to a kind of method of technical field of computer vision, be specifically related to the monitoring method of a kind of home for destitute old man's hazardous act.
Background technology
By retrieval and indexing, the problem that obtains the prior art close with the technology of the present invention field and existence thereof is as follows:
1. Chinese patent publication No. is CN 102547216A, date of publication is on July 4th, 2012, denomination of invention is a kind of video frequency graphic monitoring system for major hazard source, speaks of a kind of video images detection system that supervisory control system is applied to major hazard source in this invention.This patent, by different types of video camera is organically combined, has formed a complicated efficient system.But this system has two shortcomings, this system cannot be effectively applied in old man's hazardous act monitoring in home for destitute: the investment of (1) this system is excessive, although the monitoring that the too much high performance video camera of quantity can comprehensive multi-angle, but the in the situation that of the limited fund of home for destitute, common CCTV camera is only optimal selection.(2) full manual monitoring, psychologic research shows, when employee is engaged in monotonous work for a long time, employee's decreased attention, reaction speed is also followed decline, directly caused employee to ignore the suspicious event occurring on screen, thereby security system was lost efficacy, lost the chance of very first time prevention crime, also can produce totally unfavorable impact to employee's physical and mental health simultaneously, in addition, home for destitute is for cost reason, watch-dog generally by nurse on duty on behalf of supervision, nurse is when being busy with one's work, be easy to ignore monitored picture, supervisory control system was lost efficacy.
2. Chinese patent publication No. is CN 102764131A, date of publication is on November 7th, 2012, denomination of invention is that system and method is taken care of in a kind of long-range old man monitoring of living, in this invention, speak of a kind of various device that comprehensively passes through, comprise: a series of equipment such as watch-dog, medical sign equipment forms the system of remote monitoring old man life, this system has the advantage without installation and measuring equipment with it old man, old man is lived noiseless, and can different alarm threshold values be set according to old man's Different Individual setting or self study, make this system more intelligent.This system side overweights monitoring old man physical condition, and biological informations such as heart rate, blood pressure, blood sugar, blood oxygen, and between WA is taken medicine the rule of life information such as time, finds in time body abnormality, ahead of time diagnosis.But the real-time of this system is not high, the too much physical trait information that depends on, easily ignores old man's foudroyant disease, and cost of investment is too high.
3. Chinese patent publication No. is CN 101727570A, date of publication is on June 9th, 2010, denomination of invention is tracking, detects and follow the tracks for the treatment of facility and supervisory control system, in this invention, speak of a kind of system of utilizing tracking to monitor, this invention is by the possible position of target of prediction, and definite connected region, gained characteristic parameter and original characteristic parameter are compared to judge the position of target, and target signature is upgraded.This invention can solve preferably carries out the target signature variation issue in tracing process to target, and this invention can effectively be applied in supervisory control system, improves the accuracy of target following.But this invention only can be monitored by the movement locus of target, and cannot distinguish the different behaviors of target under same movement locus.
Summary of the invention
The present invention for overcome to old man follow the tracks of that accuracy rate is low, behavioural analysis weak effect, behavior recognizer call frequently, system load is large and monitoring effect too relies on the shortcoming of the aspects such as equipment performance, propose a kind of based target and follow the tracks of old man's hazardous act monitoring method of identifying the home for destitute combining with behavior.
In order to overcome the problems referred to above the present invention, adopt following technical scheme to realize: the step of described a kind of home for destitute old man's hazardous act monitoring method is as follows:
Step 1, usage monitoring video camera, video frequency collection card, individual PC obtain the old man's who enters guarded region video image;
Step 2, use the profile detection module in individual PC to carry out profile detection to the video image obtaining in step 1, the number of pixels of calculating old man's contour area of gained by frame difference method changes to judge the current residing state of old man.Use frame difference method, to entering the old man of guarded region, carry out profile detection, two continuous frames is monitored to picture X 1, X 2according to its corresponding gray value, subtract each other, i.e. X 1(i, j)-X 2the profile that (i, j) ≥δ region is old man, X 1(i, j)-X 2(i, j) background area that < δ is old man, calculates the number of pixel in the corresponding connected region of profile, and the number of pixel is when becoming less many, for old man enters guarded region, now do not need profile information to pass to tracking module; When the number of pixel no longer continues to increase, can judge that old man has entered guarded region, now old man's profile information is passed to tracking module, this module is every operation in 1 second once.
Tracking module in step 3, individual PC is followed the tracks of the old man's who obtains profile information, the rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and the ORB feature of extracting in tracking frame is local feature description's subcharacter of rotational invariance, and ORB characteristic is local feature description's subcharacter number of rotational invariance, the center position coordinate of continuous several frames is input in the fitting function in tracking module, and the result drawing by fitting function is predicted old man's next frame position; Detailed process is as follows:
The rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and ORB feature (the English full name: Oriented FAST and Rotated BRIEF in frame is followed the tracks of in extraction, local feature description's subcharacter of rotational invariance) and ORB characteristic (local feature description's subcharacter number of rotational invariance), to follow the tracks of the detailed process of the ORB feature in frame as follows for described extraction present frame:
First present frame is carried out to piecemeal: the image in current tracking frame is extracted separately, with I, represent this picture, if image is the matrix of a m * n dimension, the fritter that image is divided to several k * k pixels, like this, whole picture is divided into altogether (m/k) * (n/k) individual fritter, the pixel value that the mean value of getting the pixel of each fritter is this fritter, I (x, y) be expressed as x be listed as y capable the average gray value of corresponding fritter.
Then select the ORB characteristic point of present frame: according to m pq=∑ x,yx py pi (x, y), p wherein, q respectively value be 0 and 1, x ∈ [1 ..., m/k], y ∈ [1 ..., n/k] and obtain gradient θ=argtan (m of characteristic point 01, m 10). all characteristic points, according to the descending arrangement of θ value, are chosen to 8 characteristic points that conduct is chosen of θ maximum from (m/k) * (n/k) individual coordinate points.
Finally calculate the ORB characteristic of present frame: 8 characteristic points choosing are calculated according to following formula:
Figure BDA0000397539930000031
Figure BDA0000397539930000032
go here and there corresponding 10 system numbers as extracted ORB characteristic.
To be extractedly to after the ORB characteristic of present frame, the center position coordinate of continuous several frames is input in the fitting function in tracking module, the result drawing by fitting function predicts old man's next frame position, and the detailed process of prediction is:
A. in the tracking module in people's PC, be provided with three fitting functions, 1. fitting function is a beeline y=kx+b, and 2. fitting function is sine curve y=asin (cx-t)+kx+b, and 3. fitting function is parabola y=ax 2+ bx+c, by by the center position (x of continuous several frames 1, y 1), (x 2, y 2), (x 3, y 3) as parameter, be brought in three fitting functions, the parameter in three functions is all solved.Every 10 seconds, automatically upgrade the parameter of once fitting function.
B. the central point of establishing front cross frame is (x 1, y 1), (x 2, y 2), can calculate old man at the component v of x axle x=| x 1-x 2|/T, wherein T is the interval of two video pictures, and then can calculate next frame abscissa x 3estimated position be: x 3=2x 2-x 1, by by x 3be brought into fitting function y 3=f (x 3), can obtain three pre-judgement coordinates
Figure BDA0000397539930000033
wherein
Figure BDA0000397539930000034
for 1. corresponding ordinate of fitting function, wherein for 2. corresponding ordinate of fitting function, wherein
Figure BDA0000397539930000036
for 3. corresponding ordinate of fitting function.
C. for the priority of three coordinates according to
Figure BDA0000397539930000037
priority >
Figure BDA0000397539930000038
priority >
Figure BDA0000397539930000039
priority select, in the time of in 1. fitting function is predicted not, 1. fitting function is designated as miss, in the time of in 2. fitting function is predicted not, 2. fitting function is designated as miss, in the time of in 3. fitting function is predicted not, 3. fitting function is designated as miss, is allly marked as miss function, will no longer participate in selecting, when three functions are all marked as miss time, return to matching miss.
Judge that the method whether fitting function hit is as follows:
A. by the image of next frame to judge in advance centered by coordinate, take that the size of following the tracks of frame is border, extracts next frame and follow the tracks of the image in frame, and the dividing method of taking according to present frame cuts apart, be divided into (m/k) * (n/k) individual fritter.
B. according to next frame, follow the tracks of 8 characteristic points that frame extracts, use formula
Figure BDA0000397539930000041
C. by the ORB characteristic (64 strings of binary characters) of the ORB characteristic of present frame (64 strings of binary characters) and next frame, make comparisons, count the identical number of numerical value (0 or 1) on correspondence position, if be greater than a default numerical value, mean and follow the tracks of successfully, matching is hit.
Step 4, use the tracking module in individual PC, according to the above-mentioned old man who predicts, in the position of next frame, target old man is followed the tracks of, the detailed process of following the tracks of is that the tracking module in individual PC calculates ORB feature and the ORB characteristic that previous frame is followed the tracks of frame, and under present frame, follow the tracks of frame in ORB feature and the ORB characteristic of the position of prediction gained, and compare whether the element on same position in the two characteristic is identical, if identical number is more than or equal to threshold value, show to trace into target, retain current ORB feature and ORB characteristic, continue to process next frame, if identical number is less than threshold value, show that target is miss, need again to obtain next possible position and recalculate correspondence position ORB characteristic according to described approximating method, and relatively contrast with current ORB characteristic, until matching is hit, continue to process next frame, or all possible Fitting Coordinate System is all miss, this step returning tracking is lost.
If above-mentioned tracing process can trace into old man by continuous 10 frames, carry out frame-skipping operation,, every three frames performance objective track algorithm again, can greatly improve tracking efficiency like this; If above-mentioned tracing process returning tracking is lost, need to get back to before three frames, reuse above-mentioned tracing process and process, if followed the tracks of successfully, continue next frame; If followed the tracks of unsuccessfully, returning tracking is lost.
If returning tracking is lost, return to previous frame, the current old man that will follow the tracks of is carried out to 360 degree samplings around, and find the tracking frame of matching degree maximum, if the matching degree of this tracking frame is greater than threshold value, show that tracking target picks up after loss and huge change occurs in old man's path, call behavior recognizer old man's present case is identified; If the matching degree of this tracking frame is less than threshold value, show that old man follows the tracks of loss, delete current tracking frame, and call frame difference method and again detect the old man's motion conditions in guarded region.
The old man's that behavior identification module in step 5, individual PC traces into tracking module behavior is identified, if the behavior classification results identifying is normal behaviour, do not need to report to the police, if the behavior classification results identifying is abnormal behaviour, current recognition result need to be passed to display and warning audio amplifier is reported to the police, according to the difference of the abnormal behaviour classification results identifying, to display and warning audio amplifier, provide corresponding different signal.
