CN105279526B - Divide the method and apparatus of track - Google Patents

Abstract

The invention discloses a kind of method and apparatus for dividing track.The track dividing method includes: obtaining step, for obtaining multiple track sampled points；Segmentation step, for dividing the track and obtaining 1 dividing candidate points；Characteristic extraction step, for extracting one group of feature for describing the relativeness between at least two dividing candidates point；Classifying step, for at least two dividing candidates point to be classified as true cut-point or false cut-point by classifier based on extracted one group of feature.

Description

Divide the method and apparatus of track

Technical field

The present invention relates generally to track identification fields, and in particular to divides the method and device thereof of track.

Background technique

In recent years, track identification is widely used in man-machine interactive system.The technology of gesture identification allows users to It is instructed to transmissions such as TV, computer, game cartridge and mobile phones.Character, sentence or coded command can be expressed by gesture.In order to It decodes the meaning of behind or controls the further operating of purpose equipment, need to identify these gestures.

Existing track recognizing method usually requires that the position of significant track as input.It is, however, required that this input It is feasible usually unrealistic, so that being difficult to configure the system based on track identification in many real-world scenes.Therefore, Segmentation input trajectory is very important to distinguish significant section and non-targeted section, and will influence next identification As a result.

The prior art 1 [document 1] proposes a kind of automated process to divide 3D hand track and transcribe based on these hand tracks Phoneme, in this, as the step of identifying American Sign Language.Become first using partitioning algorithm detection minimum speed and the maximum of deflection Change, to divide the hand exercise track of Natural check sentence.This generates by trained naive Bayesian detector into One step processing with distinguish true cut-point and eliminate false alarm over-segmentation track.

The problem to be solved in the present invention includes the target-dependent and limitation of local feature.Naive Bayesian network is used to The probability of query node is inferred in the case where the observation of four nodes of the local message of given description segmentation candidates point.Institute Four nodes of observation are by conditional probability table (CPT) Lai Daibiao, and the conditional probability table is by using all nodes during training Value maximal possibility estimation and be trained to.The calculating of CPT depends on target (i.e. significant orbit segment).When change target When, it needs to recalculate CPT.

For example, providing the continuous path for indicating " 246 " sequence in the training stage.In test phase, " 135 " sequence is provided Column.Since the prior art 1 has target-dependent, the prior art 1 only learns how to separate 2,4 and 6.In test phase, Change target.Therefore, how to separate untrained target 1,3 and 5 have it is to be solved.The CPT scheme of the prior art 1 determines its mesh Mark dependence.

Illustrative trace is as shown in Figure 1.After segmentation, three cut-points (SP) are obtained.Two are true cut-points, such as SP1 and Shown in SP2；Another is noise, or referred to as false alarm.The prior art 1 consider local feature (such as local velocity or Local direction angle).As shown in Figure 1, SP1, SP2 and noise spot all have Local Minimum speed and local maxima deflection.Therefore, How the minor change of noise spot and other two true cut-points to be distinguished still unresolved.

The prior art 2 [document 2] proposes the track dividing method based on CHMM (cascade hidden Markov model).It is first First, the training gesture model of the prior art 2 and transition gesture model, then definition cascades hidden Markov model (HMM) structure.It connects , compare the likelihood between gesture model and transition model to obtain terminal, and traced to obtain using viterbi algorithm To starting point.This method needs to trace back to the segmentation candidates point of front, therefore it is very high to calculate cost.

As described above, it is still desirable to being capable of effective Ground Split track and the target independence to noise and interference with robustness Property method.

[citation list]

1.W.W.Kong and Surendra Ranganath,Automatic Hand Trajectory Segmentation and Phoneme Transcription for Sign Language,8th IEEE International Conference on Automatic Face and Gesture Recognition,2008

2.Hee-Deok Yang,A-Yeon Park,and Seong-Whan Lee,Gesture Spotting and Recognition for Human-Robot Interaction,IEEE Transactions on Robotics,Vol.23, No.2,April2007

Summary of the invention

It is proposed the present invention at least one of in view of the above problems.

In an aspect, a kind of track dividing method is provided comprising: obtaining step is adopted for obtaining multiple tracks Sampling point；Segmentation step, for dividing the track and obtaining 1 dividing candidate points；Characteristic extraction step, for mentioning Take one group of feature for describing the relativeness between at least two dividing candidates point；Classifying step is extracted for being based on One group of feature at least two dividing candidates point is classified as by true cut-point or false cut-point by classifier.

