CN109993118B - Action recognition method and system - Google Patents
Action recognition method and system Download PDFInfo
- Publication number
- CN109993118B CN109993118B CN201910253883.XA CN201910253883A CN109993118B CN 109993118 B CN109993118 B CN 109993118B CN 201910253883 A CN201910253883 A CN 201910253883A CN 109993118 B CN109993118 B CN 109993118B
- Authority
- CN
- China
- Prior art keywords
- information
- action
- motion
- weak
- meta
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/29—Graphical models, e.g. Bayesian networks
- G06F18/295—Markov models or related models, e.g. semi-Markov models; Markov random fields; Networks embedding Markov models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The application discloses a method for recognizing actions and a recognition system thereof, wherein the method for recognizing the actions specifically comprises the following steps: collecting at least one action, decomposing the action and determining a meta-action; preprocessing the element action to remove weak information in the element action; the determination of the element action comprises the steps of establishing a coordinate system according to the original action, dividing the collected action into a plurality of quadrants, determining the quadrant where the new action is located and the motion track of the new action when the new action occurs, and determining the element action according to the new quadrant and the motion track of the new action. The action recognition method and the action recognition system can filter weak information by using the principle of information entropy, eliminate the weak information in the element action sequence and integrate the information of the element action sequence, so that the problem of excessive classification in recognition is avoided, and the recognition accuracy and efficiency are improved.
Description
Technical Field
The present application relates to the field of motion recognition, and in particular, to a motion recognition method and a motion recognition system.
Background
In the natural world, a person manipulates an object by touch, expresses a desire by a gesture. It is desirable that a person interact with a computer in the same way. Therefore, in the research on the "multi-dimensional machine vision action perception", it is necessary to firstly study how to collect the three-dimensional action of the human body, then to process the collected complex information to identify the effective action, and finally to judge the effective action and analyze the intention of the human. However, in the conventional effective analysis process, the actions with excessive classifications cannot be quickly and effectively identified, so that the possibility of action analysis errors or estimation errors may occur.
Disclosure of Invention
The method aims to process the element action information before action recognition, model the action collection and the action recognition, discriminate weak information by utilizing a wiener process and a Markov-like algorithm according to the characteristics of human body actions, filter the weak information by utilizing the principle of information entropy, reject the weak information in an element action sequence and integrate the information of the element action sequence, thereby avoiding the problem of excessive classification in the recognition and further improving the accuracy and the efficiency of the recognition.
In order to achieve the above object, the present application provides a method for motion recognition and a recognition system thereof, wherein the method for motion recognition specifically includes the following steps: collecting at least one action, decomposing the action and determining a meta-action; preprocessing the element action to remove weak information in the element action; the determination of the element action comprises the steps of establishing a coordinate system according to the original action, dividing the collected action into a plurality of quadrants, determining the quadrant where the new action is located and the motion track of the new action when the new action occurs, and determining the element action according to the new quadrant and the motion track of the new action.
The method comprises the steps of utilizing an action preprocessing layer to preprocess the meta-action by adopting an entropy-based information extraction method and removing weak information in the meta-action.
As above, wherein the removing of the weak information comprises the following sub-steps: decomposing the meta-motion to obtain posture information and motion information; weak information is searched according to the action information; and searching a weak information base, and judging whether the current weak information can be removed.
As above, wherein in finding weak information, X (n) ═ X is assumed1,X2,X3...XnIs the meta-motion information of a part of the human body, where n is the time, X1,X2,X3...XnThe meta-action information is a sub-action of the meta-action at each time, and satisfies the following condition: x (n) ═ X1,X2,X3...XnIs an independent incremental process; at any time m, n>0, X (m + N) -X (N) -N (0, c ^2 ^ N), X (m + N) -X (N) is the increment of random motion in the interval (m, N), namely X (m + N) -X (N) is expected to be 0, and the variance is a normal distribution of c ^2 ^ N; x (n) is a continuous function with respect to n.