When the old man's hazardous act monitoring method of first use home for destitute of the present invention, need to use individual PC to determine needed initial parameter in behavior identifying, initial parameter deterministic process is: first choose and have the sample of label χ={ χ 1, χ 2..., χ n, χ wherein ibe the video that a segment mark has label, the old man that label substance records for this section of video behavior in daily life, for example C 1for walking label, C 2for the label of standing, C 3for the label of falling, C 4for the label etc. of too bending over, using these videos as training sample, several training samples are taked in each labeling requirement under varying environment, and the video of each training sample has three dimensions, and image transverse axis, the image longitudinal axis, time shaft, input data χ iwei San rank tensor, then adopts the PCA based on tensor to learn these training samples, thereby obtains three transition matrix: U 1, U 2, U 3, then to three rank tensor: Y after each sample calculation dimensionality reduction i=χ i* 1u 1* 2u 2* 3u 3, by belonging to respectively walking, stand, fall, too bend over etc., the behavior of N kind is defined as C 1, C 2c nclass, the classification center of each class is: y wherein jfor belonging to C iclass sample χ jin tensor after all dimensionality reductions, for belonging to C ithe number of dvielement.The classification center of three transition matrixes that obtain and each class behavior is inputted to behavior identification module in individual PC as the parameter of behavior relative discern;
The step of the described PCA based on tensor is as follows:
A. for the video segment of input, can be understood as a three-dimensional matrice, this matrix has three dimensions, i.e. transverse axis x, and longitudinal axis y, time shaft t, establishing each video segment is X i, the video segment of all inputs is formed to four bit matrix X=[X 1, X 2..., X n]. random initializtion matrix U 1, U 2, U 3
B. compute matrix D 1=X * 2u 2* 3u 3, to matrix D 1* D 1 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 1, compute matrix D 2=X * 1u 1* 3u 3, to matrix D 2* D 2 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 2, compute matrix D 3=X * 1u 1* 2u 2, to matrix D 3* D 3 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 3, wherein, X * iu ifor the multiplication of tensor, its implication is by high dimension vector X, retains its i dimension, and other all dimensions are launched successively, forms a two-dimensional matrix X i, and calculate U i* X i, gained matrix carries out inverse transformation according to the order of launching again, the higher dimensional matrix of gained be X * iu ivalue.
C. calculate double iteration, gained matrix
Figure BDA0000397539930000053
with
Figure BDA0000397539930000054
poor:
Figure BDA0000397539930000055
Figure BDA0000397539930000056
with
Figure BDA0000397539930000057
three matrixes of gained, by all elements summed square, gained and be the variable quantity of twice iteration, if variable quantity is less than threshold value, iteration stops, current be the result of algorithm gained, otherwise foregoing b process.
Behavior relative discern process is: the abnormal behaviour of first autotracking algorithm in video image being found is as the time started point of behavior relative discern sample, centered by the central point in current tracking frame region, take current tracking frame longest edge as the length of side, thereby obtain a square area, by the video image in current region, read the monitor video in 3 seconds, as sample χ to be sorted, by formula Y=χ * 1u 1* 2u 2* 3u 3, U wherein 1, U 2, U 3for the transition matrix of initial parameter deterministic process gained, * i(i=1,2,3) are the oeprator that tensor multiplies each other, and obtain three rank tensor Y after this sample dimensionality reduction, by the classification center of Y and above-mentioned each class
Figure BDA0000397539930000061
on three dimensions, calculate its Euclidean distance, the nearest class of chosen distance is as its classification results, thereby complete classification, according to current behavior by classification gained result be recognition result, if recognition result is normal behaviours such as walking, stand, do not need to display and warning impression transmission of signal, if recognition result is for the abnormal behaviour such as falling, too bend over, signal corresponding to current recognition result need to be passed to display and warning audio amplifier.
On step 6, display, show in real time the picture of current monitoring, when receiving old man's abnormal behaviour classification results report, signal corresponding to behavior classification results that display provides according to behavior identification module, surrounding old man shows different color boxes flicker, and the audio amplifier of simultaneously reporting to the police sends corresponding alarm sound and reports to the police.
Compared with prior art the invention has the beneficial effects as follows:
1. the present invention has utilized target following and the advantage of two kinds of methods of behavior identification in supervisory control system fully, makes the two reach the effect of having complementary advantages; Target tracking module can provide the information such as movement objective orbit, speed, and can filter out a large amount of normal behaviours for behavior identification module, thereby has reduced the load of behavior identification module; Behavior identification module provides concrete analysis result for whole supervisory control system, thereby makes system can distinguish efficiently the situation of different behaviors movement locus of the same race.
2. target tracking module of the present invention adopts the target tracking algorism based on ORB feature, has improved the accuracy rate of following the tracks of; Adopt the pre-determination methods of behavior, made each frame only need once sampling, reduced the load of target tracking module.
3. the present invention, by adopting tracking, can effectively get rid of normal regular behavior, due in daily life, most behaviors are all normal, regular behavior, therefore, the present invention can reduce the number of run of behavior recognizer effectively.
Accompanying drawing explanation
Fig. 1. the schematic diagram forming for implementing monitoring system structure that a kind of home for destitute of the present invention old man's hazardous act monitoring method adopts;
Fig. 2. the allomeric function module frame figure of the computer program adopting for a kind of home for destitute of the present invention old man's hazardous act monitoring method;
Fig. 3. be the functional sequence block diagram of a kind of home for destitute of the present invention old man's hazardous act monitoring method;
Fig. 4. the FB(flow block) that judges in advance analytical method for the track algorithm that adopts in the old man's hazardous act monitoring method of a kind of home for destitute of the present invention and behavior;
Fig. 5. for being the schematic diagram of a kind of typical moving target movement locus that discloses of embodiment mono-in the old man's hazardous act monitoring method of a kind of home for destitute of the present invention;
Fig. 6. for being the schematic diagram of a kind of track of falling that embodiment bis-discloses in the old man's hazardous act monitoring method of a kind of home for destitute of the present invention;
In figure: 1, CCTV camera, 2, video frequency collection card, 3, individual PC, 4, display, 5, warning audio amplifier, 6, No. 1 movement locus, 7, No. 2 movement locus, 8, No. 3 movement locus, 9, No. 4 movement locus.
Embodiment
Below in conjunction with accompanying drawing, the present invention is explained in detail:
The basic principle of home for destitute of the present invention old man's hazardous act monitoring method is:
Referring to Fig. 1, the monitoring system that the present invention adopts comprises N CCTV camera 1, video frequency collection card 2, individual PC 3, display 4 and the warning audio amplifier 5 that structure is identical.Wherein, N is less than or equal to 48 for being more than or equal to 10, CCTV camera 1 is placed to the place of the daily processes of old man such as home for destitute corridor and outdoor indoor square, when at corridor, it is equidistant apart from side walls that video camera is placed to corridor ceiling, when being placed in indoor and outdoor square, video camera is placed to four jiaos of square, apart from ground about 3.5 meters, and guarantee to greatest extent when laying, the monitoring dead angle area of each CCTV camera 1 can be monitored by the identical CCTV camera 1 of other structures; The bnc interface of the CCTV camera 1 that N structure is identical is connected on the corresponding bnc interface of video frequency collection card 2, again the pci interface of video frequency collection card 2 is connected with corresponding pci interface on individual PC 3, make individual PC 3 can read the video image that all CCTV cameras 1 collect, and by the functional module of computer program of the present invention, vision signal is processed; Display 4 is by VGA interface (or DVI interface, the video image collecting for display monitoring video camera 1 and when finding that old man is abnormal, old man is marked and follows the tracks of frame and glimmer to report to the police with specific color is provided with the VGA interface (or DVI interface) of individual PC 3 interface providing depending on PC); Warning audio amplifier 5 is connected with the audio output interface of individual PC 3 by the audio interface of 3.5mm plug, for sounding and report to the police when discovery old man being abnormal.
The video frequency collection card 2 of considering current main flow is 4 tunnels, 8 Huo16 roads, road, a video frequency collection card can connect 4,8 or 16 CCTV cameras 1, an individual PC 3 has 3 PCI slots conventionally, therefore, the method can connect at most 16 * 3=48 platform CCTV camera 1, can meet the demand in middle-size and small-size home for destitute completely.
Consult Fig. 2. the functional module construction that is arranged on the computer program in individual PC 3 of the present invention consists of 5 modules, i.e. image capture module, profile detection module, tracking module, behavior identification module and display alarm module.
1) image capture module:
Consult Fig. 1, image capture module is the program module operating on individual PC 3, for controlling CCTV camera 1, video frequency collection card 2 reads in the monitoring image of guarded region, and monitoring image is passed to profile detection module.
2) profile detection module:
Profile detection module is the program module operating on individual PC 3, acting as of this profile detection module used frame difference method to detect in real time the old man's who enters guarded region profile, when finding that there is old man and enter into guarded region, detect old man's boundary information, and this information is passed to tracking module, tracking module needs to process in real time the video image that each CCTV camera 1 collects, and even current have a N CCTV camera 1, needs N profile detection module.In actual motion, due to the time with respect to the every frame period of video camera, human motion speed is slow, therefore do not need in real time every frame of each CCTV camera to be carried out to profile detection, therefore, the image that the present invention collects each CCTV camera 1 carried out a profile and detects every 1 second, in order to disperse the load to system, to the N of a current active CCTV camera, adopt the mode of rotation to process: within a current second, the 1st profile detection module moved at once, second profile detection module moved after second at the 1st the postrun 1/N of module, N profile detection module moved after second at the 1st postrun (the N-1)/N of module.
3) tracking module:
Tracking module is the program module operating on individual PC 3, acting as of this tracking module receives the boundary information that profile detection module collects, and according to these information, moving object in guarded region is followed the tracks of, and according to old man's behavior track, the possible position of current old man is judged in advance, when finding that the result of tracking results with pre-judgement is not inconsistent, current behavior is considered to suspicious actions, need to call behavior identification module and carry out behavior identification; Otherwise continue next frame image to follow the tracks of.Each tracking module is corresponding with each profile detection module, when a certain profile detection module finds that there is old man and enters in guarded region, corresponding tracking module brings into operation with it, when a certain profile detection module finds that all old men move out guarded region, corresponding tracking module is out of service with it.
4) behavior identification module
Behavior identification module is the program module operating on individual PC 3, and acting as of this program module receives old man's suspicious actions that tracking module is found, current suspicious actions are identified.This module only exists one in system, and any tracking module all can call behavior identification module.
5) display alarm module
Display alarm module is the program module operating on individual PC 3.The acting as of this display alarm module shows monitoring image that all CCTV cameras collect and the operation interface of some hommizations on display 4; And the recognition result of behavior identification module can be carried out to classification according to the order of severity; When needs are reported to the police, according to different ranks, with the tracking frame of different colours, by warning old man, glimmered with it, and by warning audio amplifier 5 alarm sounds, sending the corresponding alarm sound of different stage and report to the police.
Consult Fig. 3, the step of a kind of home for destitute of the present invention old man's hazardous act monitoring method is as follows:
Step 1, usage monitoring video camera 1, video frequency collection card 2, individual PC 3 obtain the old man's who enters guarded region video image;