In one embodiment, one group of feature includes: to represent the feature or representation speed characteristic of dimensional properties Feature, or represent the feature of directional characteristic.

In one embodiment, the relativeness is the relativeness between adjacent segmentation candidate point；Or it is described Relativeness is the relativeness between non-conterminous dividing candidate point.

In one embodiment, the classifier is generated by following steps: obtaining step, for obtaining multiple training Track sampled point；Segmentation step, for dividing the trained track and obtaining 1 training dividing candidate points；Obtain step Suddenly, for obtaining the true value of at least two training dividing candidate point, wherein each expression in the true value is instructed accordingly Practicing dividing candidate point is true cut-point or false cut-point；Extraction step is waited for extracting the description at least two training segmentation One group of feature of relativeness between reconnaissance；Training step, for being waited based on the description at least two training segmentation The true value of one group of feature of relativeness between reconnaissance and at least two training dividing candidate point, Lai Xunlian Classifier.

In one embodiment, the relativeness is between the identical at least two training dividing candidate point of true value Relativeness；Or the relativeness is the relativeness between 1 different training dividing candidate points of true value.

In one embodiment, the relativeness is the adjacent relativeness really trained between cut-point；Or The relativeness is the relativeness between adjacent vacation training cut-point.

In an aspect, a kind of track recognizing method is provided comprising: segmentation step, by input trajectory be divided into A few orbit segment；Identification step identifies at least one described orbit segment.

In another aspect, a kind of track segmenting device is provided comprising: acquiring unit is configured as obtaining multiple rails Mark sampled point；Cutting unit is configured as dividing the track and obtains 1 dividing candidate points；Feature extraction list Member is configured as extracting one group of feature for describing the relativeness between at least two dividing candidates point；Taxon, quilt Be configured to extracted one group of feature by classifier by at least two dividing candidates point be classified as true cut-point or False cut-point.

In another aspect, a kind of track identification device is provided comprising: cutting unit is configured as input trajectory It is divided at least one orbit segment；Recognition unit is configured as identifying at least one described orbit segment.

In another aspect, a kind of display equipment is provided comprising: track generation unit is configurable to generate at least one A track；Track identification device；It is corresponding with the result of the track identification device interior to be configured as display for display unit Hold.

In one embodiment, the display equipment is mobile phone, camera, game cartridge, television set or computer Any of.

Detailed description of the invention

The attached drawing being included in the description and forms part of the description instantiates the embodiment of the present invention, and with text Word illustrates principle used to explain the present invention together.

Fig. 1 diagrammatically illustrates the example of segmentation track.

Fig. 2 is the schematic block diagram according to the first exemplary system configuration that can implement the embodiment of the present invention.

Fig. 3 is the schematic block diagram according to the second exemplary system configuration that can implement the embodiment of the present invention.

Fig. 4 is to illustrate the block diagram of the exemplary hardware arrangement of calculating equipment 220 in figure 2 and figure 3.

Fig. 5 is to illustrate the general flow figure of track dividing method according to the present invention.

Fig. 6 is the flow chart for illustrating classifier according to the present invention and generating processing.

Fig. 7 A to 7D, which is instantiated, wants divided two tracks.Fig. 7 A instantiates twice the x-axis space-time curve of gesture to the left. Fig. 7 B instantiates x, the corresponding track 2D on y plane.Fig. 7 C instantiates the x-axis of alternating gesture to the left, to the right, to the left and to the right Space-time curve.Fig. 7 D instantiates x, the corresponding track 2D on y plane.

Fig. 8 A to 8D instantiates the primary segmentation result for Fig. 7 A to 7D.

Fig. 9 A to 9C instantiates the example calculation of scale feature, velocity characteristic and direction character.

Figure 10 A to 10D instantiates the final segmentation result of the track for Fig. 7 A into 7D.

Figure 11 is the functional configuration of track segmenting device according to the present invention.