As above, wherein by pairing the sequence X (n) ═ { X1,X2,X3...XnCalculating probability to find weak information, setting state S (S)1,S2) For the set of strong and weak information, the computational model is as follows:
where P (S | X) is the probability of the meta-action set based on state S, P (X) is the initial probability distribution of the actions, T is the length of the observation time, T is the initial observation time,
as initial state probability matrix, S0In an initial state, B＝(B0,...,BN-1) The probability distribution of observations is generated for states in the traditional continuous hidden markov model, but in this application B ═ B (B)0,...,BN-1) To generate a probability distribution of meta-motion information in a current state of a person, where B (X)0) Is the initial state probability distribution;
is the probability distribution, S, of a certain unary action in a certain statetIn a certain state, XtIn order to perform a certain action, the user can select the action,
as hidden states in the Markov model, St-1T is a constant value in a state immediately preceding a certain state.
As above, the following process is performed for the result P (S | X):
the threshold is a weak threshold, and when the value of P (S | X) is greater than the threshold, it means that the difference between the current action and the previous action is smaller than a specified threshold, that is, when the value of relut is 1, the information is weak information, and when the value of relut is 0, the information is strong information.
The above, wherein the hamming distance is used to calculate the similarity between the current weak information and the weak information in the weak information database.
As described above, the motion image of the human body is acquired simultaneously with the acquisition of the motion, and the posture information and the motion information are obtained from the motion image.
A motion recognition system comprises an acquisition and analysis unit and a removal unit; the acquisition and analysis unit decomposes the collected actions to determine the meta-actions; the removing unit is connected with the acquisition and analysis unit and is used for preprocessing the meta-motion and removing weak information in the meta-motion.
As above, wherein the removing unit includes the following modules; an information classification module: the system is used for classifying according to the information in the collected meta-actions and decomposing posture information and action information; a searching module: for finding weak information; a search module: the weak information database is used for searching the weak information database and judging whether the current weak information can be removed; an integration module: and the method is used for adding the unremoved weak information into the strong information for information integration.
The application has the following beneficial effects:
the method and the system for recognizing the actions can process the meta-action information before action recognition, model the action collection and the action recognition, discriminate weak information by utilizing a wiener process and a Markov-like algorithm according to the action characteristics of a human body, filter the weak information by utilizing the principle of information entropy, eliminate the weak information in the meta-action sequence and integrate the information of the meta-action sequence, thereby avoiding the problem of excessive classification in the recognition and further improving the accuracy and the efficiency of the recognition.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art according to the drawings.
FIG. 1 is a flow chart of a method of motion recognition provided according to an embodiment of the present application;
FIG. 2 is a sub-flowchart of a method for motion recognition provided according to an embodiment of the present application;
FIG. 3 is a schematic diagram of actions provided according to an embodiment of the present application;
FIG. 4 is a schematic diagram of an internal structure of an identification system provided according to an embodiment of the present application;
fig. 5 is a schematic diagram of sub-modules of an identification system provided according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application are clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are some, but not all, embodiments of the present application. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
According to the method, meta-motion information is processed before motion recognition, motion collection and motion recognition are modeled, weak information is discriminated by using a wiener process and a Markov-like algorithm according to human motion characteristics, and then the weak information is filtered by using the principle of information entropy, so that the weak information in a meta-motion sequence is eliminated, and the information of the meta-motion sequence is integrated, so that the problem of excessive classification in recognition is avoided, and the accuracy and efficiency of recognition are improved.
Fig. 1 is a flowchart illustrating a method for recognizing an action provided by the present application.
Step S110: and collecting a series of actions, and decomposing the actions to determine meta-actions.
In particular, since the actions of people are very coordinated and random, when a person completes an action, many parts of the body are often used to cooperatively complete the action, and the ways in which the same person completes the action are different even when the same person completes the action for many times.
Firstly, collecting actions, establishing a coordinate system according to the original actions, dividing the collected actions into a plurality of quadrants, such as common four or three quadrants, when a new action occurs, determining the quadrant where the new action is located and the motion trail of the new action, and determining the element action according to the new quadrant and the motion trail of the new action.
For example, if a person wants to pick up a cup, there are many possible actions, and there are many meta-actions, for example, pick up the cup in front of the left of the body, the main action for completing the action is to extend the left hand and pick up the cup (case a), and the process of extending the left hand and picking up the cup is the meta-action of the new action a. However, the general action of a person is that the body slightly turns to the left to take the cup, then the left hand is extended to take the cup (situation B), and the process of turning left, extending right hand and taking the cup is the element action of the new action B; certainly, a person may turn right slightly and get a personal call to a colleague, and extend the left hand to pick up the cup (situation C), and the process of turning right, extending left hand and picking up the cup is the primary action of the new action C.