Step 2, use the profile detection module in individual PC 3 to carry out profile detection to the video image obtaining in step 1, the number of pixels of calculating old man's contour area of gained by frame difference method changes to judge the current residing state of old man.Use frame difference method, to entering the old man of guarded region, carry out profile detection, two continuous frames is monitored to picture X 1, X 2according to its corresponding gray value, subtract each other, i.e. X 1(i, j)-X 2the profile that (i, j) ≥δ region is old man, X 1(i, j)-X 2(i, j) background area that < δ is old man, calculates the number of pixel in the corresponding connected region of profile, and the number of pixel is when becoming less many, for old man enters guarded region, now do not need profile information to pass to tracking module; When the number of pixel no longer continues to increase, can judge that old man has entered guarded region, now old man's profile information is passed to tracking module, this module is every operation in 1 second once.
Tracking module in step 3, individual PC 3 is followed the tracks of the old man's who obtains profile information, the rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and the ORB feature of extracting in tracking frame is local feature description's subcharacter of rotational invariance, and ORB characteristic is local feature description's subcharacter number of rotational invariance, the center position coordinate of continuous several frames is input in the fitting function in tracking module, and the result drawing by fitting function is predicted old man's next frame position; Detailed process is as follows:
The rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and ORB feature (the English full name: Oriented FAST and Rotated BRIEF in frame is followed the tracks of in extraction, local feature description's subcharacter of rotational invariance) and ORB characteristic (local feature description's subcharacter number of rotational invariance), extracting present frame, to follow the tracks of the detailed process of the ORB feature in frame as follows:
First present frame is carried out to piecemeal: the image in current tracking frame is extracted separately, with I, represent this picture, if image is the matrix of a m * n dimension, the fritter that image is divided to several k * k pixels, like this, whole picture is divided into altogether (m/k) * (n/k) individual fritter, the pixel value that the mean value of getting the pixel of each fritter is this fritter, I (x, y) be expressed as x be listed as y capable the average gray value of corresponding fritter.
Then select the ORB characteristic point of present frame: according to m pq=∑ x,yx py pi (x, y), p wherein, q respectively value be 0 and 1, x ∈ [1 ..., m/k], y ∈ [1 ..., n/k] and obtain gradient θ=argtan (m of characteristic point 01, m 10). all characteristic points, according to the descending arrangement of θ value, are chosen to 8 characteristic points that conduct is chosen of θ maximum from (m/k) * (n/k) individual coordinate points.
Finally calculate the ORB characteristic of present frame: 8 characteristic points choosing are calculated according to following formula:
Figure BDA0000397539930000102
go here and there corresponding 10 system numbers as extracted ORB characteristic.
To be extractedly to after the ORB characteristic of present frame, the center position coordinate of continuous several frames is input in the fitting function in tracking module, the result drawing by fitting function predicts old man's next frame position, and the detailed process of prediction is:
A. in the tracking module in people's PC 3, be provided with three fitting functions, 1. fitting function is a beeline y=kx+b, and 2. fitting function is sine curve y=asin (cx-t)+kx+b, and 3. fitting function is parabola y=ax 2+ bx+c, by by the center position (x of continuous several frames 1, y 1), (x 2, y 2), (x 3, y 3) as parameter, be brought in three fitting functions, the parameter in three functions is all solved.Every 10 seconds, automatically upgrade the parameter of once fitting function.
B. the central point of establishing front cross frame is (x 1, y 1), (x 2, y 2), can calculate old man at the component v of x axle x=| x 1-x 2|/T, wherein T is the interval of two video pictures, and then can calculate next frame abscissa x 3estimated position be:
Figure BDA0000397539930000103
by by x 3be brought into fitting function y 3=f (x 3), can obtain three pre-judgement coordinates
Figure BDA0000397539930000104
wherein
Figure BDA0000397539930000105
for 1. corresponding ordinate of fitting function, wherein
Figure BDA0000397539930000106
for 2. corresponding ordinate of fitting function, wherein
Figure BDA0000397539930000107
for 3. corresponding ordinate of fitting function.
C. for the priority of three coordinates according to
Figure BDA0000397539930000108
priority >
Figure BDA0000397539930000109
priority >
Figure BDA00003975399300001010
priority select, in the time of in 1. fitting function is predicted not, 1. fitting function is designated as miss, in the time of in 2. fitting function is predicted not, 2. fitting function is designated as miss, in the time of in 3. fitting function is predicted not, 3. fitting function is designated as miss, is allly marked as miss function, will no longer participate in selecting, when three functions are all marked as miss time, return to matching miss.
The described method that judges whether fitting function hit is as follows:
A. by the image of next frame to judge in advance centered by coordinate, take that the size of following the tracks of frame is border, extracts next frame and follow the tracks of the image in frame, and the dividing method of taking according to present frame cuts apart, be divided into (m/k) * (n/k) individual fritter.
B. according to next frame, follow the tracks of 8 characteristic points that frame extracts, use formula
Figure BDA0000397539930000111
C. by the ORB characteristic (64 strings of binary characters) of the ORB characteristic of present frame (64 strings of binary characters) and next frame, make comparisons, count the identical number of numerical value (0 or 1) on correspondence position, if be greater than a default numerical value, mean and follow the tracks of successfully, matching is hit.
Step 4, use the tracking module in individual PC 3, according to the above-mentioned old man who predicts, in the position of next frame, target old man is followed the tracks of, the detailed process of following the tracks of is that the tracking module in individual PC calculates ORB feature and the ORB characteristic that previous frame is followed the tracks of frame, and under present frame, follow the tracks of frame in ORB feature and the ORB characteristic of the position of prediction gained, and compare whether the element on same position in the two characteristic is identical, if identical number is more than or equal to threshold value, show to trace into target, retain current ORB feature and ORB characteristic, continue to process next frame, if identical number is less than threshold value, show that target is miss, need again to obtain next possible position and recalculate correspondence position ORB characteristic according to described approximating method, and relatively contrast with current ORB characteristic, until matching is hit, continue to process next frame, or all possible Fitting Coordinate System is all miss, this step returning tracking is lost.
If above-mentioned tracing process can trace into old man by continuous 10 frames, carry out frame-skipping operation,, every three frames performance objective track algorithm again, can greatly improve tracking efficiency like this; If above-mentioned tracing process returning tracking is lost, need to get back to before three frames, reuse above-mentioned tracing process and process, if followed the tracks of successfully, continue next frame; If followed the tracks of unsuccessfully, returning tracking is lost.
If returning tracking is lost, return to previous frame, the current old man that will follow the tracks of is carried out to 360 degree samplings around, and find the tracking frame of matching degree maximum, if the matching degree of this tracking frame is greater than threshold value, show that tracking target picks up after loss and huge change occurs in old man's path, call behavior recognizer old man's present case is identified; If the matching degree of this tracking frame is less than threshold value, show that old man follows the tracks of loss, delete current tracking frame, and call frame difference method and again detect the old man's motion conditions in guarded region.
The old man's that behavior identification module in step 5, individual PC 3 traces into tracking module behavior is identified, if the behavior classification results identifying is normal behaviour, do not need to report to the police, if the behavior classification results identifying is abnormal behaviour, current recognition result need to be passed to display 4 and warning audio amplifier 5 is reported to the police, according to the difference of the abnormal behaviour classification results identifying, to display 4 and warning audio amplifier 5, provide corresponding different signal.
When first use this method, need to use needed initial parameter in 4 pairs of behavior identifyings of individual PC to determine, initial parameter deterministic process is: first choose and have the sample of label χ={ χ 1, χ 2..., χ n, χ wherein ibe the video that a segment mark has label, the old man that label substance records for this section of video behavior in daily life, for example C 1for walking label, C 2for the label of standing, C 3for the label of falling, C 4for the label etc. of too bending over, using these videos as training sample, several training samples are taked in each labeling requirement under varying environment, and the video of each training sample has three dimensions, and image transverse axis, the image longitudinal axis, time shaft, input data χ iwei San rank tensor, then adopts the PCA based on tensor to learn these training samples, thereby obtains three transition matrix: U 1, U 2, U 3, then to three rank tensor: Y after each sample calculation dimensionality reduction i=χ i* 1u 1* 2u 2* 3u 3, by belonging to respectively walking, stand, fall, too bend over etc., the behavior of N kind is defined as C 1, C 2c nclass, the classification center of each class is:
Figure BDA0000397539930000121
y wherein jfor belonging to C iclass sample χ jin tensor after all dimensionality reductions,
Figure BDA0000397539930000122
for belonging to C ithe number of dvielement.The classification center of three transition matrixes that obtain and each class behavior is inputted to behavior identification module in individual PC 3 as the parameter of behavior relative discern;
The step of the described PCA based on tensor is as follows:
A. for the video segment of input, can be understood as a three-dimensional matrice, this matrix has three dimensions, i.e. transverse axis x, and longitudinal axis y, time shaft t, establishing each video segment is X i, the video segment of all inputs is formed to four bit matrix X=[X 1, X 2..., X n]. random initializtion matrix U 1, U 2, U 3
B. compute matrix D 1=X * 2u 2* 3u 3, to matrix D 1* D 1 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 1, compute matrix D 2=X * 1u 1* 3u 3, to matrix D 2* D 2 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 2, compute matrix D 3=X * 1u 1* 2u 2, to matrix D 3* D 3 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 3, wherein, X * iu ifor the multiplication of tensor, its implication is by high dimension vector X, retains its i dimension, and other all dimensions are launched successively, forms a two-dimensional matrix X i, and calculate U i* X i, gained matrix carries out inverse transformation according to the order of launching again, the higher dimensional matrix of gained be X * iu ivalue.
C. calculate double iteration, gained matrix
Figure BDA0000397539930000123
with poor:
Figure BDA0000397539930000125
Figure BDA0000397539930000126
with three matrixes of gained, by all elements summed square, gained and be the variable quantity of twice iteration, if variable quantity is less than threshold value, iteration stops, current
Figure BDA0000397539930000128
be the result of algorithm gained, otherwise foregoing b process.
Behavior relative discern process is: the abnormal behaviour of first autotracking algorithm in video image being found is as the time started point of behavior relative discern sample, centered by the central point in current tracking frame region, take current tracking frame longest edge as the length of side, thereby obtain a square area, by the video image in current region, read the monitor video in 3 seconds, as sample χ to be sorted, by formula Y=χ * 1u 1* 2u 2* 3u 3, U wherein 1, U 2, U 3for the transition matrix of initial parameter deterministic process gained, * i(i=1,2,3) are the oeprator that tensor multiplies each other, and obtain three rank tensor Y after this sample dimensionality reduction, by the classification center of Y and above-mentioned each class on three dimensions, calculate its Euclidean distance, the nearest class of chosen distance is as its classification results, thereby complete classification, according to current behavior by classification gained result be recognition result, if recognition result is normal behaviours such as walking, stand, do not need to display 4 and warning audio amplifier 5 transmission of signals, if recognition result is for the abnormal behaviour such as falling, too bend over, signal corresponding to current recognition result need to be passed to display 4 and warning audio amplifier 5.
On step 6, display 4, show in real time the picture of current monitoring, when receiving old man's abnormal behaviour classification results report, signal corresponding to behavior classification results that display 4 provides according to behavior identification module, surrounding old man shows different color boxes flicker, and the audio amplifier 5 of simultaneously reporting to the police sends corresponding alarm sound and reports to the police.
Provide two embodiments to describe the detailed process of old man's hazardous act method for supervising in home for destitute of the present invention below:
One, the present embodiment is for being used the daily behavior of old man of camera head monitor under the environment of home for destitute.In the present embodiment, we have simulated the normal walking states of old man, and detailed process is as follows:
Step 1, image capture module are controlled CCTV camera acquisition monitoring image:
CCTV camera 1 passes to the monitoring image of the old man's activity in the guarded region of collection in video frequency collection card 2, through video frequency collection card 2, the profile detection module being transferred to after image compression in individual PC 3 interior operations is processed.
Step 2, profile detection module are used frame difference method to carry out profile detection to old man:
Profile detection module is processed the monitoring image being passed over by image capture module.Use frame difference method, to entering the old man of guarded region, carry out profile detection, two continuous frames is monitored to picture X 1, X 2according to its corresponding gray value, subtract each other, i.e. X 1(i, j)-X 2the profile that (i, j) ≥δ region is old man, X 1(i, j)-X 2(i, j) background area that < δ is old man, δ is a default threshold value, in the present embodiment, establish δ=50, calculate the number of old man's the interior pixel of the corresponding connected region of profile, the number of pixel when becoming less many, for old man enters guarded region, does not now need profile information to pass to tracking module; When the number of pixel no longer continues to increase, can judge that old man has entered guarded region, now old man's profile information is passed to tracking module, tracking module is every operation in 1 second once.
Step 3, tracking module prediction old man's next frame position:
Consult Fig. 5, as shown in FIG., old man first directly walks the movement locus that old man goes down in normal row, turns, more directly walks, until walk out guarded region.First old man enters into guarded region, and is detected by profile detection module, and tracking module obtains after old man's profile information, starts old man's movement locus to follow the tracks of.First, old man directly walks in guarded region, and now, the direction that old man advances is straight line, as shown in No. 1 movement locus 6 in figure.
First present frame is carried out to piecemeal: the image in current tracking frame is extracted separately, with I, represent this picture, if image is the matrix of a m * n dimension, the fritter that image is divided to several k * k pixels, like this, whole picture is divided into altogether (m/k) * (n/k) individual fritter, the pixel value that the mean value of getting the pixel of each fritter is this fritter, I (x, y) be expressed as x be listed as y capable the average gray value of corresponding fritter.
Then select the ORB characteristic point of present frame: according to m pq=∑ x,yx py pi (x, y), p wherein, q respectively value be 0 and 1, x ∈ [1 ..., m/k], y ∈ [1 ..., n/k] and obtain gradient θ=argtan (m of characteristic point 01, m 10). all characteristic points, according to the descending arrangement of θ value, are chosen to 8 characteristic points that conduct is chosen of θ maximum from (m/k) * (n/k) individual coordinate points.
Finally calculate the ORB characteristic of present frame: 8 characteristic points choosing are calculated according to following formula:
Figure BDA0000397539930000141
Figure BDA0000397539930000142
go here and there corresponding 10 system numbers as extracted ORB characteristic.
To be extractedly to after the ORB characteristic of present frame, the center position coordinate of continuous several frames is input in the fitting function in tracking module, the result drawing by fitting function predicts old man's next frame position, and the detailed process of prediction is:
A. in the tracking module in people's PC, be provided with three fitting functions, 1. fitting function is a beeline y=kx+b, and 2. fitting function is sine curve y=asin (cx-t)+kx+b, and 3. fitting function is parabola y=ax 2+ bx+c, by by the center position (x of continuous several frames 1, y 1), (x 2, y 2), (x 3, y 3) as parameter, be brought in three fitting functions, the parameter in three functions is all solved.Every 10 seconds, automatically upgrade the parameter of once fitting function.
B. the central point of establishing front cross frame is (x 1, y 1), (x 2, y 2), can calculate old man at the component v of x axle x=| x 1-x 2|/T, wherein T is the interval of two video pictures, and then can calculate next frame abscissa x 3estimated position be: x 3=2x 2-x 1, by by x 3be brought into fitting function y 3=f (x 3), can obtain three pre-judgement coordinates
Figure BDA0000397539930000143
wherein
Figure BDA0000397539930000144
for 1. corresponding ordinate of fitting function, wherein for 2. corresponding ordinate of fitting function, wherein for 3. corresponding ordinate of fitting function.
C. for the priority of three coordinates according to priority >
Figure BDA0000397539930000148
priority >
Figure BDA0000397539930000149
priority select, in the time of in 1. fitting function is predicted not, 1. fitting function is designated as miss, in the time of in 2. fitting function is predicted not, 2. fitting function is designated as miss, in the time of in 3. fitting function is predicted not, 3. fitting function is designated as miss, is allly marked as miss function, will no longer participate in selecting, when three functions are all marked as miss time, return to matching miss.
Judge that the method whether fitting function hit is as follows:
A. by the image of next frame to judge in advance centered by coordinate, take that the size of following the tracks of frame is border, extracts next frame and follow the tracks of the image in frame, and the dividing method of taking according to present frame cuts apart, be divided into (m/k) * (n/k) individual fritter.
B. according to next frame, follow the tracks of 8 characteristic points that frame extracts, use formula
Figure BDA0000397539930000151
C. by the ORB characteristic (64 strings of binary characters) of the ORB characteristic of present frame (64 strings of binary characters) and next frame, make comparisons, count the identical number of numerical value (0 or 1) on correspondence position, if be greater than a default numerical value, mean and follow the tracks of successfully, matching is hit.
Follow the tracks of target old man the position of old man's next frame that step 4, tracking module are predicted according to step 3:
Use the tracking module in individual PC 3, according to the above-mentioned old man who predicts, in the position of next frame, target old man is followed the tracks of, the detailed process of following the tracks of is that the tracking module in individual PC 3 calculates ORB feature and the ORB characteristic that previous frame is followed the tracks of frame, and under present frame, follow the tracks of frame in ORB feature and the ORB characteristic of the position of prediction gained, and compare whether the element on same position in the two characteristic is identical, if identical number is more than or equal to threshold value, show to trace into target, retain current ORB feature and ORB characteristic, continue to process next frame, if identical number is less than threshold value, show that target is miss, need again to obtain next possible position and recalculate correspondence position ORB characteristic according to described approximating method, and relatively contrast with current ORB characteristic, until matching is hit, continue to process next frame, or all possible Fitting Coordinate System is all miss, this step returning tracking is lost.Now, due to old man's straight line of going on along, so use fitting function y=kx+b just to find easily old man, if above-mentioned tracing process can trace into old man by continuous 10 frames, carry out frame-skipping operation,, every three frames performance objective track algorithm again, can greatly improve tracking efficiency like this.
When old man has moved to the end of No. 1 movement locus 6 in figure, when preparation enters into the initiating terminal of No. 2 movement locus 7, movement locus changes, now, fitting function y=kx+b is miss, this function is labeled as miss, replacing fitting function is: y=asin (cx-t)+kx+b, and whether the frequency that judges this SIN function is abnormal, when old man turns, the frequency of SIN function is excessive, surpasses the scope of rocking of the normal walking of people, so give up just profound function, and to change fitting function be y=ax 2+ bx+c, notices that old man's movement locus when turning is parabola, as No. 2 movement locus 7 in figure, therefore, and in the process old man through the track shown in movement locus 7, fitting function y=ax 2+ bx+c can effectively predict next step position of old man.
When old man moves to the end of No. 2 movement locus 7 in figure, when preparing to enter into the initiating terminal of No. 3 movement locus 8, there is variation in movement locus, now, and fitting function y=ax 2+ bx+c is miss, considers that now all fitting functions have all been marked as miss, therefore, using this frame as start frame, the behavior recognizer of calling is identified, and recognition result is for walking, for normal behaviour, now, by three fitting function cancel all marks.Reselect y=kx+b and carry out matching.
Notice old man in the drawings the movement locus in region shown in No. 3 movement locus 8 be straight line, therefore, the present invention can effectively dope old man in the position of next frame.When old man moves to the border of guarded region, the area of following the tracks of frame diminishes.Because the situation of losing is followed the tracks of in as easy as rolling off a log appearance at this moment, and at this moment old man is followed the tracks of and behavior identification nonsensical, when occurring following the tracks of when losing, this algorithm is deleted and is followed the tracks of frame, calls frame difference method and detects.While finding to follow the tracks of the situation that frame area reduces and a lateral boundaries is fixing, can conclude that this old man is walking out tracked region, frame difference method does not follow the tracks of this of frame as parameter call track algorithm, but continues monitoring, until old man walks out tracing area.
In the present embodiment, the present invention only needs No. 1 movement locus 6 in the drawings, No. 2 movement locus 7, the starting point of No. 3 movement locus 8 occurs to change fitting function in the miss situation of some fitting functions, thereby and the process that enters into movement locus 8 from movement locus 7 because all fitting functions are all marked as the miss behavior recognizer of need to calling identify, visible, this algorithm is when monitoring normal behaviour, can get rid of most normal behaviours, improve the operational efficiency of system.
In actual use, because old man's movement locus may not be than more complete straight line, especially adopt every three frames again after the processing of performance objective track algorithm, easily there is following the tracks of Loss, if above-mentioned tracing process returning tracking is lost, need to get back to before three frames, reuse above-mentioned tracing process and process, if followed the tracks of successfully, continue next frame; If followed the tracks of unsuccessfully, returning tracking is lost.
If returning tracking is lost, return to previous frame, the current old man that will follow the tracks of is carried out to 360 degree samplings around, and find the tracking frame of matching degree maximum, if the matching degree of this tracking frame is greater than threshold value, show that tracking target picks up after loss and huge change occurs in old man's path, call behavior recognizer old man's present case is identified; If the matching degree of this tracking frame is less than threshold value, show that old man follows the tracks of loss, delete current tracking frame, and call frame difference method and again detect the old man's motion conditions in guarded region.
Step 5, behavior identification module are identified old man's suspicious actions:
1) parameter initialization process:
In the present embodiment, behavior is divided into 6 classes: C 1class is walking, C 2class is for standing, C 3class is for falling, C 4class is for too bending over, C 5class is for impacting mutually, C 6class is overlapping and the behaviors that cannot identification of two people.
A. by gathering old man at the video sample of above-mentioned 6 behaviors in actual life, the video segment for input, can be understood as a three-dimensional matrice, and this matrix has three dimensions, i.e. transverse axis x, and longitudinal axis y, time shaft t, establishing each video segment is X i, the video segment of all inputs is formed to four bit matrix X=[X 1, X 2..., X n]. random initializtion matrix U 1, U 2, U 3.
B. compute matrix D 1=X * 2u 2* 3u 3, to matrix D 1* D 1 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 1, compute matrix D 2=X * 1u 1* 3u 3, to matrix D 2* D 2 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 2, compute matrix D 3=X * 1u 1* 2u 2, to matrix D 3* D 3 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 3, wherein, X * iu ifor the multiplication of tensor, its implication is by high dimension vector X, retains its i dimension, and other all dimensions are launched successively, forms a two-dimensional matrix X i, and calculate U i* X i, gained matrix carries out inverse transformation according to the order of launching again, the higher dimensional matrix of gained be X * iu ivalue.
C. calculate double iteration, gained matrix
Figure BDA0000397539930000171
with
Figure BDA0000397539930000172
poor:
Figure BDA0000397539930000173
Figure BDA0000397539930000174
with three matrixes of gained, by all elements summed square, gained and be the variable quantity of twice iteration, if variable quantity is less than threshold value, iteration stops, current
Figure BDA0000397539930000176
be the result of algorithm gained, otherwise repeat B step.
By said method, obtained transition matrix U 1, U 2, U 3, by calculate the classification center of each class
