CN101607138B - Action recognition method based on finite automata model - Google Patents
Action recognition method based on finite automata model Download PDFInfo
- Publication number
- CN101607138B CN101607138B CN2008100435135A CN200810043513A CN101607138B CN 101607138 B CN101607138 B CN 101607138B CN 2008100435135 A CN2008100435135 A CN 2008100435135A CN 200810043513 A CN200810043513 A CN 200810043513A CN 101607138 B CN101607138 B CN 101607138B
- Authority
- CN
- China
- Prior art keywords
- action
- sub
- son
- data
- finite automata
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an action recognition method based on a finite automata model, which comprises the following steps: defining a sub-action; dividing motion operation into a plurality of the sub-actions, and describing the sequence of the sub-actions by an action expression; dividing motion data acquired by an acceleration sensor into a plurality of continuous data subsets, and judging the data subsets as corresponding sub-actions to obtain the sequence of the sub-actions of the motion data; and introducing the finite automata model, and according to the sub-actions of the motion data and the action expression of the motion operation, performing action recognition. The action recognition method can process input action data in real time and improve the efficiency of the action recognition.
Description
Technical field
The present invention relates to the action induction technology, particularly a kind of action identification method based on finite automata model.
Background technology
Generally, can be divided into two kinds to the action induction technology according to the size of movement range: the induction of small action and action induction significantly, the induction of common small action is just like flight simulator etc., if the small variation of handle can both be experienced in game; Common significantly action induction is just like the electronic game machine Wii of Nintendo 2006 issue etc., significantly action as up/down/left/right etc., and common arm has action.
The application of action induction technology needs to realize in conjunction with corresponding hardware, take game paddle as example, for the small action induction technology of the first, needing in handle has the gyroscope that can measure rotational speed, only need to carry out the angular velocity of rotation that gyroscope is measured simple the processing at the software end and get final product.
For action induction technology significantly, corresponding hardware needs to measure XYZ three axis linear acceleration at least, can also increase the gyroscope of measuring angular velocity of rotation on this basis.
Significantly the dynamic analytical algorithm of action induction technology is take 3 axles of dynamic handle or 5 axle acceleration sensor influence values as input, by the software algorithm final analysis, goes out user's motion operation.
Significantly dynamic analytical algorithm also is in the starting stage at present, can only be generally some simple motion of identification, such as up, down, left, right, before and after etc.All need to spend the plenty of time and energy goes identification for the identification of some compound actions, and action of every interpolation is not always the case.And have a large amount of floating-point operations in algorithm, have a strong impact on efficiency, Latency response time.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of action identification method based on finite automata model, adopts this action induction method, can process in real time the input action data, the enhancing action recognition efficiency.
For solving the problems of the technologies described above, the action identification method based on finite automata model of the present invention comprises the following steps:
(1) definition action;
(2) motion operation is divided into some son actions, with the manual expression formula, describes its sub-action sequence;
(3) the dynamic data of acceleration transducer collection are divided into some continuous data subsets, each data subset is judged as corresponding son action, obtain the sub-action sequence of dynamic data;
(4) introduce finite automata model,, according to the son action of dynamic data and the manual expression formula of motion operation, carry out action recognition.
The sub-action sequence of the dynamic data that step (3) obtains can first carry out carrying out step (4) after the error shielding again.
Space displacement amount, the average acceleration that can move according to the son of dynamic data are carried out the error shielding.
Action identification method based on finite automata model of the present invention, introduce finte-state machine (being called for short FA) model, can identify complicated action.For compound action, it is divided into some continuous son actions, represent the action of this reality with sub-action sequence.This sequence is input to FA one by one, and FA changes through certain state according to the internal logic of oneself, the output action result of determination.Can process in real time the input action data, the enhancing action recognition efficiency.
Description of drawings
Below in conjunction with the drawings and the specific embodiments, the present invention is described in further detail.
Fig. 1 is the one embodiment process flow diagram of the action identification method based on finite automata model of the present invention;
Fig. 2 is a motion operation " circle clockwise " schematic diagram;
Fig. 3 is a sub-action definition mode example;
Fig. 4 is that a motion operation is divided son action example;
Fig. 5 is two seed action division methods examples to " circle clockwise ".