Although the three cases are different in actions, the target is to pick up the cup, but from the perspective of data, the difference of the data in the new meta-action A, B, C is very large, which cannot accurately and quickly identify the target action, and therefore the meta-action needs to be processed to execute step S120.
Step S120: and preprocessing the meta-action to remove useless information in the meta-action.
Specifically, by using the action preprocessing layer, the meta-action is preprocessed by using an entropy-based information extraction method to remove useless information in the meta-action, and the useless information can also be expressed as weak information, as exemplified by the case A, B, C in step S120, the slight left turn and the slight right turn of the body in the case B, C are useless information that increases difficulty or makes identification meaningless for action identification, and therefore needs to be removed, wherein the removal of the useless information specifically includes the following steps:
step S210: and (5) information classification.
And classifying according to the collected information in the meta-motion, and decomposing the posture information and the motion information in the meta-motion.
Preferably, in step S110, the motion image of the human body is acquired simultaneously with the motion, the posture information and the motion information are obtained from the motion image, whether there is corresponding posture information and motion information in the posture library and the motion library is searched, and if the similarity between the acquired posture information and the posture in the posture library is higher than a predetermined threshold, the two are considered as the same posture. And if the similarity between the collected motion information and the motion in the motion library is higher than a preset threshold value, the collected motion information and the motion in the motion library are considered to be the same motion. The gesture library and the action library are respectively stored with various gestures and actions of the human body.
Step S220: weak information is sought.
Specifically, weak information is searched for in the action information, and the weak information (weak action) is the above-mentioned useless information.
Further, in the above case A, B, C, since the motion of the person is continuous and newton's second law is applied, the motion of the person at each moment is considered to be related to the motion at the previous moment, and since the motion of the person is subject to subjective control of the person and limitation of bones and muscles, the motion increment is independent and has a certain range. It can therefore be assumed that the human action is a wiener process, i.e. the human action process satisfies the following conditions:
X(n)＝{X1,X2,X3...Xnis the meta-motion information of a part of the human body, where n is the time, X1,X2,X3...XnA sub-action that is a meta-action at each time. Which satisfies the following conditions:
(1)X(n)＝{X1,X2,X3...Xnis an independent incremental process;
(2) at any time m, N >0, X (m + N) -X (N) to N (0, c ^2 ^ N), X (m + N) -X (N) is the increment of the random motion in the section (m, N), and m is the time. I.e., X (m + n) -X (n) is a normal distribution with a variance of c ^ 2X n, desirably 0;
(3) x (n) is a continuous function with respect to n.
When a person is in different states (e.g., sitting, standing, walking, etc.), c in the above formula N (0, c ^2 x N) is different. Wherein c is a self-setting parameter, and the larger the action amplitude of the person is, the larger c is, preferably, c can be 1 if the state of the person is a static sitting posture or a standing posture.
With the above convention, the weak actions can be separated, specifically, the sequence probability is calculated to judge that the sequence probability is strong information or weak information, so as to search the weak information. Preferably, the hidden markov method is applied, and the observation sequence is meta-motion information X (n) { X1,X2,X3...XnState S (S)1,S2) Namely the set of strong information and weak information, the calculation model is as follows:
The following processing is performed for the result P (S | X):
As shown in fig. 3, for the action diagram provided in the embodiment of the present application, when the upper arm of the human is in the a state at time t, and the next action does not change much based on the a state, the probability that the upper arm is in the B state at time t +1 is very high (i.e., the probability is very high), and therefore the B state is weak information. Similarly, at time t +1, the possibility that the upper arm changes from the state a to the state C is low unless someone intentionally performs this action, and therefore, it is strong information that the probability is lower than the threshold, that is, the state C is strong information.
Step S230: and searching a weak information base, and judging whether the current weak information can be removed.
Specifically, in some cases, the found weak information is not really removable weak information, and therefore, the searched weak information needs to be searched in the existing weak information base to judge whether the found weak information is really removable weak information. The weak information base stores a plurality of pieces of weak information obtained in advance, during searching, the current weak information is compared with the weak information in the weak information base, whether the current weak information is similar to the weak information in the weak information base or not is judged, and optionally, the similarity between the current weak information and the weak information in the weak information base is calculated by using the hamming distance. If the similarity is higher than the preset distance, step S240 is performed: the weak information is removed.