Figure BDA0000397539930000178
the data of parameter initialization process gained are written in system when system is dispatched from the factory, and for behavior identifying, call at any time.
2) old man's normal row being walked to behavior identifies
When tracking module is found suspicious actions, and while calling behavior identification module, behavior identification module be take current time frame as starting point, centered by the central point in current tracking frame region, take current tracking frame longest edge as the length of side, thereby obtain a square area, the video image in current region is read to 3 seconds, thereby obtain one section of video image, this video has three dimensions: image transverse axis, the image longitudinal axis, time shaft, Ji San rank tensor χ, by formula Y=X * 1u 1* 2u 2* 3u 3(wherein, X is input video, U 1, U 2, U 3for the transition matrix of parameter initialization process gained, * i(i=1,2,3) are oeprator, obtain the tensor Y after dimensionality reduction, by calculating Y and six classification center
Figure BDA0000397539930000179
distance, i.e. Euclidean distance between higher-dimension array, the C that chosen distance is nearest iclass is as classification results.
In the present embodiment, old man is when moving track significant change, and tracking module calls behavior identification module and identifies.The classification results of behavior identification module is for belonging to C iclass, is normal behaviour, does not need to report to the police, and the result of identification must be returned to tracking module.
Step 6, display alarm module are reported to the police to the old man's abnormal behaviour identifying
In the present embodiment, old man's behavior is normal behaviour, therefore, in display 4 screens, shows current monitoring image, does not do any warning and processes.
Two, our kind of representative abnormal behaviour of having selected to fall in the present embodiment, the implementation process of the present embodiment is as follows:
Step 1, use image capture module are controlled CCTV camera acquisition monitoring image:
CCTV camera 1 passes to the monitoring image of the old man's activity in the guarded region of collection in video frequency collection card 2, through video frequency collection card 2, the profile detection module being transferred to after image compression in individual PC 3 interior operations is processed.
Step 2, use profile detection module are carried out profile detection to old man:
Profile detection module is processed the monitoring image being passed over by image capture module.Use frame difference method, to entering the old man of guarded region, carry out profile detection, two continuous frames is monitored to picture X 1, X 2according to its corresponding gray value, subtract each other, i.e. X 1(i, j)-X 2the profile that (i, j) ≥δ region is old man, X 1(i, j)-X 2(i, j) background area that < δ is old man, δ is a default threshold value, in the present embodiment, establish δ=50, calculate the number of old man's the interior pixel of the corresponding connected region of profile, the number of pixel when becoming less many, for old man enters guarded region, does not now need profile information to pass to tracking module; When the number of pixel no longer continues to increase, can judge that old man has entered guarded region, now old man's profile information is passed to tracking module, tracking module is every operation in 1 second once.
Step 3, use tracking module prediction old man's next frame position:
The front and back that old man is falling, there is obvious variation in its movement locus, in reality, the movement locus that old man is falling, consult shown in the head of the movement locus 9 shown in Fig. 6, it is a flex point of slightly swinging to a side, before old man is falling, running orbit is straight line, first present frame is carried out to piecemeal: the image in current tracking frame is extracted separately, with I, represent this picture, if image is the matrix of a m * n dimension, the fritter that image is divided to several k * k pixels, like this, whole picture is divided into altogether (m/k) * (n/k) individual fritter, the pixel value that the mean value of getting the pixel of each fritter is this fritter, I (x, y) be expressed as x be listed as y capable the average gray value of corresponding fritter.
Then select the ORB characteristic point of present frame: according to m pq=∑ x,yx py pi (x, y), p wherein, q respectively value be 0 and 1, x ∈ [1 ..., m/k], y ∈ [1 ..., n/k] and obtain gradient θ=argtan (m of characteristic point 01, m 10). all characteristic points, according to the descending arrangement of θ value, are chosen to 8 characteristic points that conduct is chosen of θ maximum from (m/k) * (n/k) individual coordinate points.
Finally calculate the ORB characteristic of present frame: 8 characteristic points choosing are calculated according to following formula:
Figure BDA0000397539930000192
go here and there corresponding 10 system numbers as extracted ORB characteristic.
To be extractedly to after the ORB characteristic of present frame, the center position coordinate of continuous several frames is input in the fitting function in tracking module, the result drawing by fitting function predicts old man's next frame position, and the detailed process of prediction is:
A. in the tracking module in people's PC 3, be provided with three fitting functions, 1. fitting function is a beeline y=kx+b, and 2. fitting function is sine curve y=asin (cx-t)+kx+b, and 3. fitting function is parabola y=ax 2+ bx+c, by by the center position (x of continuous several frames 1, y 1), (x 2, y 2), (x 3, y 3) as parameter, be brought in three fitting functions, the parameter in three functions is all solved.Every 10 seconds, automatically upgrade the parameter of once fitting function.
B. the central point of establishing front cross frame is (x 1, y 1), (x 2, y 2), can calculate old man at the component v of x axle x=| x 1-x 2|/T, wherein T is the interval of two video pictures, and then can calculate next frame abscissa x 3estimated position be: x 3=2x 2-x 1, by by x 3be brought into fitting function y 3=f (x 3), can obtain three pre-judgement coordinates
Figure BDA0000397539930000193
wherein
Figure BDA0000397539930000194
for 1. corresponding ordinate of fitting function, wherein
Figure BDA0000397539930000195
for 2. corresponding ordinate of fitting function, wherein
Figure BDA0000397539930000196
for 3. corresponding ordinate of fitting function.
C. for the priority of three coordinates according to
Figure BDA0000397539930000197
priority >
Figure BDA0000397539930000198
priority >
Figure BDA0000397539930000199
priority select, in the time of in 1. fitting function is predicted not, 1. fitting function is designated as miss, in the time of in 2. fitting function is predicted not, 2. fitting function is designated as miss, in the time of in 3. fitting function is predicted not, 3. fitting function is designated as miss, is allly marked as miss function, will no longer participate in selecting, when three functions are all marked as miss time, return to matching miss.
Judge that the method whether fitting function hit is as follows:
A. by the image of next frame to judge in advance centered by coordinate, take that the size of following the tracks of frame is border, extracts next frame and follow the tracks of the image in frame, and the dividing method of taking according to present frame cuts apart, be divided into (m/k) * (n/k) individual fritter.
B. according to next frame, follow the tracks of 8 characteristic points that frame extracts, use formula
Figure BDA00003975399300001910
C. by the ORB characteristic (64 strings of binary characters) of the ORB characteristic of present frame (64 strings of binary characters) and next frame, make comparisons, count the identical number of numerical value (0 or 1) on correspondence position, if be greater than a default numerical value, mean and follow the tracks of successfully, matching is hit.
Follow the tracks of target old man the position of old man's next frame that step 4, use tracking module are predicted according to step 3:
Use the tracking module in individual PC 3, according to the above-mentioned old man who predicts, in the position of next frame, target old man is followed the tracks of, the detailed process of following the tracks of is that the tracking module in individual PC 3 calculates ORB feature and the ORB characteristic that previous frame is followed the tracks of frame, and under present frame, follow the tracks of frame in ORB feature and the ORB characteristic of the position of prediction gained, and compare whether the element on same position in the two characteristic is identical, if identical number is more than or equal to threshold value, show to trace into target, retain current ORB feature and ORB characteristic, continue to process next frame, if identical number is less than threshold value, show that target is miss, need again to obtain next possible position and recalculate correspondence position ORB characteristic according to described approximating method, and relatively contrast with current ORB characteristic, until matching is hit, continue to process next frame, or all possible Fitting Coordinate System is all miss, this step returning tracking is lost.Now, due to old man's straight line of going on along, so use fitting function y=kx+b just to find easily old man, if above-mentioned tracing process can trace into old man by continuous 10 frames, carry out frame-skipping operation,, every three frames performance objective track algorithm again, can greatly improve tracking efficiency like this.
In scene due to Falls Among Old People, occurred speed and direction of motion variation issue clearly, matching is miss, needs that present image information is passed to behavior identification module and identifies.
Step 5, behavior identification module are identified old man's suspicious actions
1) parameter initialization process:
In the present embodiment, behavior is divided into 6 classes: C 1class is walking, C 2class is for standing, C 3class is for falling, C 4class is for too bending over, C 5class is for impacting mutually, C 6class is overlapping and the behaviors that cannot identification of two people.
A. by gathering old man at the video sample of above-mentioned 6 behaviors in actual life, the video segment for input, can be understood as a three-dimensional matrice, and this matrix has three dimensions, i.e. transverse axis x, and longitudinal axis y, time shaft t, establishing each video segment is X i, the video segment of all inputs is formed to four bit matrix X=[X 1, X 2..., X n]. random initializtion matrix U 1, U 2, U 3.
B. compute matrix D 1=X * 2u 2* 3u 3, to matrix D 1* D 1 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 1, compute matrix D 2=X * 1u 1* 3u 3, to matrix D 2* D 2 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 2, compute matrix D 3=X * 1u 1* 2u 2, to matrix D 3* D 3 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 3, wherein, X * iu ifor the multiplication of tensor, its implication is by high dimension vector X, retains its i dimension, and other all dimensions are launched successively, forms a two-dimensional matrix X i, and calculate U i* X i, gained matrix carries out inverse transformation according to the order of launching again, the higher dimensional matrix of gained be X * iu ivalue.
C. calculate double iteration, gained matrix
Figure BDA0000397539930000211
with
Figure BDA0000397539930000212
poor:
Figure BDA0000397539930000213
Figure BDA0000397539930000214
with
Figure BDA0000397539930000215
three matrixes of gained, by all elements summed square, gained and be the variable quantity of twice iteration, if variable quantity is less than threshold value, iteration stops, current
Figure BDA0000397539930000216
be the result of algorithm gained, otherwise repeat b step.
By said method, obtained transition matrix U 1, U 2, U 3, by
Figure BDA0000397539930000217
calculate the classification center of each class
Figure BDA0000397539930000218
the data of parameter initialization process gained are written in system when system is dispatched from the factory, and for behavior identifying, call at any time.
2) Falls Among Old People situation is carried out to behavior identification
When tracking module is found suspicious actions, and while calling behavior identification module, behavior identification module be take current time frame as starting point, centered by the central point in current tracking frame region, take current tracking frame longest edge as the length of side, thereby obtain a square area, the video image in current region is read to 3 seconds, thereby obtain one section of video image, this video has three dimensions: image transverse axis, the image longitudinal axis, time shaft, Ji San rank tensor χ, by formula Y=X * 1u 1* 2u 2* 3u 3(wherein, X is input video, U 1, U 2, U 3for the transition matrix of parameter initialization process gained, * i(i=1,2,3) are oeprator, obtain the tensor Y after dimensionality reduction, by calculating Y and six classification center
Figure BDA0000397539930000219
distance, i.e. Euclidean distance between higher-dimension array, the C that chosen distance is nearest iclass is as classification results.
In the present embodiment, old man is when moving track significant change, and tracking module calls behavior identification module and identifies.The classification results of behavior identification module is for belonging to C 3class is the behavior of falling, and needs to report to the police, and the result of identification must be returned to tracking module.
Step 6, display alarm module are reported to the police to the old man's abnormal behaviour identifying
In the present embodiment, the recognition result of old man's behavior is for belonging to C 3class is the behavior of falling, and needs to report to the police, and according to the predefined order of severity, reports to the police, the mode of reporting to the police and adopting image and sound to combine, follow the tracks of frame and show on the screen of display 4, and change with corresponding warning color and flicker, warning audio amplifier 5 sends corresponding sound.