Specific embodiment

Detailed description of the present invention exemplary embodiment below with reference to accompanying drawings.It should be noted that being described below substantially only It is illustrative and exemplary, and is in no way intended to limit the present invention and its application or purposes.Unless otherwise expressly specified, In The relative configuration of component and step described in the present embodiment, numerical expression and numerical value does not limit the scope of the invention.In addition, Technology well known by persons skilled in the art, method and apparatus can be not discussed in detail, but be confirmed as this explanation in due course A part of book.

Then, it will illustrate details of the invention.It has been observed that the relativeness between segmentation candidates point makes with distinguishing characteristics True cut-point and false alarm can be distinguished by obtaining.Firstly, term involved in defining.Track includes target phase and non-targeted section.Mesh Bid section is meaningful.Rather than target phase is transitional or full of noise.

For example, operator forms the track including " 8642 " character string with a finger.Due to desired character The stroke of " 8 ", " 6 ", " 4 " or " 2 " is bent, so forming four orbit segments of each independent character " 8 ", " 6 ", " 4 " or " 2 " Bending.After the gesture that operator completes a character, in general, his finger will directly and quickly move on to character late Initial point.Orbit segment in this transient process is generally less bent, and completes the orbit segment with higher average speed.

These features are very unique to distinguish true cut-point and form the point of character.Therefore, reflection segmentation candidates point is utilized Between relativeness these features come accurate Ground Split track be effective.Certainly, above situation is merely exemplary, and Target is not limited to number and character.

The present invention has excavated the feature between the target of target transition period, and considers the pass between segmentation candidates point The feature of system, these features have bigger scale in test trails.

Since our method has quantified reflection segmentation candidates point (that is, the feature between target, rather than the spy in target Sign) between relativeness feature, therefore our method do not have target-dependent.

Continue to continue to use example above-mentioned to illustrate method of the invention.Instruction " 246 " sequence is provided in the training stage Continuous path.In test phase, " 135 " sequence is provided.Even if not learning any test target in the training stage, the present invention Method still be able to generate correct segmentation.

In addition, interference or the minor change in relation to local feature as caused by noise all have an impact very much.However, to retouching These interference for stating the feature of relativeness between segmentation candidates point are suitably inhibited by method of the invention.

Turning now to specific implementation of the invention.First, it should be noted that the present invention be capable of handling various tracks (2D or 3D)。

Fig. 2 is the schematic block diagram according to the first exemplary system configuration that can implement the embodiment of the present invention.It is hand-held Equipment 200 includes alignment sensor 210, the calculating equipment 220 connecting with 210 and the operating unit 230 connecting with 220.

Alignment sensor 210 can be gyroscope, accelerometer, depth transducer or RGB sensor.Alignment sensor 210 Obtain the track related data of handheld device 200.Calculating equipment 220 can be the form of IC chip, compact and be easy to It is embedded in handheld device 200.Operating unit 230 is connected to 220.In typical applications, handheld device 200 can be photograph Camera or mobile phone.The track data that sensor 210 will acquire, which is sent to, calculates equipment 220.It calculates equipment 220 and divides rail The control command of hint to identify track, and is output to operating unit 230 by mark.Operating unit 230 can have various shapes Formula, this depends on the purpose of application.For example, operating unit 230 can be display.When the user for holding equipment 200 make to When the gesture moved left, the display of equipment 200 will show next picture.When the user for holding equipment 200 makes to the right When mobile gesture, the display of equipment 200 will show previous picture.

Fig. 3 is the schematic block diagram according to the second exemplary system configuration that can implement the embodiment of the present invention.Second Sensor 210, calculating equipment 220 and the operating unit 230 of exemplary system physically disperse, and connect as shown in Figure 3, Rather than it concentrates in the handheld device.Structure shown in Fig. 3 can use in such as game cartridge, projector or gesture control The application such as television set in.

By taking projector applications as an example.User's hand-held remote controller makes various gestures to control projected picture.Alignment sensor In 210 insertion remote controlers.Sensor 210 obtains track related data, and is sent out track related data by computer network 250 It is sent to and calculates equipment 220.It calculates equipment 220 and divides track, to identify track, and the control command of hint is output to operation Unit 230.In this example, operating unit 230 corresponds to projector.The projector will implement what the track by remote controler implied Control command.

Fig. 4 is the block diagram of the exemplary hardware arrangement of the calculating equipment 220 in diagrammatic illustration 2 or Fig. 3.