If the similarity is lower than the preset distance, step S250 is executed: and adding the current weak information into the strong information for information integration.
Fig. 4 is a schematic view of an internal structure of the motion recognition system according to the embodiment of the present application, specifically, the recognition system includes an acquisition and analysis unit 401 and a removal unit 402.
The removing unit 402 includes the following sub-modules, as shown in fig. 5, specifically including an information classifying module 501, a searching module 502, an extracting module 503, a searching module 504, and an integrating module 506.
The collection and analysis unit 401 is configured to collect an action, decompose the action, and determine a meta-action.
The removing unit 402 is connected to the collecting and analyzing unit 401, and is configured to perform preprocessing on the meta-motion and remove information (i.e., weak information) that is not used in the meta-motion.
The information classification module 501 is configured to classify the meta-motion according to the collected information, and decompose the posture information and the motion information.
The searching module 502 is connected to the information classifying module 501 for searching weak information.
The extracting module 503 is connected to the finding module 502 for extracting the weak information features.
The searching module 504 is connected to the extracting module 503, and is configured to search for weak information according to the weak information feature, and search out whether the weak information is present.
The integration module 506 is connected to the search module 504, and is configured to add weak information that is not already present to strong information for information integration.
The application has the following beneficial effects:
the method and the system for recognizing the actions can process the meta-action information before action recognition, model the action collection and the action recognition, discriminate weak information by utilizing a wiener process and a Markov-like algorithm according to the action characteristics of a human body, filter the weak information by utilizing the principle of information entropy, eliminate the weak information in the meta-action sequence and integrate the information of the meta-action sequence, thereby avoiding the problem of excessive classification in the recognition and further improving the accuracy and the efficiency of the recognition.
Although the present application has been described with reference to examples, which are intended to be illustrative only and not to be limiting of the application, changes, additions and/or deletions may be made to the embodiments without departing from the scope of the application.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (8)
1. A method for recognizing actions is characterized by comprising the following steps:
collecting a series of actions, decomposing the actions and determining meta-actions;
preprocessing the meta-motion to remove useless information in the meta-motion, wherein the useless information is weak information and is information which increases difficulty or makes identification meaningless for motion identification;
wherein the determining of the meta-action comprises:
establishing a coordinate system according to the original action;
dividing the collected motion into a plurality of quadrants;
when a new action occurs, determining a quadrant where the new action is located and a motion track of the new action, and determining a meta-action according to the new quadrant and the motion track of the new action;
the removing of the weak information comprises the following substeps:
decomposing the meta-motion to obtain posture information and motion information;
weak information is searched according to the action information;
searching a weak information base, and judging whether the current weak information can be removed;
set state S (S)1,S2) For the set of strong and weak information, the computational model is as follows:
wherein P (S | X) is the probability of the meta-action set X based on the state S, P (X) is the initial probability distribution of the actions, T is the length of the observation time, T is the initial observation time,
as initial state probability matrix, S0Is in an initial state, B ═ B0,...,BN-1) To generate a probability distribution of meta-motion information in the current state, where B (X)0) Is the initial state probability distribution;
is the probability distribution, S, of a certain unary action in a certain statetIn a certain state, XtIn order to perform a certain action, the user can select the action,
as hidden states in the Markov model, St-1T is a constant value in a state immediately preceding a certain state.
2. The method of action recognition as recited in claim 1, wherein the meta-actions are preprocessed by an entropy-based information extraction method using an action preprocessing layer to remove weak information therein.
3. The method of motion recognition according to claim 1, wherein in finding weak information, X (n) { X is assumed1,X2,X3...XnIs the meta-motion information of a part of the human body, where n is the time, X1,X2,X3...XnThe meta-action information is a sub-action of the meta-action at each time, and satisfies the following condition:
X(n)＝{X1,X2,X3...Xnis an independent incremental process;
at any time m, N >0, X (m + N) -X (N) to N (0, c ^2 ^ N), X (m + N) -X (N) is increment of random motion in the interval (m, N), namely X (m + N) -X (N) is expected to be 0, and the variance is normal distribution of c ^2 ^ N; x (n) is a continuous function with respect to n.