Claims (5)

1.Yi Zhong home for destitute old man's hazardous act monitoring method, is characterized in that, the step of described a kind of home for destitute old man's hazardous act monitoring method is as follows:
Old man's monitoring image the monitoring image of collection is passed to video frequency collection card (2) in step 1, CCTV camera (1) acquisition monitoring region, video frequency collection card (2) is processed being transferred to after Surveillant Image Compression at individual PC (3);
Profile detection module in step 2, individual PC (3) is carried out profile detection to the monitoring image gathering, and the number of pixels of calculating old man's contour area by frame difference method changes to judge the residing state of old man;
Tracking module in step 3, individual PC (3) is followed the tracks of the old man's who obtains profile information, the rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and the ORB feature of extracting in tracking frame is local feature description's subcharacter of rotational invariance, and ORB characteristic is local feature description's subcharacter number of rotational invariance, the center position coordinate of continuous several frames is input in the fitting function in trace routine, and the result drawing by fitting function is predicted old man's next frame position;
Tracking module in step 4, individual PC (3), according to the old man that predicts in the position of next frame, target old man is followed the tracks of;
The old man's that behavior identification module in step 5, individual PC (3) is followed the tracks of tracking module behavior is identified, if the behavior classification results identifying is normal behaviour, do not need to report to the police, if the behavior classification results identifying is abnormal behaviour, current recognition result need to be passed to display (4) and warning audio amplifier (5) is reported to the police, and according to the difference of the abnormal behaviour classification results identifying, to display (4) and warning audio amplifier (5), provide different signals;
Step 6, the upper picture that shows in real time current monitoring of display (4), when receiving old man's abnormal behaviour classification results report, signal corresponding to behavior classification results that display (4) provides according to behavior identification module, surrounding old man shows different color boxes flicker, and the audio amplifier (5) of simultaneously reporting to the police sends corresponding alarm sound and reports to the police.
2. home for destitute according to claim 1 old man's hazardous act monitoring method, the detailed process that it is characterized in that the current status of judgement old man described in step 2 is: use frame difference method, to entering the old man of guarded region, carry out profile detection, two continuous frames is monitored to picture X 1, X 2according to its corresponding gray value, subtract each other, i.e. X 1(i, j)-X 2the profile that (i, j) < δ region is old man, X 1(i, j)-X 2(i, j) background area that < δ is old man, calculates the number of pixel in the corresponding connected region of profile, and the number of pixel is when becoming less many, for old man enters guarded region, now do not need profile information to pass to tracking module; When the number of pixel no longer continues to increase, can judge that old man has entered guarded region, now old man's profile information is passed to tracking module, this module is every operation in 1 second once.
3. home for destitute according to claim 1 old man's hazardous act monitoring method, is characterized in that the process of the prediction old man next frame position described in step 3 is:
The rectangular shaped rim surrounding according to old man's profile information is as following the tracks of frame, and extracts ORB feature and the ORB characteristic of following the tracks of in frame, and extracting present frame, to follow the tracks of the detailed process of ORB feature in frame and ORB characteristic as follows:
First present frame is carried out to piecemeal: the image in current tracking frame is extracted separately, with I, represent this picture, if image is the matrix of a m * n dimension, the fritter that image is divided to several k * k pixels, like this, whole picture is divided into altogether (m/k) * (n/k) individual fritter, the pixel value that the mean value of getting the pixel of each fritter is this fritter, I (x, y) be expressed as x be listed as y capable the average gray value of corresponding fritter.
Then select the ORB characteristic point of present frame: according to m pq=∑ x,yx py pi (x, y), p wherein, q respectively value be 0 and 1, x ∈ [1 ..., m/k], y ∈ [1 ..., n/k] and obtain gradient θ=argtan (m of characteristic point 01, m 10). all characteristic points, according to the descending arrangement of θ value, are chosen to 8 characteristic points that conduct is chosen of θ maximum from (m/k) * (n/k) individual coordinate points.
Finally calculate the ORB characteristic of present frame: 8 characteristic points choosing are calculated according to following formula:
&tau; ( I , x 1 , y 1 , x 2 , y 2 ) = 1 : I ( x 1 , y 1 ) < I ( x 2 , y 2 ) 0 : I ( x 1 , y 1 , ) &GreaterEqual; I ( x 2 , y 2 )
(x wherein i, y i) be respectively 8 characteristic points selected under B step, by formula
Figure FDA0000397539920000022
the corresponding 10 system numbers of symbol string are as extracted ORB characteristic.
To be extractedly to after the ORB characteristic of present frame, the center position coordinate of continuous several frames is input in the fitting function in trace routine, the result drawing by fitting function predicts old man's next frame position, and the detailed process of prediction is:
A. in the trace routine in people's PC (3), be provided with three fitting functions, 1. fitting function is a beeline y=kx+b, and 2. fitting function is sine curve y=asin (cx-t)+kx+b, and 3. fitting function is parabola y=ax 2+ bx+c, by by the center position (x of continuous several frames 1, y 1), (x 2, y 2), (x 3, y 3) as parameter, be brought in three fitting functions, the parameter in three functions is all solved.Every 10 seconds, automatically upgrade the parameter of once fitting function.
B. the central point of establishing front cross frame is (x 1, y 1), (x 2, y 2), can calculate old man at the component v of x axle x=| x 1-x 2|/T, wherein T is the interval of two video pictures, and then can calculate next frame abscissa x 3estimated position be: x 3=2x 2-x 1, by by x 3be brought into fitting function y 3=f (x 3), can obtain three pre-judgement coordinate (x 3,
Figure FDA0000397539920000032
), (x 3,
Figure FDA0000397539920000033
), (x 3,
Figure FDA0000397539920000034
), wherein
Figure FDA0000397539920000035
for 1. corresponding ordinate of fitting function, wherein
Figure FDA0000397539920000036
for 2. corresponding ordinate of fitting function, wherein
Figure FDA0000397539920000037
for 3. corresponding ordinate of fitting function.
C. for the priority of three coordinates according to (x 3,
Figure FDA0000397539920000038
) priority > (x 3,
Figure FDA0000397539920000039
) priority > (x 3,
Figure FDA00003975399200000310
) priority select, in the time of in 1. fitting function is predicted not, 1. fitting function is designated as miss, in the time of in 2. fitting function is predicted not, 2. fitting function is designated as miss, in the time of in 3. fitting function is predicted not, 3. fitting function is designated as miss, is allly marked as miss function, will no longer participate in selecting, when three functions are all marked as miss time, return to matching miss.
Judge that the method whether fitting function hit is as follows:
First by the image of next frame to judge in advance centered by coordinate, take that the size of following the tracks of frame is border, extracts next frame and follow the tracks of the image in frame, and the dividing method of taking according to present frame cuts apart, be divided into (m/k) * (n/k) individual fritter.
Then according to next frame, follow the tracks of 8 characteristic points that frame extracts, use formula
Finally by the ORB characteristic (64 strings of binary characters) of the ORB characteristic of present frame (64 strings of binary characters) and next frame, make comparisons, count the identical number of numerical value (0 or 1) on correspondence position, if be greater than a default numerical value, mean and follow the tracks of successfully, matching is hit.
4. home for destitute according to claim 1 old man's hazardous act monitoring method, it is characterized in that the detailed process of in the position of next frame, target old man being followed the tracks of according to predicted old man described in step 4 is, use the trace routine in individual PC (3), according to the above-mentioned old man who predicts, in the position of next frame, target old man is followed the tracks of, the detailed process of following the tracks of is that the trace routine in individual PC (3) calculates ORB feature and the ORB characteristic that previous frame is followed the tracks of frame, and under present frame, follow the tracks of frame in ORB feature and the ORB characteristic of the position of prediction gained, and compare whether the element on same position in the two characteristic is identical, if identical number is more than or equal to threshold value, show to trace into target, retain current ORB feature and ORB characteristic, continue to process next frame, if identical number is less than threshold value, show that target is miss, need again to obtain next possible position and recalculate correspondence position ORB characteristic according to described approximating method, and relatively contrast with current ORB characteristic, until matching is hit, continue to process next frame, or all possible Fitting Coordinate System is all miss, this step returning tracking is lost.
If above-mentioned tracing process can trace into old man by continuous 10 frames, carry out frame-skipping operation,, every three frames performance objective track algorithm again, can greatly improve tracking efficiency like this; If above-mentioned tracing process returning tracking is lost, need to get back to before three frames, reuse above-mentioned tracing process and process, if followed the tracks of successfully, continue next frame; If followed the tracks of unsuccessfully, returning tracking is lost.
If returning tracking is lost, return to previous frame, the current old man that will follow the tracks of is carried out to 360 degree samplings around, and find the tracking frame of matching degree maximum, if the matching degree of this tracking frame is greater than threshold value, show that tracking target picks up after loss and huge change occurs in old man's path, call behavior recognizer old man's present case is identified; If the matching degree of this tracking frame is less than threshold value, show that old man follows the tracks of loss, delete current tracking frame, and call frame difference method and again detect the old man's motion conditions in guarded region.
5. home for destitute according to claim 1 old man's hazardous act monitoring method, is characterized in that initial parameter deterministic process when process that the old man's that the behavior recognizer in the use individual's PC (3) described in step 5 traces into trace routine behavior is identified comprises first use this method and each behavior relative discern process while using this method;
Initial parameter deterministic process during described first use this method is: first choose and have the sample of label χ={ χ 1, χ 2..., χ n, χ wherein ibe the video that a segment mark has label, the old man that label substance records for this section of video behavior in daily life, for example C 1for walking label, C 2for the label of standing, C 3for the label of falling, C 4for the label etc. of too bending over, using these videos as training sample, several training samples are taked in each labeling requirement under varying environment, and the video of each training sample has three dimensions, and image transverse axis, the image longitudinal axis, time shaft, input data χ iwei San rank tensor, then adopts differentiation tensor and canonical correlation analysis algorithm based on increasing progressively to learn these training samples, thereby obtains three transition matrix: U 1, U 2, U 3, then to three rank tensor: Y after each sample calculation dimensionality reduction i=χ i* 1u 1* 2u 2* 3u 3, by belonging to respectively walking, stand, fall, too bend over etc., the behavior of N kind is defined as C 1, C 2c nclass, the classification center of each class is:
Figure FDA0000397539920000041
y wherein jfor belonging to C iclass sample χ jin tensor after all dimensionality reductions,
Figure FDA0000397539920000042
for belonging to C ithe number of dvielement.The classification center of three transition matrixes that obtain and each class behavior is inputted to behavior recognizer in individual PC (3) as the parameter of behavior relative discern;
The step of the described PCA based on tensor is as follows:
A. for the video segment of input, can be understood as a three-dimensional matrice, this matrix has three dimensions, i.e. transverse axis x, and longitudinal axis y, time shaft t, establishing each video segment is X i, the video segment of all inputs is formed to four bit matrix X=[X 1, X 2..., X n]. random initializtion matrix U 1, U 2, U 3
B. compute matrix D 1=X * 2u 2* 3u 3, to matrix D 1* D 1 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 1, compute matrix D 2=X * 1u 1* 3u 3, to matrix D 2* D 2 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 2, compute matrix D 3=X * 1u 1* 2u 2, to matrix D 3* D 3 task its eigenvalue λ iand characteristic vector v i, select eigenvalue λ ithe set that the corresponding characteristic vector of > 0 forms, by the value assignment of this set to U 3, wherein, X * iu ifor the multiplication of tensor, its implication is by high dimension vector X, retains its i dimension, and other all dimensions are launched successively, forms a two-dimensional matrix X i, and calculate U i* X i, gained matrix carries out inverse transformation according to the order of launching again, the higher dimensional matrix of gained be X * iu ivalue.
C. calculate double iteration, gained matrix
Figure FDA0000397539920000051
with
Figure FDA0000397539920000052
poor:
Figure FDA0000397539920000053
with
Figure FDA0000397539920000055
three matrixes of gained, by all elements summed square, gained and be the variable quantity of twice iteration, if variable quantity is less than threshold value, iteration stops, current
Figure FDA0000397539920000056
be the result of algorithm gained, otherwise repeat foregoing b process.
Behavior relative discern process during described each use this method is: the abnormal behaviour of first autotracking algorithm in video image being found is as the time started point of behavior relative discern sample, centered by the central point in current tracking frame region, take current tracking frame longest edge as the length of side, thereby obtain a square area, by the video image in current region, read the monitor video in 3 seconds, as sample X to be sorted, by formula Y=χ * 1u 1* 2u 2* 3u 3, U wherein 1, U 2, U 3for the transition matrix of initial parameter deterministic process gained, * i(i=1,2,3) are the oeprator that tensor multiplies each other, and obtain three rank tensor Y after this sample dimensionality reduction, by the classification center of Y and above-mentioned each class
Figure FDA0000397539920000057
on three dimensions, calculate its Euclidean distance, the nearest class of chosen distance is as its classification results, thereby complete classification, according to current behavior by classification gained result be recognition result, if recognition result is normal behaviours such as walking, stand, do not need to display (4) and warning audio amplifier (5) transmission of signal, if recognition result is for the abnormal behaviour such as falling, too bend over, signal corresponding to current recognition result need to be passed to display (4) and warning audio amplifier (5).
CN201310489096.8A 2013-10-17 2013-10-17 A kind of nursing house old man's hazardous act monitoring method Expired - Fee Related CN103517042B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201310489096.8A CN103517042B (en) 2013-10-17 2013-10-17 A kind of nursing house old man's hazardous act monitoring method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201310489096.8A CN103517042B (en) 2013-10-17 2013-10-17 A kind of nursing house old man's hazardous act monitoring method

Publications (2)

Publication Number Publication Date
CN103517042A true CN103517042A (en) 2014-01-15
CN103517042B CN103517042B (en) 2016-06-29

Family

ID=49898967

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201310489096.8A Expired - Fee Related CN103517042B (en) 2013-10-17 2013-10-17 A kind of nursing house old man's hazardous act monitoring method

Country Status (1)

Country Link
CN (1) CN103517042B (en)

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105162851A (en) * 2015-08-12 2015-12-16 罗令 Method and system for achieving instant transmission of positioning prompt information
CN105844663A (en) * 2016-03-21 2016-08-10 中国地质大学（武汉） Adaptive ORB object tracking method
CN106454277A (en) * 2016-11-30 2017-02-22 杭州联络互动信息科技股份有限公司 Image analysis method and device for video monitoring
CN106571014A (en) * 2016-10-24 2017-04-19 上海伟赛智能科技有限公司 Method for identifying abnormal motion in video and system thereof
CN106598856A (en) * 2016-12-14 2017-04-26 广东威创视讯科技股份有限公司 Path detection method and path detection device
CN107045623A (en) * 2016-12-30 2017-08-15 厦门瑞为信息技术有限公司 A kind of method of the indoor dangerous situation alarm based on human body attitude trace analysis
CN107072548A (en) * 2014-11-03 2017-08-18 皇家飞利浦有限公司 For people orientation and/or position automatic detection equipment, system and method
CN107346407A (en) * 2016-05-04 2017-11-14 东北林业大学 Cheating determination methods based on MK60 recognitions of face
CN107710281A (en) * 2015-06-11 2018-02-16 柯尼卡美能达株式会社 Motion detection system, action detection device, motion detection method and motion detection program
CN107735813A (en) * 2015-06-10 2018-02-23 柯尼卡美能达株式会社 Image processing system, image processing apparatus, image processing method and image processing program
CN107729876A (en) * 2017-11-09 2018-02-23 重庆医科大学 Fall detection method in old man room based on computer vision
CN108134922A (en) * 2017-12-25 2018-06-08 李锐 A kind of behavior label and retroactive method based on artificial intelligence
CN108256479A (en) * 2018-01-17 2018-07-06 百度在线网络技术（北京）有限公司 Face tracking method and device
CN108280843A (en) * 2018-01-24 2018-07-13 新华智云科技有限公司 A kind of video object detecting and tracking method and apparatus
CN109271999A (en) * 2018-09-06 2019-01-25 北京京东尚科信息技术有限公司 Processing method, device and the computer readable storage medium of image
CN109473173A (en) * 2018-09-30 2019-03-15 华中科技大学 A kind of the elderly's Cognitive deficiency assessment system and device based on video
CN109934845A (en) * 2019-03-15 2019-06-25 腾讯科技（深圳）有限公司 Based on the timing behavior capture frame generation method and device from attention network
CN109947793A (en) * 2019-03-20 2019-06-28 深圳市北斗智能科技有限公司 Analysis method, device and the storage medium of accompanying relationship
WO2019179024A1 (en) * 2018-03-20 2019-09-26 平安科技（深圳）有限公司 Method for intelligent monitoring of airport runway, application server and computer storage medium
CN110490171A (en) * 2019-08-26 2019-11-22 睿云联(厦门)网络通讯技术有限公司 A kind of danger gesture recognition method, device, computer equipment and storage medium
CN110837529A (en) * 2019-11-27 2020-02-25 王培根 Big data analysis monitoring method and device, server and readable storage medium
CN111149129A (en) * 2017-09-21 2020-05-12 株式会社安德科技术 Abnormality detection device and abnormality detection method
CN111246177A (en) * 2020-01-23 2020-06-05 深圳市大拿科技有限公司 User safety alarm prompting method and system
CN111292504A (en) * 2020-01-23 2020-06-16 深圳市大拿科技有限公司 Method and system for carrying out safety alarm through image identification
CN111681149A (en) * 2020-06-05 2020-09-18 视云融聚（广州）科技有限公司 Nursing home guardianship object refined monitoring method, storage device and mobile terminal
CN113826150A (en) * 2019-04-11 2021-12-21 诺比有限责任公司 Old person nursing and security system
CN114724335A (en) * 2022-03-24 2022-07-08 慧之安信息技术股份有限公司 Nursing home safety monitoring system and method based on edge calculation
CN114758417A (en) * 2021-04-12 2022-07-15 沈阳工业大学 Intelligent old-age-protecting sensing control method
CN116188785A (en) * 2023-05-04 2023-05-30 福建环宇通信息科技股份公司 Polar mask old man contour segmentation method using weak labels

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110415823B (en) * 2019-07-30 2020-08-04 杭州思锐信息技术股份有限公司 Machine learning-based safety state data processing method and system

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060225120A1 (en) * 2005-04-04 2006-10-05 Activeye, Inc. Video system interface kernel
CN101098465A (en) * 2007-07-20 2008-01-02 哈尔滨工程大学 Moving object detecting and tracing method in video monitor
CN101727570A (en) * 2008-10-23 2010-06-09 华为技术有限公司 Tracking method, track detection processing unit and monitor system
CN102663452A (en) * 2012-04-14 2012-09-12 中国人民解放军国防科学技术大学 Suspicious act detecting method based on video analysis
CN102799873A (en) * 2012-07-23 2012-11-28 青岛科技大学 Human body abnormal behavior recognition method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060225120A1 (en) * 2005-04-04 2006-10-05 Activeye, Inc. Video system interface kernel
CN101098465A (en) * 2007-07-20 2008-01-02 哈尔滨工程大学 Moving object detecting and tracing method in video monitor
CN101727570A (en) * 2008-10-23 2010-06-09 华为技术有限公司 Tracking method, track detection processing unit and monitor system
CN102663452A (en) * 2012-04-14 2012-09-12 中国人民解放军国防科学技术大学 Suspicious act detecting method based on video analysis
CN102799873A (en) * 2012-07-23 2012-11-28 青岛科技大学 Human body abnormal behavior recognition method

Cited By (39)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107072548A (en) * 2014-11-03 2017-08-18 皇家飞利浦有限公司 For people orientation and/or position automatic detection equipment, system and method
CN107072548B (en) * 2014-11-03 2020-09-01 皇家飞利浦有限公司 Device, system and method for automatic detection of orientation and/or position of a person
CN107735813A (en) * 2015-06-10 2018-02-23 柯尼卡美能达株式会社 Image processing system, image processing apparatus, image processing method and image processing program
CN107710281A (en) * 2015-06-11 2018-02-16 柯尼卡美能达株式会社 Motion detection system, action detection device, motion detection method and motion detection program
CN105162851A (en) * 2015-08-12 2015-12-16 罗令 Method and system for achieving instant transmission of positioning prompt information
CN105844663B (en) * 2016-03-21 2018-11-27 中国地质大学（武汉） A kind of adaptive ORB method for tracking target
CN105844663A (en) * 2016-03-21 2016-08-10 中国地质大学（武汉） Adaptive ORB object tracking method
CN107346407A (en) * 2016-05-04 2017-11-14 东北林业大学 Cheating determination methods based on MK60 recognitions of face
CN106571014A (en) * 2016-10-24 2017-04-19 上海伟赛智能科技有限公司 Method for identifying abnormal motion in video and system thereof
CN106454277B (en) * 2016-11-30 2019-09-27 杭州联络互动信息科技股份有限公司 A kind of image analysis method and device for video monitoring
CN106454277A (en) * 2016-11-30 2017-02-22 杭州联络互动信息科技股份有限公司 Image analysis method and device for video monitoring
CN106598856B (en) * 2016-12-14 2019-03-01 威创集团股份有限公司 A kind of path detection method and device
WO2018107748A1 (en) * 2016-12-14 2018-06-21 威创集团股份有限公司 Path detection method and apparatus
CN106598856A (en) * 2016-12-14 2017-04-26 广东威创视讯科技股份有限公司 Path detection method and path detection device
CN107045623A (en) * 2016-12-30 2017-08-15 厦门瑞为信息技术有限公司 A kind of method of the indoor dangerous situation alarm based on human body attitude trace analysis
CN107045623B (en) * 2016-12-30 2020-01-21 厦门瑞为信息技术有限公司 Indoor dangerous condition warning method based on human body posture tracking analysis
CN111149129A (en) * 2017-09-21 2020-05-12 株式会社安德科技术 Abnormality detection device and abnormality detection method
CN107729876A (en) * 2017-11-09 2018-02-23 重庆医科大学 Fall detection method in old man room based on computer vision
CN108134922A (en) * 2017-12-25 2018-06-08 李锐 A kind of behavior label and retroactive method based on artificial intelligence
CN108134922B (en) * 2017-12-25 2020-09-25 李锐 Behavior marking and tracing method based on artificial intelligence
CN108256479A (en) * 2018-01-17 2018-07-06 百度在线网络技术（北京）有限公司 Face tracking method and device
CN108280843A (en) * 2018-01-24 2018-07-13 新华智云科技有限公司 A kind of video object detecting and tracking method and apparatus
WO2019179024A1 (en) * 2018-03-20 2019-09-26 平安科技（深圳）有限公司 Method for intelligent monitoring of airport runway, application server and computer storage medium
CN109271999A (en) * 2018-09-06 2019-01-25 北京京东尚科信息技术有限公司 Processing method, device and the computer readable storage medium of image
CN109473173A (en) * 2018-09-30 2019-03-15 华中科技大学 A kind of the elderly's Cognitive deficiency assessment system and device based on video
CN109934845B (en) * 2019-03-15 2022-11-29 腾讯科技（深圳）有限公司 Time sequence behavior capturing frame generation method and device based on self-attention network
CN109934845A (en) * 2019-03-15 2019-06-25 腾讯科技（深圳）有限公司 Based on the timing behavior capture frame generation method and device from attention network
CN109947793A (en) * 2019-03-20 2019-06-28 深圳市北斗智能科技有限公司 Analysis method, device and the storage medium of accompanying relationship
CN113826150A (en) * 2019-04-11 2021-12-21 诺比有限责任公司 Old person nursing and security system
US11749085B2 (en) 2019-04-11 2023-09-05 Nobi Bv Elderly care and security system
CN110490171A (en) * 2019-08-26 2019-11-22 睿云联(厦门)网络通讯技术有限公司 A kind of danger gesture recognition method, device, computer equipment and storage medium
CN110837529A (en) * 2019-11-27 2020-02-25 王培根 Big data analysis monitoring method and device, server and readable storage medium
CN111292504A (en) * 2020-01-23 2020-06-16 深圳市大拿科技有限公司 Method and system for carrying out safety alarm through image identification
CN111246177A (en) * 2020-01-23 2020-06-05 深圳市大拿科技有限公司 User safety alarm prompting method and system
CN111681149A (en) * 2020-06-05 2020-09-18 视云融聚（广州）科技有限公司 Nursing home guardianship object refined monitoring method, storage device and mobile terminal
CN114758417A (en) * 2021-04-12 2022-07-15 沈阳工业大学 Intelligent old-age-protecting sensing control method
CN114758417B (en) * 2021-04-12 2024-04-02 沈阳工业大学 Intelligent aging protection sensing and controlling method
CN114724335A (en) * 2022-03-24 2022-07-08 慧之安信息技术股份有限公司 Nursing home safety monitoring system and method based on edge calculation
CN116188785A (en) * 2023-05-04 2023-05-30 福建环宇通信息科技股份公司 Polar mask old man contour segmentation method using weak labels

Also Published As

Publication number Publication date
CN103517042B (en) 2016-06-29

Similar Documents

Publication Publication Date Title
CN103517042A (en) Nursing home old man dangerous act monitoring method
CN102163290B (en) Method for modeling abnormal events in multi-visual angle video monitoring based on temporal-spatial correlation information
EP1955285B1 (en) Detection of abnormal crowd behavior
CN107679471B (en) Indoor personnel air post detection method based on video monitoring platform
CN105426820B (en) More people&#39;s anomaly detection methods based on safety monitoring video data
CN102201146B (en) Active infrared video based fire smoke detection method in zero-illumination environment
CN103077423B (en) To run condition detection method based on crowd&#39;s quantity survey of video flowing, local crowd massing situation and crowd
CN103942850A (en) Medical staff on-duty monitoring method based on video analysis and RFID (radio frequency identification) technology
Ma et al. Real-time monitoring of water quality using temporal trajectory of live fish
CN202257856U (en) Driver fatigue-driving monitoring device
CN110728252B (en) Face detection method applied to regional personnel motion trail monitoring
CN104966304A (en) Kalman filtering and nonparametric background model-based multi-target detection tracking method
CN101576952A (en) Method and device for detecting static targets
CN102254394A (en) Antitheft monitoring method for poles and towers in power transmission line based on video difference analysis
CN111681382A (en) Method for detecting temporary fence crossing in construction site based on visual analysis
CN115484263A (en) Cross-regional intelligent factory real-time safety monitoring system based on edge calculation
CN115457446A (en) Abnormal behavior supervision system based on video analysis
CN110458198A (en) Multiresolution target identification method and device
CN112800975A (en) Behavior identification method in security check channel based on image processing
CN115691034A (en) Intelligent household abnormal condition warning method, system and storage medium
CN109086717B (en) Violent behavior detection system and method based on human skeleton and motion signal characteristics
CN106941580A (en) Method and system of the teacher student from motion tracking is realized based on single detective camera lens
CN114067396A (en) Vision learning-based digital management system and method for live-in project field test
CN116419059A (en) Automatic monitoring method, device, equipment and medium based on behavior label
CN115272967A (en) Cross-camera pedestrian real-time tracking and identifying method, device and medium

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
C14 Grant of patent or utility model
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20160629

Termination date: 20181017