JP2013152595A - Information processing apparatus and method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To achieve fast and robust recognition of a recognition object.SOLUTION: A recognition unit recognizes a recognition object from an input image, and a learning unit learns a feature quantity of a recognition object for each frame using a feature quantity having a score changing in time series. The recognition unit recognizes, on the basis of the feature quantity of a previous frame obtained as the learning results of the learning unit, the recognition object of a current frame. This technique can be applied to, for instance, an information processing apparatus such as a personal computer.

Description

  The present technology relates to an information processing device and method, and a program, and more particularly, to an information processing device and method, and a program that enable a recognition object to be recognized at high speed and robustly.

  In recent years, various methods for tracking an object have been proposed in order to realize gesture control, an automatic monitoring system, and the like.

  For example, as in Massachusetts Institute of Technology (MIT) "SixthSense" and Microsoft Corporation "Kinect" (trademark), shapes and textures of hands and bodies are changed using features such as color and depth. There are practical examples of tracking objects.

  As a technique for recognizing a hand gesture, first, an image in which only the hand is captured is used, or an image including only the hand is extracted by specifying the position of the hand in the image. As for the extracted image, there has been proposed a technique for recognizing a hand gesture by a technique called skin color information, motion detection, and pattern matching (see Patent Document 1).

  In addition, the posture and gestures such as multiple hand shapes are defined by prior learning as multiple dictionaries, and when recognizing, the multiple dictionaries can be switched and used at the same time depending on the state of the recognized object and time-series changes. A technique to be used has also been proposed.

JP 2007-333690 A

  However, in the technique such as Patent Document 1, it is necessary to previously learn recognition objects in various states, and only a recognition object in the same state as the state when learning in advance can be recognized.

  In other words, if the shape or color of the hand as a recognized object changes greatly, the recognition performance will be affected, and even if a large number of dictionaries are defined so as to cope with changes in the recognized object, There was a limit to including the state space (feature space).

  In particular, when recognizing a recognized recognition object online, it is necessary to recognize the recognition object at high speed and robustly after learning in advance the various states and time-series changes of the recognition object.

  The present technology has been made in view of such a situation, and makes it possible to recognize a recognition object at high speed and robustly.

  An information processing apparatus according to an aspect of the present technology learns the feature amount of the recognition object for each frame using a recognition unit that recognizes a recognition object from an input image and a feature amount having a score that changes in time series. A learning unit, and the recognition unit recognizes the recognition object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit.

  The information processing apparatus includes: a pre-learning unit that learns in advance the feature amount of the recognized object for each frame from a learning image including the recognized object whose state changes in time series; and A calculation unit that pre-calculates the score of the feature amount for each frame obtained as a learning result is further provided, and the learning unit includes the feature amount corresponding to the score calculated in advance by the calculation unit. It is possible to learn the feature amount of the recognized object for each frame of the input image.

  The information processing apparatus further includes a selection unit that selects the feature amount having a score higher than a predetermined threshold from the feature amount corresponding to the current frame having the score calculated in advance by the calculation unit, The learning unit can learn the feature amount of the recognized object in the current frame using the feature amount selected by the selection unit.

  The information processing apparatus includes a storage unit that stores the feature amount obtained as a learning result of the learning unit, and a frame that stores the feature amount stored in the storage unit according to the learning result of the learning unit. An update unit that updates each time is further provided, and the recognition unit can recognize the recognition object of the current frame based on the feature amount of the previous frame updated by the update unit.

  The selection unit is configured to select the feature amount having a score lower than another threshold value from the feature amount of the previous frame stored in the storage unit, and the learning unit is selected by the selection unit The feature amount of the recognized object in the current frame can be learned using the feature amount excluding the feature amount.

  The update unit can update the feature amount obtained as a learning result of the pre-learning unit for each frame according to the learning result of the learning unit.

  An information processing method according to an aspect of the present technology learns the feature amount of the recognized object for each frame using a recognition unit that recognizes a recognized object from an input image and a feature amount having a score that changes in time series. An information processing method of an information processing device comprising a learning unit, wherein the information processing device learns the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series, Recognizing the recognized object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit.

A program according to one aspect of the present technology includes a recognition step of recognizing a recognition object from an input image;
A learning step of learning the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series, and the recognition step is a learning result of the learning step. The recognition object of the current frame is recognized based on the feature amount of the previous frame obtained as.