4. The method of motion recognition according to claim 3, wherein the following is done for the result P (S | X):
the threshold is a weak threshold, and when the value of P (S | X) is greater than the threshold, it means that the difference between the current action and the previous action is smaller than a specified threshold, that is, when the value of relut is 1, the information is weak information, and when the value of relut is 0, the information is strong information.
5. The method of motion recognition according to claim 2, wherein the hamming distance is used to calculate the similarity between the current weak information and the weak information in the weak information repository.
6. The motion recognition method according to claim 1, wherein the motion image of the human body is acquired simultaneously with the acquisition of the motion, and the posture information and the motion information are obtained from the motion image.
7. The motion recognition system is characterized by comprising a collecting and analyzing unit and a removing unit;
the acquisition and analysis unit is used for acquiring a series of actions and decomposing the actions to determine meta-actions;
the removing unit is connected with the acquisition and analysis unit and is used for preprocessing the meta-motion and removing useless information in the meta-motion, wherein the useless information is weak information and is information for increasing difficulty or making identification meaningless for motion identification;
the removing of the weak information comprises the following substeps:
decomposing the meta-motion to obtain posture information and motion information;
weak information is searched according to the action information;
searching a weak information base, and judging whether the current weak information can be removed;
the removing of the weak information comprises the following substeps:
decomposing the meta-motion to obtain posture information and motion information;
weak information is searched according to the action information;
searching a weak information base, and judging whether the current weak information can be removed;
set state S (S)1,S2) For the set of strong and weak information, the computational model is as follows:
wherein P (S | X) is the probability of the meta-action set X based on the state S, P (X) is the initial probability distribution of the actions, T is the length of the observation time, T is the initial observation time,
as initial state probability matrix, S0Is in an initial state, B ═ B0,...,BN-1) To generate probability score of meta-action information in current stateCloth of which B (X)0) Is the initial state probability distribution;
is the probability distribution, S, of a certain unary action in a certain statetIn a certain state, XtIn order to perform a certain action, the user can select the action,
as hidden states in the Markov model, St-1T is a constant value in a state immediately preceding a certain state.
8. The motion recognition system of claim 7, wherein the removal unit comprises the following modules;
an information classification module: the system is used for classifying according to the information in the collected meta-actions and decomposing posture information and action information;
a searching module: for finding weak information;
a search module: the weak information database is used for searching the weak information database and judging whether the current weak information can be removed;
an integration module: and the method is used for adding the unremoved weak information into the strong information for information integration.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910253883.XA CN109993118B (en) | 2019-03-30 | 2019-03-30 | Action recognition method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910253883.XA CN109993118B (en) | 2019-03-30 | 2019-03-30 | Action recognition method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109993118A CN109993118A (en) | 2019-07-09 |
| CN109993118B true CN109993118B (en) | 2021-08-20 |
Family
ID=67131993
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910253883.XA Active CN109993118B (en) | 2019-03-30 | 2019-03-30 | Action recognition method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109993118B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112582063A (en) * | 2019-09-30 | 2021-03-30 | 长沙昱旻信息科技有限公司 | BMI prediction method, device, system, computer storage medium, and electronic apparatus |
| CN116108391B (en) * | 2023-04-12 | 2023-06-30 | 江西珉轩智能科技有限公司 | Human body posture classification and recognition system based on unsupervised learning |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104898831A (en) * | 2015-05-08 | 2015-09-09 | 中国科学院自动化研究所北仑科学艺术实验中心 | Human action collection and action identification system and control method therefor |
| CN107016342A (en) * | 2017-03-06 | 2017-08-04 | 武汉拓扑图智能科技有限公司 | A kind of action identification method and system |