Input/output (IO) interface is transmitted through from alignment sensor 210 to the track related data for calculating equipment 220 310 promote, which, which can be, meets universal serial bus (USB) standard and have corresponding The universal serial bus of USB connector.Track data can also be from can include being locally stored for SIM card, SD card, USB memory card etc. Equipment 240 is downloaded.

The track data is obtained by I/O interface 310 and is sent to memory 350.Arrangement processor 320 is deposited with retrieving Store up the software program of the published method in memory 350.Processor 320 can also be arranged to fetch, decode and execute basis All steps of published method, such as the step shown in Fig. 5,6,8,9 and 10.Processor 320 uses system bus 330, The result of each operation is recorded in memory 350.In addition to memory 350, output also can be via I/O interface 360 more forever It is stored on memory devices 240 long.Alternatively, output may be used as the order of control operating unit 230.At certain In a little situations, operating unit 230 can be the display of game cartridge, projector or gesture control TV, and use sound Frequently/video interface 368 sends output to operating unit 230.

Calculating equipment 220 can be various forms, for example, being embedded in the processing system in the handheld device in Fig. 2, or figure Independent calculating equipment 220 in 3 may remove one or more unnecessary components, or increase one or more additional Component.

It hereinafter, will be referring to first embodiment and attached drawing detailed description of the present invention track dividing method.

In the present embodiment, the track dividing method to gesture is proposed, two continuous gesture paths are especially to discriminate between. Gesture to the left is repeatedly presented in one track；Another track alternately present to the left, to the right, to the left, the gestures such as to the right.

Fig. 5 schematically shows the general flow figure of track dividing method according to the present invention.In the step s 100, it obtains Step is taken, multiple track sampled points are obtained.As described above, alignment sensor 210 obtains the track dependency number of handheld device 200 According to.Then, which is converted into the coordinate of track sampled point and corresponding temporal information.

Fig. 7 A instantiates twice the x-axis space-time curve of gesture to the left.Fig. 7 B instantiates x, the corresponding track 2D on y plane. Fig. 7 C instantiates the x-axis space-time curve of alternating gesture to the left, to the right, to the left and to the right.Fig. 7 D instantiates x, the phase on y plane Answer the track 2D.Much less, Fig. 7 A and Fig. 7 C can replace with y-axis space-time curve.

The embodiment has very high industrial applicability.When user wants from the image of camera browsing shooting, mention The corresponding two sets of browsings gesture of the browsing instructions different from two is supplied.Fig. 7 A and Fig. 7 B indicate in order image browsing one by one； In contrast, first image of Fig. 7 C and Fig. 7 D instruction browsing, then second image is browsed, then alternately browse first figure Picture and the second last image.

In step s 200, the track in difference segmentation figure 7B and Fig. 7 D.Cut-point (SP) candidate can sample from track It is selected in point.Track generally includes target phase and non-targeted section.For next track identification, the purpose of segmentation is in target phase Between carry out divide and divided between target phase and non-targeted section.

It is candidate that the method proposed obtains initial SP based on the motion change of space-time curvature is used.To in Fig. 7 A or Fig. 7 C The analytical table of space-time curvature of space-time curve suddenly change is illustrated position in continuous track occurs.If space-time curvature Greater than given threshold value, then there is motion change, to obtain SP candidate.

Processing in this way, the initial SP for obtaining Fig. 7 B and Fig. 7 D respectively are candidate.These SP candidates are by Fig. 8 A into 8D Open circles illustrate.It note that the method for not limiting and obtaining initial SP candidate.The space-time curvature threshold method proposed only shows Example property.Unquestionably, other dividing methods are also suitable.

The result of step S200 be can be seen that into 8D from Fig. 8 A by over-segmentation, thus the input as identification in future It is undesirable.In addition, implementation steps S300, is divided into true SP and false alarm for SP candidate.Extract description SP candidate's Between relativeness one group of feature.It will be referring to the original for illustrating feature extraction by the processing of the track defined Fig. 7 A and Fig. 7 B Reason.It carries out in an identical manner by the processing of the track defined Fig. 7 C and Fig. 7 D, and does not need to provide here duplicate thin Section.

1. two scale features.As shown in Figure 9 A, arc length 1 is between given SP and its previous SP, and arc length 2 is being given Between fixed SP and its latter SP.