  In one aspect of the present technology, the feature amount of the recognition object for each frame is learned using the feature amount having a score that changes in time series, and the feature amount of the previous frame obtained as a learning result of the learning unit is used. Thus, the recognition object of the current frame is recognized.

  According to one aspect of the present technology, a recognition object can be recognized at high speed and robustly.

It is a block diagram which shows the structure of one Embodiment of the information processing apparatus to which this technique is applied. It is a block diagram which shows the structure of one Embodiment of the personal computer to which this technique is applied. It is a flowchart explaining an offline learning process. It is a figure which shows the time-sequential change of the score of a feature-value. It is a flowchart explaining an object tracking process. It is a flowchart explaining an object recognition process. It is a flowchart explaining a learning process. It is a figure explaining replacement of the feature-value according to a score.

Hereinafter, embodiments of the present technology will be described with reference to the drawings. The description will be given in the following order.
1. 1. Configuration of information processing apparatus 2. Configuration of personal computer Offline learning process Object tracking processing Object recognition processing Learning process Other

<1. Configuration of information processing apparatus>
It is a block diagram which shows the structure of one Embodiment of the information processing apparatus to which this technique is applied.

  The information processing apparatus 1 in FIG. 1 recognizes a recognition object that is a recognition target from an input image, and outputs the recognition result. The information processing apparatus 1 includes an offline processing unit 2 and an online processing unit 3. The offline processing unit 2 learns information on the recognized object in advance (offline learning) based on learning data prepared in advance, and supplies the learning result to the online processing unit 3. The online processing unit 3 recognizes the recognized object from the input image based on the learning result from the offline processing unit 2 and learns information on the recognition target (online learning).

  The offline processing unit 2 includes a learning data storage unit 11 and a learning processing unit 12, and the online processing unit 3 includes an offline dictionary storage unit 13, an input unit 14, a recognition unit 15, and a learning processing unit 16.

  The learning data storage unit 11 stores learning data collected in advance for learning the information of the recognized object. The learning data is a moving image including a recognition object (hereinafter also referred to as a learning image).

  The learning processing unit 12 learns the feature amount of the recognized object in advance (offline learning) from the learning image stored in the learning data storage unit 11 as the recognition object information for each frame, and the learning result is obtained. The data is supplied to the offline dictionary storage unit 13.

  The learning processing unit 12 includes a learning unit 21 and a calculation unit 22. The learning unit 21 learns the feature amount of the recognized object for each frame in advance, and the calculation unit 22 calculates the score of the feature amount of the recognized object for each frame. The score is a parameter that represents the degree to which the recognized object can be recognized by the feature quantity, and the feature quantity having a higher score is a (good) feature quantity suitable for accurately recognizing the recognized object.

  The offline dictionary storage unit 13 stores the learning result from the learning processing unit 12 as a dictionary (offline dictionary). The dictionary is a recognizer that is given for each pattern of changes in the shape and state of a recognized object, and has parameters including a feature amount as a learning result and its score.

  The input unit 14 inputs a moving image acquired by imaging a subject with a camera or the like, and outputs the moving image (input image) to the recognition unit 15. Note that the input unit 14 may itself be a camera that captures an image of a subject. The camera is a fixed camera whose shooting direction is fixed.

  The recognition unit 15 recognizes a recognition object for each frame from the input image, and supplies the recognition result to the learning processing unit 16.

  The learning processing unit 16 uses the dictionary stored in the offline dictionary storage unit 13 to learn (online learning) the feature quantity of the recognized object in the input image based on the recognition result from the recognition unit 15.

  The learning processing unit 16 includes a selection unit 23, a learning unit 24, a storage unit 25, and an update unit 26. The selection unit 23 selects a feature amount included in the dictionary of the offline dictionary storage unit 13 according to the score. Details of the score will be described later. The learning unit 24 learns the feature amount of the recognized object using the feature amount selected by the selection unit 23. The storage unit 25 stores the learning result of the learning unit 24 as a dictionary (online dictionary). The update unit 26 updates the online dictionary stored in the storage unit 25 according to the learning result of the learning unit 24.

  The information processing apparatus 1 can also be configured by a personal computer 31 that realizes a predetermined function by executing software as shown in FIG.

<2. Configuration of personal computer>
FIG. 2 is a block diagram illustrating a configuration of an embodiment of a personal computer to which the present technology is applied.

  A personal computer 31 as an information processing apparatus includes a bus 41, a CPU (Central Processing Unit) 42, a ROM (Read Only Memory) 43, a RAM (Random Access Memory) 44, an input unit 45, an output unit 46, a storage unit 47, and a communication. The unit 48, the drive 49, and the removable medium 50 are configured.

  The bus 41 connects the CPU 42, ROM 43, RAM 44, input unit 45, output unit 46, storage unit 47, communication unit 48, and drive 49 to each other.

  The CPU 42 realizes various functions of the information processing apparatus 1 in FIG. 1 by controlling various operations of the personal computer 31.

  The ROM 43 records various processing programs executed in the personal computer 31 and data necessary for the processing. The RAM 44 is used as a work area for various processes such as temporarily recording and holding data obtained in various processes.

  The input unit 45 includes a keyboard, a mouse, a microphone, and the like. The output unit 46 includes a display, a speaker, and the like. The storage unit 47 includes a hard disk, a nonvolatile memory, and the like.

  The communication unit 48 includes a network interface. The drive 49 drives a removable medium 50 such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory.

  In the personal computer 31 configured as described above, the CPU 42 loads various programs, for example, stored in the ROM 43 or the storage unit 47 to the RAM 44 via the bus 41 and executes them. Is done.

  The program executed by the CPU 42 is provided, for example, by being recorded on a removable medium 50 as a package medium or the like.

  As the package medium, a magnetic disk (including a flexible disk), an optical disk (CD-ROM (Compact Disc-Read Only Memory), DVD (Digital Versatile Disc), etc.), a magneto-optical disk, or a semiconductor memory is used. .

  The program can be provided via a wired or wireless transmission medium such as a local area network, the Internet, or digital satellite broadcasting.

  In the personal computer 31, the program can be installed in the storage unit 47 via the bus 41 by attaching the removable medium 50 to the drive 49.

  The program can be received by the communication unit 48 via a wired or wireless transmission medium and installed in the storage unit 47. In addition, the program can be installed in the ROM 43 or the storage unit 47 in advance.

  Note that the program executed by the personal computer 31 may be a process in which processing is performed in time series in the order described in this specification, or may be necessary in parallel or when a call is made. It may be a program that performs processing at timing.

<3. Offline learning process>
Next, offline learning processing by the offline processing unit 2 of the information processing apparatus 1 will be described with reference to the flowchart of FIG.

  In step S <b> 11, the learning processing unit 12 acquires a learning image stored in the learning data storage unit 11. The learning image is a moving image including a recognition object whose shape and state change in time series. Here, it is assumed that the recognition object is a hand, and the learning image is a moving image including a hand whose shape changes in time series. Specifically, for example, a moving image in which a hand making a gesture for paying an object from the shape of a par is captured a plurality of times is prepared as a learning image.

  In step S <b> 12, the learning unit 21 of the learning processing unit 12 learns the recognition target time-series state from the learning image acquired from the learning data storage unit 11. Specifically, the learning unit 21 learns a time-series change in the feature amount of the recognized object from the learning image. In this learning, clustering, interpolation, etc. may be performed based on label information including information according to the position and / or size of the recognized object, or modeling such as HMM (Hidden Markov Model) A technique may be used.

  In step S <b> 13, the learning processing unit 12 normalizes the learning image in time series using the learning result of the learning unit 21.

  In step S14, the calculation unit 22 of the learning processing unit 12 calculates a feature amount score for each frame using the normalized learning image.

  FIG. 4 shows an example of the time-series change of the feature amount score.

  In FIG. 4, among n feature quantities f1, f2,..., Fn, changes in the scores of the feature quantities f1, f2, fn with respect to time t, that is, changes for each frame are shown.

  As shown in FIG. 4, the score of the feature quantity f1 is always higher than the scores of the feature quantities f2 and fn. That is, it can be said that the feature amount f1 is more suitable for recognizing a hand shape change in the learning image than the feature amounts f2 and fn.

  In step S <b> 15, the learning processing unit 12 supplies the feature quantity and the score of the recognized object for each frame obtained as a learning result to the offline dictionary storage unit 13 and stores it as an offline dictionary (offline recognizer).

  In this way, offline learning is performed on the recognized object using the learning image.

  In the present embodiment, online learning is performed using the learning result of offline learning. In online learning, the same hand shape change as that learned by offline learning is learned.

  Here, in the conventional learning, the feature amount learning (parameter adjustment) and the score calculation (evaluation) are generally performed for the same frame, but in the offline learning of the present embodiment, Considering the feature of online learning described later, it is assumed that the feature amount is learned in the current frame of interest, and the feature amount score is calculated in the next frame temporally after the current frame.

  In the above description, offline learning is performed for each feature amount, but may be performed for each feature amount group that is a combination of a plurality of feature amounts.

<4. Object tracking processing>
Next, the object tracking process by the online processing unit 3 of the information processing apparatus 1 will be described with reference to the flowchart of FIG. In the object tracking process, a recognition object that changes in the same manner as the shape and state change pattern of the recognition object learned in the offline learning process is tracked. Specifically, for example, in the object tracking process, a hand that makes a gesture of paying an object from the shape of a par is tracked in the input image.

  In step S31, the input unit 14 sets a recognition object. Here, the hand in the shape of a par, which is the initial state of the shape change of the hand that is the recognition object, is set.

  In step S32, the input unit 14 acquires an image. That is, an input image obtained by imaging the subject is acquired.

  In step S33, the input unit 14 detects a recognition object. That is, the hand in the shape of the par set in the process of step S31 is detected from the input image acquired by the process of step S32. The detection of the recognized object here may be performed in accordance with a user operation, or may be performed by the input unit 14 using an object detection method.

  In step S <b> 34, the recognition unit 15 performs object recognition processing on the input image (frame) from the input unit 14 by calculating the hand shape probability and the hand color pattern. For this object recognition processing, for example, a technology such as Japanese Patent Application Laid-Open No. 2010-108475 that configures a recognizer using Boosting with a Steerable Filter response as a feature amount can be used. Further, an SSD (Sum of Squared Difference), a color histogram template matching method, or the like may be used.

<5. Details of object recognition processing>
Here, the details of the object recognition processing in step S34 of the flowchart of FIG. 5 will be described with reference to the flowchart of FIG.

  In step S <b> 51, the recognition unit 15 acquires an online dictionary that is a learning result of online learning for the previous frame from the storage unit 25 of the learning processing unit 16. When the frame of the input image is the first frame, the recognition unit 15 acquires a learning result (offline dictionary) of offline learning for the frame corresponding to the first frame from the offline dictionary storage unit 13.

  In step S52, the recognition unit 15 recognizes the recognized object by calculating a feature amount score based on the acquired dictionary.

  In step S53, the recognition unit 15 generates a recognition result based on the calculated feature amount score, and returns to the process of step S34 in FIG.

  Returning to the flowchart of FIG. 5, in step S <b> 35, the learning processing unit 16 executes a recognition object learning process using the recognition result generated by the recognition unit 15.

<6. Details of the learning process>
Here, the details of the learning process in step S35 of the flowchart of FIG. 5 will be described with reference to the flowchart of FIG.

  In step S <b> 71, the selection unit 23 of the learning processing unit 16 acquires the recognition result generated by the recognition unit 15.

  In step S72, the selection unit 23 has a score lower than the first threshold in the online dictionary of the previous frame temporally prior to the current frame (online learning result for the previous frame) stored in the storage unit 25. Select feature quantity. If the current frame is the first frame of the input image, the process of step S72 is skipped.

  In step S73, the selection unit 23 selects, in the offline dictionary (offline learning result) stored in the offline dictionary storage unit 13, a feature amount having a score higher than the second threshold value from among the feature amounts corresponding to the current frame. To do.

  Here, in the online learning result of the previous frame, the feature amount obtained by adding the feature amount selected in step S73 excluding the feature amount selected in step S72 is referred to as a learning feature amount.

  Note that the feature quantity for learning may be obtained by exchanging the feature quantity selected in step S72 with the feature quantity selected in step S73.

  Specifically, the feature quantities f1 and f2 having high scores at the time T corresponding to the current frame in the score of the feature quantity changing in time series as the offline learning result shown in the upper part of FIG. 8 are shown in the middle part of FIG. The feature values fy and fz having low scores in the online learning result of the previous frame shown may be replaced. As a result, the learning feature quantity including the feature quantities f1 and f2 and excluding the feature quantities fy and fz shown in the lower part of FIG. 8 is obtained.

In step S <b> 74, the learning unit 24 learns the feature amount of the recognized object using the learning feature amount obtained as a result of selection by the selection unit 23. Learning is performed using an online boosting technique or the like. The online boosting technique is disclosed in the following document, for example.
Helmut Grabner and Horst Bischof, "On-line Boosting and Vision", In Proceedings IEEE Conference on Computer Vision and Pattern Recognition (CVPR), volume 1, pages 260-267, 2006

  In step S75, the updating unit 26 updates the online dictionary stored in the storage unit 25 based on the learning result of the learning unit 24, and the process returns to step S35 in FIG.

  Returning to the flowchart of FIG. 5, in step S <b> 36, the update unit 26 updates the offline learning result stored in the offline dictionary storage unit 13 with the learning result of the learning unit 24. Thereby, the recognition performance with respect to a specific recognition object can be improved. Note that the process of step S36 may not be performed as necessary.

  In step S37, the recognition unit 15 outputs the recognition result to a display device (not shown), for example. That is, the recognition result generated by the process of step S34 is displayed on a display device (not shown). Specifically, a frame surrounding the recognized hand is displayed in the input image displayed by the display device.

  Thereby, the user can confirm tracking of the hand as a recognition object.

  Note that the processes of steps S35 and S36 in FIG. 5 and the process of step S37 can be executed in parallel. That is, the process of step S37 may be executed before the processes of steps S35 and S36 are completed.

  In step S38, the input unit 14 acquires an image. That is, a frame of a new input image is acquired.

  Note that the newly acquired image may be an image of a frame next to an image acquired in the past, or may be an image after a predetermined number of frames from an image acquired in the past.

  That is, a new image may be acquired for each frame or may be acquired for each predetermined number of frames.

  After the process of step S38, the process returns to step S34, and the subsequent processes are repeated.

  In the above, the recognition object is assumed to be a hand that is a part of the user's body, but the shape and state in time series such as a part of the body other than the hand, the entire body, an object other than the user's body, As long as it changes.

According to the above processing, the following effects can be realized.
(1) Although it is difficult to perform online learning for the entire feature space, it is fast and robust to recognize objects whose shape and appearance change greatly by performing learning using prior knowledge of the recognized objects (offline learning). Can be recognized.
(2) In off-line learning, it is possible to evaluate a score, which is a goodness for each feature amount of a recognized object, or a goodness of a feature amount group which is a combination of a plurality of feature amounts, using a temporal distribution.
(3) In online learning, feature quantities suitable for learning are selected and calculated based on the learning results of offline learning, so that learning can be performed at high speed, and there are feature quantities specialized for recognition objects. Since it is used, it is possible to learn robustly.
(4) Since the learning result of online learning is fed back to the offline learning result (offline dictionary), a dictionary for a specific recognition object, that is, a personalized dictionary can be generated.
(5) Since a recognition object whose shape and appearance greatly change can be recognized at high speed and robustly, it can be easily applied to an application in gesture recognition.

The present technology can be applied to a case where an information processing apparatus such as a television receiver or a personal computer is remotely operated with a gesture.
<7. Other>

  Embodiments of the present technology are not limited to the above-described embodiments, and various modifications can be made without departing from the gist of the present technology.

  For example, the present technology can take a configuration of cloud computing in which one function is shared by a plurality of devices via a network and is jointly processed.

  In addition, each step described in the above flowchart can be executed by being shared by a plurality of apparatuses in addition to being executed by one apparatus.

  Further, when a plurality of processes are included in one step, the plurality of processes included in the one step can be executed by being shared by a plurality of apparatuses in addition to being executed by one apparatus.

Moreover, this technique can take the following structures.
(1) a recognition unit that recognizes a recognition object from an input image;
A learning unit that learns the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series, and
The recognition unit recognizes the recognition object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit.
(2) a pre-learning unit that learns in advance the feature amount of the recognized object for each frame from a learning image including the recognized object whose state changes in time series;
A calculation unit that calculates in advance the score of the feature amount for each frame obtained as a learning result of the pre-learning unit;
The information according to (1), wherein the learning unit learns the feature amount of the recognition object for each frame of the input image using the feature amount corresponding to the score calculated in advance by the calculation unit. Processing equipment.
(3) a selection unit that selects the feature amount having a score higher than a predetermined threshold from the feature amount corresponding to the current frame having the score calculated in advance by the calculation unit;
The information processing apparatus according to (2), wherein the learning unit learns the feature amount of the recognized object in the current frame using the feature amount selected by the selection unit.
(4) a storage unit that stores the feature amount obtained as a learning result of the learning unit;
An update unit that updates the feature amount stored in the storage unit for each frame according to a learning result of the learning unit;
The information processing apparatus according to any one of (1) to (3), wherein the recognition unit recognizes the recognition object of the current frame based on the feature amount of the previous frame updated by the update unit.
(5) The selection unit selects the feature amount having a score lower than another threshold from the feature amount of the previous frame stored in the storage unit,
The learning unit learns the feature amount of the recognized object in the current frame using the feature amount excluding the feature amount selected by the selection unit. (1) to (4) Information processing device.
(6) The update unit updates the feature amount obtained as a learning result of the pre-learning unit for each frame according to the learning result of the learning unit. (4) or (5) Processing equipment.
(7) a recognition unit that recognizes a recognition object from the input image;
An information processing method of an information processing apparatus including a learning unit that learns the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series,
The information processing apparatus is
Using the feature amount having a score that changes in time series, learning the feature amount of the recognized object for each frame,
An information processing method including a step of recognizing the recognized object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit.
(8) a recognition step of recognizing a recognition object from the input image;
Using a feature amount having a score that changes in time series, causing a computer to execute a process including learning step for learning the feature amount of the recognized object for each frame,
The recognition step is a program for recognizing the recognition object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning step.

  DESCRIPTION OF SYMBOLS 1 Information processing apparatus, 2 Offline processing part, 3 Online processing part, 11 Learning data storage part, 12 Learning processing part, 13 Offline dictionary preservation | save part, 14 Input part, 15 Recognition part, 16 Learning processing part, 21 Learning part, 22 calculation units, 23 selection units, 24 learning units, 25 storage units, 26 update units

Claims (8)

A recognition unit for recognizing a recognition object from an input image;
A learning unit that learns the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series, and
The recognition unit recognizes the recognition object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit. A pre-learning unit that learns in advance the feature amount of the recognized object for each frame from a learning image including the recognized object whose state changes in time series;
A calculation unit that calculates in advance the score of the feature amount for each frame obtained as a learning result of the pre-learning unit;
The information according to claim 1, wherein the learning unit learns the feature amount of the recognized object for each frame of the input image using the feature amount corresponding to the score calculated in advance by the calculation unit. Processing equipment. A selection unit that selects the feature amount having a score higher than a predetermined threshold from the feature amount corresponding to the current frame having the score calculated in advance by the calculation unit;
The information processing apparatus according to claim 2, wherein the learning unit learns the feature amount of the recognized object in the current frame using the feature amount selected by the selection unit. A storage unit for storing the feature amount obtained as a learning result of the learning unit;
An update unit that updates the feature amount stored in the storage unit for each frame according to a learning result of the learning unit;
The information processing apparatus according to claim 3, wherein the recognition unit recognizes the recognition object of the current frame based on the feature amount of the previous frame updated by the update unit. The selection unit selects the feature amount having a score lower than another threshold from the feature amount of the previous frame stored in the storage unit,
The information processing apparatus according to claim 4, wherein the learning unit learns the feature amount of the recognized object in the current frame using the feature amount excluding the feature amount selected by the selection unit. The information processing apparatus according to claim 4, wherein the update unit updates the feature amount obtained as a learning result of the pre-learning unit for each frame according to a learning result of the learning unit. A recognition unit for recognizing a recognition object from an input image;
An information processing method of an information processing apparatus including a learning unit that learns the feature amount of the recognized object for each frame using a feature amount having a score that changes in time series,
The information processing apparatus is
Using the feature amount having a score that changes in time series, learning the feature amount of the recognized object for each frame,
An information processing method including a step of recognizing the recognized object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning unit. A recognition step for recognizing a recognition object from an input image;
Using a feature amount having a score that changes in time series, causing a computer to execute a process including learning step for learning the feature amount of the recognized object for each frame,
The recognition step is a program for recognizing the recognition object of the current frame based on the feature amount of the previous frame obtained as a learning result of the learning step.

Images