Embodiment
Significantly action induction based on finite automata model method of the present invention one embodiment as shown in Figure 1, comprises the following steps:
(1) definition action;
(2) motion operation is divided into some son actions, describes its sub-action sequence with mathematic(al) representation in configuration file;
(3) the dynamic data of the dynamic handle of acceleration transducer collection are divided into some continuous data subsets, each data subset is judged as corresponding son action, obtain the sub-action sequence of dynamic data;
(4) the sub-action sequence that obtains according to dynamic data judgement is carried out error shielding, the error of being brought to eliminate dynamic handle state induction and small shake according to space displacement amount, the average acceleration of son action;
(5) introduce finite automata model,, according to the son action of dynamic data and the sub-action sequence of motion operation of describing with mathematic(al) representation, carry out action recognition in configuration file.
Action identification method based on finite automata model of the present invention, introduce finte-state machine (being called for short FA) model, can identify complicated action.For compound action, it is divided into some continuous son actions, represent the action of this reality with sub-action sequence.This sequence is input to FA one by one, and FA changes through certain state according to the internal logic of oneself, the output action result of determination.Can process in real time the input action data, the enhancing action recognition efficiency.
No matter be 3 axles or the acceleration transducer of 5 axles, its action induction is to occur with the form of space 3 axle accelerations, and this just means that it does not have influence value when dynamic handle is made uniform motion in space, is as good as when static at this moment.In technical scheme of the present invention, can or only have the action of small induction to ignore non-inductive value.The present invention is applicable to the significantly action induction of 3 dimension spaces, but, for the ease of accurately describing technology of the present invention, below take 2 " circles clockwise " tieed up in action plane, as example, technical scheme of the present invention is described.
1. definition action
Can define 8 son actions in plane, be respectively " →
↑
↓
".But this is not unique definition, can be defined as 4,12 even more yet, is mainly by application program, the different demands of motion operation accuracy to be determined.The various definitions mode of son action as shown in Figure 3.
For the action identification method based on finite automata model of the present invention, the action of the son of definition is more, and the action that can identify is just more careful, more accurate.But in actual implementation procedure, the dynamic handle of for example playing, due to the error of dynamic handle hardware output parameter, and the restriction of the each side factors such as actual operation ability of PC, this degree of accuracy will be affected.Thereby can not unrestrictedly too much define the son action.Here hypothesis is defined as 8.
2. motion operation is divided into some height actions
A fundamental issue of action induction technology is " how identifying motion operation ".In the present invention,, based on defined son action in the 1st, whole motion operation is divided into some continuous sons moves.Again this sub-action sequence is mated finally.
A general method of dividing motion operation is based on the time and divides, and suppose motion operation as shown in Figure 4, and the curve in Fig. 4 represents the track of motion operation, and 2 dotted lines wherein are this division of operations 3 sections, namely are divided into 3 sons and move.According to defined son action in the 1st, can be judged to be 3 stages in Fig. 3 successively " →
→ ".The manual expression formula that this motion operation is described is pre-defined to be left in configuration file, after the dynamic handle of user is done an action, the dynamic data of the dynamic handle that acceleration transducer gathers can with predefined manual expression formula in configuration file " →
→ " coupling, just being considered to the target action, then application program just reads the description of manual expression formula in configuration file, set up FA dynamicly, and then carries out action recognition, carries out corresponding logic function.Method of the present invention might not need motion operation consistent completely, and certain fault-tolerance is arranged in this process.
For example for the clockwise circle described in Fig. 2, just can divide as shown in Figure 5:
Two kinds of divisions are arranged in Fig. 5:
(1) left side: be divided into 4 son actions, can be identified as successively
(2) the right: be divided into 4 son actions, can be identified as successively " → ↓ ← ↑ ".
These two kinds of zoned formats, can be as the standard of identification.
But, from motion operation " circle clockwise " itself, it is not only these two kinds.In fact can be refined as 4 types for each top division:
(1)①→②→③→④
(2)②→③→④→①
(3)③→④→①→②
(4)④→①→②→③
So in Fig. 5 representative be exactly 8 seed action sequences.As long as the action of the dynamic handle of user can with one kind of coupling, that just can be identified as " circle clockwise " and move.These contents in the actual development process all the available profile form present, and at the inner engine that needs to realize these action of configuration expression formulas of identification of program.
This action identification method of the present invention, except identifying easily compound action, also has the another one advantage.Need now to suppose this operation of identification " n time round clockwise ", so only need simple modification action of configuration expression formula to get final product, for example for " 2 times round clockwise ", can be expressed as " 1 → 2 → 3 → 4 → 1 → 2 → 3 → 4 ".
3. dynamic data are divided into some continuous data subsets
Two of fronts step is to build analytical framework for analyzing motion operation in fact., from this step, just can carry out concrete data analysis and action recognition.
The identification for action has also been mentioned in front, at first will carry out segmentation to the data based time of innervation.After the son action of dynamic data is judged in segmentation, the more continuous sub-action sequence of dynamic data is identified.
The acceleration transducer of dynamic handle is at x for the output of action induction, y, and the acceleration on z three direction of principal axis, be designated as (ax, ay, az), thereby at first need these data are converted into the point set of corresponding space motion path.
Suppose to be sampled as 100 times/S, son action split time is 50ms, if be used for completing " circle clockwise " this action, needs 250ms, can be divided into so the data subset of 5 son actions.Then the data acquisition of every height action is carried out linear fit (linear because the son action in the above in second point all is defined as, tool has great advantage on arithmetic speed).
4. shield the error that state induction and small shake bring
Carry out in the motion operation process unavoidable some maloperation at dynamic handle, for example the user holds handle and keeps motionless, but in fact be difficult to guarantee fully static, acceleration transducer some influence value always more or less.Thereby for the processing of the error of this class, be necessary.
Demand for action induction in some application programs is the significantly action of type, and the user also can be put into the scope of shielding if any a small amount of small-sized action so.
Also have in addition a kind of situation, when a motion operation finishes, from space motion path, be almost static.But the angle of theorem of kinetic energy from mechanics, have a reverse impulse force that influence value is larger during release, and what show in dynamic data is exactly the larger values of a series of influence values.This also need to deshield.
In the present invention, can carry out the error shielding according to the son action, dual mode is roughly arranged:
(1) the space displacement amount of son action
(2) average acceleration of son action
5. introduce finte-state machine (FA) Model Identification motion operation
Finte-state machine is in computer theory, can process rationally and effectively a kind of model of complex state conversion logic, following 5 parts of its formal definition: state set, alphabet, transfer function, initial state and final state.
Introduce in the present invention finite automata model, with general action induction technology, compared following advantage:
(1) process in real time input.
Common action induction technology, in the time of need to waiting until that motion operation is completed when processing the dynamic data of induction, could process data.And in the present invention, the real time data processing ability in the enhancing action recognition technology greatly, the generation of each height action can, as the input of FA, change the internal state of FA in real time.
(2) enhancing action recognition efficiency.
The needed time of FA identification maneuver is depended on the quantity of son action.For example need 250ms completing a motion operation, be 50ms every sub-actuation time, will produce a sub-actuating signal every 50ms so., for FA, can receive the sub-actuating signal of an input every 50ms, thereby order about the variation of a series of states.
(3) effectively organize all kinds of motion operation.
For each motion operation, owing to itself there being multiple situation, as " circle clockwise " this operation of mentioning in the 2nd.Generally, can mate one by one during identification.But introduce in the present invention the FA model, just do not needed.Because " circle clockwise " this division of operations is 4 son actions,, so for the identification of this action, be exactly to be determined by sub-amount of action, be fixed as 4, there are several situations irrelevant with this operation.
(4) simplify development process.
In the specific implementation process, at first adopt manual expression formal description motion operation, its sub-action sequence is namely described, then in inside, realize building the engine of FA.At first application program reads in configuration file the manual expression form for the description of motion operation, and the dynamic FA that sets up then subsequently just can carry out action recognition.
When needs increased identification to new motion operation, the expression formula description that only need to add this action in configuration file got final product, and does not need to revise any module.
Claims (3)
1. the action identification method based on finite automata model, is characterized in that, comprises the following steps:
(1) definition action;
(2) motion operation is divided into some son actions, with the manual expression formula, describes its sub-action sequence;
(3) the dynamic data of acceleration transducer collection are divided into some continuous data subsets, each data subset is judged as corresponding son action, obtain the sub-action sequence of dynamic data;
(4) introduce finite automata model,, according to the son action of dynamic data and the manual expression formula of motion operation, carry out action recognition.
2. the action identification method based on finite automata model according to claim 1, is characterized in that, the sub-action sequence of the dynamic data that step (3) is obtained carries out carrying out step (4) after the error shielding again.
3. the action identification method based on finite automata model according to claim 2, is characterized in that, according to space displacement amount, the average acceleration of the action of the son of dynamic data, carries out the error shielding.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100435135A CN101607138B (en) | 2008-06-17 | 2008-06-17 | Action recognition method based on finite automata model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2008100435135A CN101607138B (en) | 2008-06-17 | 2008-06-17 | Action recognition method based on finite automata model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101607138A CN101607138A (en) | 2009-12-23 |
| CN101607138B true CN101607138B (en) | 2013-11-20 |
Family
ID=41481146
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2008100435135A Active CN101607138B (en) | 2008-06-17 | 2008-06-17 | Action recognition method based on finite automata model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101607138B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218062A (en) * | 2013-04-24 | 2013-07-24 | 伍斌 | Man-machine interaction method and equipment based on acceleration sensor and motion recognition |
| CN109189218B (en) * | 2018-08-20 | 2019-05-10 | 广州市三川田文化科技股份有限公司 | A kind of method, apparatus of gesture identification, equipment and computer readable storage medium |
| CN113476833A (en) * | 2021-06-10 | 2021-10-08 | 深圳市腾讯网域计算机网络有限公司 | Game action recognition method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1153576A (en) * | 1995-05-10 | 1997-07-02 | 任天堂株式会社 | Operating device with analog joystick |
| CN1474424A (en) * | 1994-05-09 | 2004-02-11 | �ֹ��� | Controller unit for electronic device |
| CN1770064A (en) * | 2005-07-15 | 2006-05-10 | 中国海洋大学 | Interactive input device for computer |
| CN1996208A (en) * | 2007-01-15 | 2007-07-11 | 解晖 | Man-machine command input device and mapping method for motion information in the same |
-
2008
- 2008-06-17 CN CN2008100435135A patent/CN101607138B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1474424A (en) * | 1994-05-09 | 2004-02-11 | �ֹ��� | Controller unit for electronic device |
| CN1153576A (en) * | 1995-05-10 | 1997-07-02 | 任天堂株式会社 | Operating device with analog joystick |
| CN1770064A (en) * | 2005-07-15 | 2006-05-10 | 中国海洋大学 | Interactive input device for computer |
| CN1996208A (en) * | 2007-01-15 | 2007-07-11 | 解晖 | Man-machine command input device and mapping method for motion information in the same |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101607138A (en) | 2009-12-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103335814B (en) | Correction method for inclination angle measurement error data of experimental model in wind tunnel | |
| CN103699715A (en) | Fluid-solid coupling method based on smoothed-particle hydrodynamics (SPH) and nonlinear finite elements | |
| CN113283082B (en) | Centroid track generation method, centroid track generation device, computer readable storage medium and robot | |
| CN101607138B (en) | Action recognition method based on finite automata model | |
| EP1966687A2 (en) | Digital effects analysis in modeling environments | |
| CN112506759A (en) | Automatic testing method and device for servo system control software and storage medium | |
| CN105353789A (en) | Continuous vibration signal time history replication control method | |
| CN100470418C (en) | Polypropylene melting index softsensoring instrument based on blind signal analysis and method thereof | |
| CN104102127A (en) | Airborne aerodynamic parameter identification system | |
| CN102306054B (en) | Attitude sensing equipment and positioning method and device thereof and method and device for controlling mouse pointer | |
| CN114211491A (en) | Collision detection method and device, computer readable storage medium and robot | |
| CN101762252A (en) | Virtual three-coordinate measuring instrument and use method thereof | |
| CN100480973C (en) | Method and apparatus for detecting position of movable device | |
| CN116451470A (en) | Robot energy consumption modeling method and device based on multiple linear regression and dynamics | |
| CN108972536A (en) | The kinetic parameter of mechanical arm determines system, method and storage medium | |
| CN109202883A (en) | A kind of position control method of self-balance robot | |
| CN113343458A (en) | Model selection method and device for engine sensor, electronic equipment and storage medium | |
| CN111475940B (en) | Flexible baseline dynamic prediction method based on fiber bragg grating sensor and wing mode | |
| CN113919170A (en) | Working condition environment simulation method and system based on multi-sensor fusion | |
| CN109795592A (en) | Riding posture recognition methods and application apparatus | |
| CN116738759B (en) | Method and device for designing and operating equipment, computer equipment and readable storage medium | |
| CN203191755U (en) | Multifunctional control instrument | |
| CN107730712A (en) | A kind of interference positioning method and device | |
| CN114218139B (en) | Simulation turntable high-speed synchronous acquisition method based on real-time operating system and FPGA | |
| CN103196446B (en) | Intelligent gyro signal filtering method of trapdown inertial navigation system in high overload environment |
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 | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20190228 Address after: 201203 7, 1 Lane 666 lane, Zhang Heng Road, Pudong New Area, Shanghai. Patentee after: SHANGHAI ZHANGMEN TECHNOLOGY CO., LTD. Address before: 201203 No. 356 GuoShoujing Road, Pudong New Area, Shanghai Patentee before: Shengle Information Technology (Shanghai) Co., Ltd. |
|
| TR01 | Transfer of patent right |