It note that scale feature is not limited to arc length, also can choose between such as given SP and its adjacent SP The other sizes of linear distance.

2. two velocity characteristics: above-mentioned two arc length is divided by respective time interval.Fig. 9 B is x-axis space-time curve.Scheming As can be seen that T1 is the time interval from previous SP to given SP in 9B, and T2 is from given SP to next SP Time interval.Then two velocity characteristics are calculated as follows.

V1=ArcLength1/T1 (1)

V2=ArcLength2/T2 (2)

Since scale feature is not limited to arc length, correspondingly velocity characteristic is also unrestricted.Also other be can choose Velocity characteristic, as long as the relativeness between these velocity characteristics reflection SP candidate.

The example of other velocity characteristics is listed as follows.

V1=Distance1/T1 (3)

V2=Distance2/T2 (4)

3. both direction feature: as shown in Figure 9 C, angle theta 1 and straight line 2 and horizontal line between straight line 1 and horizontal line Between angle theta 2.In addition, given SP connect by straight line 1 with its previous SP, and straight line 2 is by given SP and its latter SP connection.

It is to be understood that direction character is not limited to situation described above.Also it may be selected that between reflection SP candidate Other direction characters of relativeness.For example, tangent line 1 and arc 1 are tangent, and tangent line 2 and arc 2 are tangent.θ 1 can be tangent line 1 and hang down Angle between straight line, and θ 2 can be the angle between tangent line 2 and vertical line.Here, horizontal line is replaced by vertical line, this Both it is used as reference line.

In the examples described above, the feature for describing the relativeness between adjacent SP candidate is extracted.In other embodiments In, the feature for describing the relativeness between non-conterminous SP candidate can be extracted.To the form of the feature of extraction there is no limit, As long as which feature provides the differences distinguished between true cut-point and spurious alarm.

Fig. 5 is returned to, in step S400, is implemented by classifier for time will to be divided based on extracted one group of feature Reconnaissance is classified as the classifying step of true and false alarm.

Plain mode using extracted feature is to be characterized value scheduled threshold value is arranged.Can be by SP candidate classification True or false alarm.The method that threshold value is arranged is the simple form of classifier.

In fact, track is often complicated.In order to obtain good and adaptive classification results, classifier can be Line or off-line training.The exemplary process diagram for generating classifier is as shown in Figure 6.

In step slo, training track sampled point is obtained.In step S20, divides training track and obtain at least two Training SP is candidate.In step s 30, the true value (ground truth) of training SP candidate is obtained.In step s 40, instruction is extracted Practice one group of feature between SP candidate.Finally, in step s 50, one group of feature and training SP based on training SP candidate are candidate True value train classifier.

Preferably, implementation steps S30 can be carried out in another way.In step slo, while trained track sampling is obtained Point and their own true value.Then, in step S20, it is candidate to generate training SP.Therefore, in step s 30, according in step The true value of the training SP candidate obtained in rapid S10 will train SP candidate to be labeled as true or false.

To the form of classifier there is no limit, if in the case where the extraction feature of given correlation SP candidate, classifier It is true or false alarm that SP candidate, which can be exported,.In one embodiment, classifier is SVM (support vector machines) classification Device.In another embodiment, classifier is neural network classifier.

In the case where classifier training method, it is preferable that relativeness is the identical at least two training segmentation of true value Relativeness between candidate point；Or relativeness is opposite between 1 different training dividing candidate points of true value Relationship.

In the case where classifier training method, it is preferable that relativeness is opposite between adjacent very trained cut-point Relationship.

In the case where classifier training method, it is preferable that relativeness is opposite between adjacent false training cut-point Relationship.

In addition, the invention discloses a kind of track recognizing methods.This method includes two key steps.First, according to preceding It states any one track dividing method and input trajectory is divided at least one orbit segment.Second, identify at least one rail Mark section.

By above-mentioned processing, as shown in Figure 10 A to 10D, accurate Ground Split track.In Figure 10 A into 10D, Filled Rectangle Indicate final segmentation result.

Figure 11 is the functional configuration of track segmenting device according to the present invention.It track segmenting device and is included therein Unit can be configured by any hardware, firmware, software or their any combination, as long as the unit in the track segmenting device It can be realized the function of the corresponding steps of track dividing method as above.

If device 1100 is by software section or completely, configuration, the software are stored in the memory of computer (example Such as the memory 350 in Fig. 4), and computer can be in the processor (such as component 320 in Fig. 4) of computer by holding The software of row storage when being handled, realizes the function of track segmentation of the invention.In in other respects, device 1100 can By hardware or firmware portions or configuration completely.Device 1100 can be incorporated to as functional module and calculate equipment 220.

Track segmenting device 1100 may include acquiring unit 1101, cutting unit 1102,1103 and of feature extraction unit Taxon 1104.Acquiring unit 1101 is configured as obtaining multiple track sampled points；Cutting unit 1102 is configured as dividing The track simultaneously obtains 1 dividing candidate points；Feature extraction unit 1103 is configured as extracting description described at least two One group of feature of relativeness between dividing candidate point；Taxon 1104 is configured as being based on extracting one group of feature by dividing At least two dividing candidates point is divided into true or false cut-point by class device.

The invention also discloses track identification devices comprising cutting unit and recognition unit.Cutting unit is configured as Input trajectory is divided at least one orbit segment according to track segmenting device 1100；It is described extremely that recognition unit is configured as identification A few orbit segment.

Note that unit described above can represent can be various forms (for example, firmware, hardware, software or they Any combination) and implement corresponding steps in the method described earlier function any part.

[experimental result]

In order to show effect of the invention, the experiment of the performance for showing the method according to above-described embodiment is carried out.

Continuous gesture data set is tested.The data set is collected from 10 demonstrators.Each demonstrator has done 12 Continuous gesture.Each continuous gesture includes at least one significant gesture section and other non-targeted gesture sections.Solution uses SP candidate is divided into true cut-point and false alarm by SVM classifier.Term definition is listed as follows.TP indicates true positives；FP is indicated False positive；FN indicates false negative.Nicety of grading is defined by formula (5) and (6).

Test result shows that while keeping accuracy rate is 85% or more, recall rate is 90% or more.This means that this The method of invention can reach very high recall rate while false positive is maintained at a very low level.The performance also turns out The validity of the principle of the present invention, that is to say, that using the feature between target, alternatively, consider description segmentation candidates point it Between relativeness feature.

Since the calculating of disclosed method is at low cost, the speed to a SP candidate classification is about 50ms, this energy Enough meet the needs of calculating in real time, and makes it possible to use gesture and control further real-time task.

[industrial feasibility]

The present invention can be used in many applications.For example, the present invention can be implemented in the display device, the display equipment packet Include track generation unit, track identification device disclosed above and display unit.Track generation unit is configurable to generate at least One track；Display unit is configured as display content corresponding with the result of track identification device.

The display equipment can be following any form: mobile phone, camera, game cartridge, television set or calculating Machine.

It note that method and apparatus described in this specification can be as software, firmware, hardware or theirs is any Combination is to implement.Certain components can be used as the software that runs on digital signal processor or microprocessor for example to implement. Other component can be used as hardware and/or specific integrated circuit for example to implement.

Methods and apparatus of the present invention can be executed in many ways.For example, can by software, hardware, firmware or Their any combination executes methods and apparatus of the present invention.The sequence of the above-mentioned steps of this method is merely to illustrate, and The step of method of the invention, is not limited to the sequence of above-mentioned specific descriptions, unless otherwise specified.In addition, in some embodiments In, the present invention is also used as recording program in the recording medium to implement (including for executing according to the method for the present invention Machine readable instructions).Therefore, the invention also includes storages for executing record Jie of program according to the method for the present invention Matter.

Although some specific embodiments of the present invention is demonstrated with example in detail, those skilled in the art should be managed Solution, above-mentioned example are only for illustrative, and do not limit the scope of the invention.It should be appreciated by those skilled in the art, Above-described embodiment can be modified without departing from the scope and spirit of the present invention.The scope of the present invention is by appended right It is required that limiting.

Claims (16)

1. a kind of track dividing method comprising:

Obtaining step, for obtaining multiple track sampled points；

Segmentation step, for dividing the track and obtaining 1 dividing candidate points；

Characteristic extraction step, for extracting one group of feature for describing the relativeness between at least three dividing candidates point, Wherein one group of feature is the distinguishing characteristics that can distinguish true cut-point and false alarm；

Classifying step, for being classified as at least three dividing candidates point true minute by classifier based on one group of feature Cutpoint or false alarm.

2. track dividing method according to claim 1, wherein one group of feature includes:

The feature of dimensional properties is represented, or

The feature of representation speed characteristic, or

Represent the feature of directional characteristic.

3. track dividing method according to claim 1 or 2, wherein the relativeness be adjacent segmentation candidate point it Between relativeness；Or the relativeness is the relativeness between non-conterminous dividing candidate point.

4. track dividing method according to claim 1, wherein the classifier is generated by following steps:

Obtaining step, for obtaining multiple trained track sampled points；

Segmentation step, for dividing the trained track and obtaining 1 training dividing candidate points；

Obtaining step, for obtaining the true value of at least three training dividing candidate point, wherein each table in the true value Show that corresponding training dividing candidate point is true cut-point or false alarm；

Extraction step, for extracting the one group of spy for describing the relativeness between at least three training dividing candidate point Sign, wherein one group of feature is the distinguishing characteristics that can distinguish true cut-point and false alarm；

Training step, it is special for described one group based on the relativeness described between at least three training dividing candidate point The true value of sign and at least three training dividing candidate point, Lai Xunlian classifier.

5. track dividing method according to claim 4, wherein the relativeness is identical at least two instruction of true value Practice the relativeness between dividing candidate point；Or

The relativeness is the relativeness between 1 different training dividing candidate points of true value.

6. track dividing method according to claim 4, wherein the relativeness be it is adjacent really train cut-point it Between relativeness；Or

The relativeness is the relativeness between adjacent vacation training cut-point.

7. a kind of track recognizing method comprising:

Segmentation step, according to claim 1 to track dividing method described in any one in 6 by input trajectory be divided into A few orbit segment；

Identification step identifies at least one described orbit segment.

8. a kind of track segmenting device comprising:

Acquiring unit is configured as obtaining multiple track sampled points；

Cutting unit is configured as dividing the track and obtains 1 dividing candidate points；

Feature extraction unit is configured as extracting one group of spy for describing the relativeness between at least three dividing candidates point Sign, wherein one group of feature is the distinguishing characteristics that can distinguish true cut-point and false alarm；

Taxon is configured as being classified as at least three dividing candidates point by classifier based on one group of feature True cut-point or false alarm.

9. track segmenting device according to claim 8, wherein one group of feature includes:

The feature of dimensional properties is represented, or

The feature of representation speed characteristic, or

Represent the feature of directional characteristic.

10. track segmenting device according to claim 8 or claim 9, wherein the relativeness be adjacent segmentation candidate point it Between relativeness；Or the relativeness is the relativeness between non-conterminous dividing candidate point.

11. track segmenting device according to claim 8, wherein the classifier is generated by following steps:

Obtaining step, for obtaining multiple trained track sampled points；

Segmentation step, for dividing the trained track and obtaining 1 training dividing candidate points；

Obtaining step, for obtaining the true value of at least three training dividing candidate point, wherein each table in the true value Show that corresponding training dividing candidate point is true cut-point or false alarm；

Extraction step, for extracting the one group of spy for describing the relativeness between at least three training dividing candidate point Sign, wherein one group of feature is the distinguishing characteristics that can distinguish true cut-point and false alarm；

Training step, it is special for described one group based on the relativeness described between at least three training dividing candidate point The true value of sign and at least three training dividing candidate point, Lai Xunlian classifier.

12. track segmenting device according to claim 11, wherein the relativeness is true value identical at least two Relativeness between training dividing candidate point；Or

The relativeness is the relativeness between 1 different training dividing candidate points of true value.

13. track segmenting device according to claim 11, wherein the relativeness is adjacent very trained cut-point Between relativeness；Or

The relativeness is the relativeness between adjacent vacation training cut-point.

14. a kind of track identification device comprising:

Cutting unit, rail will be inputted by being configured as the track segmenting device according to any one in claim 8 to 13 Mark is divided at least one orbit segment；

Recognition unit is configured as identifying at least one described orbit segment.

15. a kind of display equipment comprising:

Track generation unit is configurable to generate at least one track；

Track identification device according to claim 14；

Display unit is configured as display content corresponding with the result of the track identification device.

16. display equipment according to claim 15, wherein the display equipment is mobile phone, camera, game Any of box, television set or computer.