| CN107290741A (en) * | 2017-06-02 | 2017-10-24 | 南京理工大学 | Combine the indoor human body gesture recognition method apart from time-frequency conversion based on weighting |
| CN108537200A (en) * | 2018-04-19 | 2018-09-14 | 佛山市长郡科技有限公司 | A kind of device and method for selectively collecting EEG data by action recognition |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9165196B2 (en) * | 2012-11-16 | 2015-10-20 | Intel Corporation | Augmenting ADAS features of a vehicle with image processing support in on-board vehicle platform |
| CN103593680B (en) * | 2013-11-19 | 2016-09-14 | 南京大学 | A kind of dynamic gesture identification method based on the study of HMM independent increment |
| CN103941869B (en) * | 2014-04-21 | 2017-07-14 | 云南电网公司普洱供电局 | A kind of body-sensing posture identification method based on action element |
| CN104699606B (en) * | 2015-03-06 | 2017-05-24 | 国网四川省电力公司电力科学研究院 | Method for predicting state of software system based on hidden Markov model |
| CN204759348U (en) * | 2015-05-08 | 2015-11-11 | 中国科学院自动化研究所北仑科学艺术实验中心 | Human action is gathered and is moved identification system |
| CN105446484B (en) * | 2015-11-19 | 2018-06-19 | 浙江大学 | A kind of electromyography signal gesture identification method based on Hidden Markov Model |
| CN108549856B (en) * | 2018-04-02 | 2021-04-30 | 上海理工大学 | Human body action and road condition identification method |
-
2019
- 2019-03-30 CN CN201910253883.XA patent/CN109993118B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104898831A (en) * | 2015-05-08 | 2015-09-09 | 中国科学院自动化研究所北仑科学艺术实验中心 | Human action collection and action identification system and control method therefor |
| CN107016342A (en) * | 2017-03-06 | 2017-08-04 | 武汉拓扑图智能科技有限公司 | A kind of action identification method and system |
| CN107290741A (en) * | 2017-06-02 | 2017-10-24 | 南京理工大学 | Combine the indoor human body gesture recognition method apart from time-frequency conversion based on weighting |
| CN108537200A (en) * | 2018-04-19 | 2018-09-14 | 佛山市长郡科技有限公司 | A kind of device and method for selectively collecting EEG data by action recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109993118A (en) | 2019-07-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jiang et al. | Multi-layered gesture recognition with Kinect. | |
| US8000500B2 (en) | System and method for analyzing of human motion based on silhouettes of real time video stream | |
| Ko et al. | A novel and efficient feature extraction method for iris recognition | |
| US10949702B2 (en) | System and a method for semantic level image retrieval | |
| Wang et al. | Automatic fall detection of human in video using combination of features | |
| CN111291865B (en) | Gait recognition method based on convolutional neural network and isolated forest | |
| CN105975932A (en) | Gait recognition and classification method based on time sequence shapelet | |
| US10949460B2 (en) | Product indexing method and system thereof | |
| CN109993118B (en) | Action recognition method and system | |
| Tafazzoli et al. | Genetic feature selection for gait recognition | |
| CN106778456A (en) | A kind of optimization method and device of handwriting input | |
| Wang et al. | Automatic shoeprint retrieval algorithm for real crime scenes | |
| CN110737788B (en) | Rapid three-dimensional model index establishing and retrieving method | |
| KR20190050551A (en) | Apparatus and method for recognizing body motion based on depth map information | |
| Safavipour et al. | A hybrid approach to multimodal biometric recognition based on feature-level fusion of face, two irises, and both thumbprints | |
| Chua et al. | Vision-based hand grasping posture recognition in drinking activity | |
| JP2019016268A (en) | Image processing apparatus, image processing method and image processing program | |
| CN113378691A (en) | Intelligent home management system and method based on real-time user behavior analysis | |
| CN109977890B (en) | Action recognition method and recognition system thereof | |
| CN112101293A (en) | Facial expression recognition method, device, equipment and storage medium | |
| CN106557523B (en) | Representative image selection method and apparatus, and object image retrieval method and apparatus | |
| Tsai et al. | VQ-HMM classifier for human activity recognition based on R-GBD sensor | |
| CN113657315B (en) | Quality screening method, device, equipment and storage medium for face image | |
| CN115713806A (en) | Falling behavior identification method based on video classification and electronic equipment | |
| CN110909678B (en) | Face recognition method and system based on width learning network feature extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20210804 Address after: Room b6-2005, No. 121, Zhongshan North 1st Road, Hongkou District, Shanghai 200080 Applicant after: SHANGHAI SHIZHUANG INFORMATION TECHNOLOGY Co.,Ltd. Address before: 100045 75 Yuetan South Street, Xicheng District, Beijing Applicant before: Qian Yifan |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |