KR20160087738A - Method and apparatus of detecting object using event-based sensor - Google Patents

Abstract

Disclosed are a method and an apparatus for detecting an object using an event-based sensor. According to an embodiment of the present invention, the method for detecting an object includes the steps of: determining a feature vector based on target pixels and neighboring pixels included in an event image; and determining a target object corresponding to the target pixels based on the feature vector.

Description

TECHNICAL FIELD [0001] The present invention relates to a method and apparatus for detecting an object using an event-based sensor,

The following embodiments relate to a method and apparatus for detecting an object using an event-based sensor.

The detection of motion objects is a research issue in computer image field and can be widely applied to military reconnaissance, surveillance system, HCI (Human Computer Interaction) and so on. For example, capture, tracking and analysis can be performed on the moving object, and the HCI experience can be improved by switching the operation pattern of the terminal equipment based on the motion pattern of the analyzed moving object. The terminal equipment may be a mobile terminal, a camcorder, smart glasses, and a smart TV.

Detection of a moving object is generally performed by using a conventional imaging device based on a CCD (Charge-coupled Device) or a CMOS (Complementary Metal Oxide Semiconductor) to perform image collection on a scene to which the motion object belongs, By separating the motion zone and the background zone, and then identifying the motion object based on the divided motion zone. However, it is difficult to divide a moving object in a scene because it takes a long time and requires rapid tracking.

In addition, the conventional motion object tracking method can quickly acquire a motion object in a state where the imaging equipment is open. Therefore, the conventional motion object tracking method has a relatively high energy consumption and is not suitable for application to portable equipment. Accordingly, there is a need for a method of detecting a motion object that has low energy consumption and quickly captures a motion object.

A method of detecting an object according to one side includes: generating an event image based on an event signal by an event-based sensor; Determining a feature vector based on the target pixels and neighboring pixels included in the event image; And determining a target object corresponding to the target pixels based on the feature vector.

Wherein the step of determining the target object based on the feature vector comprises the step of determining the target object based on the result obtained by inputting the feature vector to a classifier learned by the learning sample related to the target area and the neighboring area neighboring the target area, Based on the result of the determination. The step of determining the target object based on the feature vector may include determining a type of the target object and a position of the target object. The step of determining the target object based on the feature vector may include determining a position of the target object based on the position of the pixels corresponding to the target object.

Wherein the step of determining the feature vector comprises: dividing the event image into a plurality of regions; And sampling the neighboring pixels in a neighboring region neighboring the target region to which the target pixels belong, among the regions. The step of sampling the neighboring pixels may comprise randomly sampling a predetermined number of pixels in the neighboring region.

The object detection method may further include a step of verifying the type of the target object. The verifying the type of the target object may include verifying the type of the target object based on the relationship between the target object and a neighboring object corresponding to the neighboring pixels. Wherein the verifying the type of the target object may include verifying the type of the target object based on the valid range related to the position of the target object, Can be determined based on the expected range of operation of the type.

The object detection method includes: determining a motion trajectory of the target object based on a position of the target object; And generating an action command corresponding to the motion locus. Wherein the generating the motion command comprises: segmenting the motion trajectory by motion segment; Extracting features relating to the order of the action segments; And generating the motion command based on the feature, the feature including at least one of a position feature, a path feature, a movement direction feature, a velocity feature, and an acceleration feature. The step of generating the operation command may include combining objects of different types into one object and then determining a motion trajectory of the one object based on a motion trajectory of each type of objects; Extracting a feature relating to a motion locus of the one object; And generating the motion command based on the feature, the feature including at least one of a position feature, a path feature, a movement direction feature, a velocity feature, and an acceleration feature.

A learning method according to one side includes the steps of: generating a learning sample about a target area and a neighboring area neighboring the target area; And learning the classifier on the type of object object corresponding to the object area based on the learning sample.

Wherein the generating the learning sample comprises: generating a sample image based on an event signal by an event-based sensor; Dividing the sample image into a plurality of regions; And constructing one learning sample among the target pixels included in the target area and neighboring pixels included in the neighboring area among the areas.

The step of learning the classifier may include learning the classifier based on a Deep Belief Network (DBN). The step of learning the classifier may include adjusting parameters of the classifier based on the classification of the classifier for the learning object type of the learning sample and the learning sample. The learning object type may include a type of the target object and a type of a neighbor object corresponding to the neighboring area.

An apparatus for detecting an object according to one side includes a processing unit for generating an event image generated based on an event signal of an event-based sensor; And a classifier included in the event image to determine a feature vector based on target pixels and neighboring pixels, and to detect a target object corresponding to the target pixels based on the feature vector.

The classifier can detect the target object based on a result value obtained by inputting the feature vector to a classifier learned by a learning sample related to a target area and a neighboring area neighboring the target area.

The object detecting apparatus may further include a verifying unit that verifies the type of the target object based on a relationship between the target object and a neighboring object corresponding to the neighboring pixels.

1 is a flow chart illustrating an object detection method according to an embodiment;
2 is a diagram illustrating an event image according to an embodiment;
3 is a diagram illustrating a process of filtering an event signal according to an exemplary embodiment;
4 is a flowchart illustrating an object verification process according to an exemplary embodiment;
5 is a flowchart illustrating a process of generating an operation command according to an embodiment;
6 is a flowchart illustrating a learning process according to an embodiment;
7 is a flowchart showing a process of generating a learning sample according to an embodiment;
8 illustrates a structure of a classifier according to an embodiment;
9 is a view showing the structure of a classifier according to another embodiment;
10 is a block diagram illustrating a learner and classifier in accordance with one embodiment.
11 is a block diagram illustrating an object detection apparatus according to an embodiment;

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Like reference symbols in the drawings denote like elements. Embodiments described below may be applied to a user interface. For example, it can be used to recognize a swipe motion in a non-contact motion recognition interface. According to the embodiments to be described below, the repetitive turnover operations can be quickly recognized with less power.

1 is a flow chart illustrating an object detection method according to an embodiment.

Referring to FIG. 1, in step 100, an object detection apparatus generates an event image based on an event signal by an event-based sensor. The event-based sensor includes a plurality of sensing pixels. The event-based sensor generates an event signal including identification information of an active pixel that senses an event among a plurality of sensing pixels. Event-based sensors can be based on neuromorphic sensing techniques.

An event includes an event associated with a change in input. For example, an event may be an event in which the intensity of incident light changes, an event in which the color of the incident light changes, an event in which the magnitude of the input sound changes, an event in which the frequency of the input sound changes, And the like may be included. The plurality of sensing pixels included in the event-based sensor may sense an event (e.g., an event in which the intensity of the incident light changes). Of the plurality of sensing pixels, a sensing pixel sensing an event may be referred to as an active pixel. The active pixel may generate an activation signal in response to detecting the event.

The event-based sensor may generate an event signal that includes identification information of the active pixel. For example, the event-based sensor may generate an event signal that includes an address that identifies the active pixel based on an activation signal generated by the active pixel. Event-based sensors generate and output event signals asynchronously in time, so they can operate at lower power and higher speed than frame-based vision sensors that scan all pixels every frame.

Furthermore, through the combination of the position of the event point accumulated within a certain time interval (for example, 20 ms), the event flow signal collected by the dynamic vision sensor can be converted into a video signal, And the contour of the motion object and the segment pattern information, and directly ignored the object which can not move in the background.

The object detecting apparatus generates an event image based on the event signal. The event image will be described with reference to FIG.

2 is a view for explaining an event image according to an embodiment. Referring to FIG. 2, an event image divided into a plurality of regions is shown. The plurality of regions may include a target region 10 and a neighboring region 20. The object area 10 means an area to be subjected to object detection. The target area 10 may be at least one of a plurality of areas. The neighboring region 20 means a region adjacent to the target region 10. [ The pixels included in the target area 10 may be referred to as target pixels. In addition, the pixels included in the neighboring region 20 may be referred to as neighboring pixels. The accuracy of object detection can be improved by dividing the event image into a plurality of regions and using the neighboring region 20 to detect an object in the object region 10. [

The object detecting apparatus can generate an event image based on the event stream accumulated in a predetermined time period. The event stream may include a plurality of event signals. The object detection device may filter the event signal prior to generating the event image. The process of filtering the event signal will be described with reference to FIG.

3 is a diagram for explaining a process of filtering an event signal according to an embodiment. Referring to FIG. 3, a time stamp map in which an event signal is recorded is shown. The event image may be generated based on a valid time stamp in the time stamp recorded in the time stamp map. A valid timestamp means an unfiltered timestamp. The particular timestamp may be filtered based on the value of the neighboring timestamp. For example, whether to filter the first timestamp may be determined based on the difference between the first timestamp and the neighbor timestamps neighboring the first timestamp. A neighboring stamp having a first timestamp and a value less than the threshold value may be referred to as a supporter. A timestamp whose number of supporters exceeds a predetermined number can be determined as a valid timestamp.

For example, a timestamp having three or more supporters may be determined as a valid timestamp. In FIG. 3, when the time stamp 30 is updated with the time stamp 50 according to the occurrence of the event, the time stamp 50 has three supporters 40. Thus, the timestamp 50 can be determined as a valid event signal.

Referring again to FIG. 1, in step 200, the object detection apparatus determines a feature vector based on target pixels and neighboring pixels included in the event image. The target pixels and neighboring pixels may be used together to detect the target object. Neighboring pixels may be sampled in the neighboring region 20. For example, the neighboring pixels may be arbitrarily sampled by a predetermined number in the neighboring region 20. Also, various schemes can be used for sampling neighboring pixels.

In step 300, the object detection apparatus determines a target object corresponding to the target pixels based on the feature vector. The object detecting apparatus can determine the target object based on the result of inputting the feature vector to the classifier learned by the learning samples related to the target area 10 and the neighboring area 20. [ The target object means an object included in the target area 10. The target object may be a moving object.

The object detection apparatus can determine the type of the target object and the position of the target object. For example, when an object detection device tracks a user's gesture, the type of object may include a head, a hand, and a body. The type of the target object can be learned in advance. For example, the object detecting apparatus may determine the type of the first object corresponding to the first object area based on the first object area and the first neighboring area neighboring the first object area, with the user's hand. Further, the object detecting apparatus can determine the type of the second object corresponding to the second object area as the head of the user based on the second object area and the second neighboring area neighboring the second object area. The object detection apparatus can track the movement of the detected user's hand and head. As will be described in greater detail below, a first neighbor object corresponding to the first neighboring region may be used for the determination of the first target object. Also, the first neighbor object may be used for verification of the first object.

The object detection apparatus can determine the position of the object of the determined type when the type of the object is determined. For example, the object detection apparatus can determine the position of the target object based on the position of the pixels belonging to the target object. Also, the object detection apparatus can determine the center of the pixels belonging to the target object as the position of the target object. In addition, the location of the target object can be determined using various cluster analysis techniques. For example, the center of a target object can be determined based on a K-means cluster analysis technique.

The object detection apparatus may repeat step 200 and step 300 for a plurality of object areas included in the event image. Accordingly, the object detection apparatus can determine object objects corresponding to the respective regions included in the event image. In addition, the object detection apparatus can track the movement of the object objects.

4 is a flowchart illustrating an object verification process according to an embodiment.

Referring to FIG. 4, in step 400, the object detection apparatus verifies the type of the target object. The object detection apparatus can verify the type of the object after determining the type of the object. The object detection apparatus can improve the speed and efficiency of object detection by verifying the type of the object.

According to one aspect, the object detection apparatus can verify the type of the target object based on the valid form. Valid forms are the types that a particular type of object can have. The object detection apparatus can determine an effective form in advance for each object type. For example, the effective form may include a basic form of the head or a basic form of the hand. The object detection apparatus can determine that the type of the target object is valid when the type of the target object belongs to the valid type. The effective form can be determined based on the type of the target object for each type accumulated in advance.

According to another aspect, the object detection apparatus can verify the type of the target object based on the validity range. An effective range is a range of positions where an object of a certain type can exist. The object detection apparatus can determine a valid range in advance for each type of object. For example, the effective range may include a position where the head may exist or a position where the arm may move. The validity range can be defined by the relationship between a plurality of objects. For example, the validity range can be determined based on the fact that the head and body are directly connected and that the head and body can not be directly connected. The validity range may also be determined based on an expected range of operation of a particular type of object. For example, the effective range can be determined based on a range in which the head can move about the body or a range in which the arm can move about the body. The expected range of motion can be considered with the previous location of the target object. For example, when t = 2 based on the position of the arm when t = 1, the expected range of motion of the arm can be determined. The object detection apparatus can determine that the type of the target object is valid when the position of the target object belongs to the valid range. The validity range can be determined based on the position and the positional change of the target object for each type accumulated in advance. The object detection apparatus can trace a target object of the corresponding type after verifying the type of the target object.

5 is a flowchart showing a process of generating an operation command according to an embodiment.

Referring to FIG. 5, in step 500, the object detection apparatus determines a motion trajectory of a target object based on the position of the target object. The object detection apparatus can identify the type and position of the target object and then determine the motion locus of the target object based on the position of the target object sequentially identified. Various techniques such as a smoothing filter or a time sequence tracking algorithm can be used to determine the motion trajectory. Due to the specificity of the event-based sensor, the motion trajectory may temporarily disappear when the target object is temporarily stopped. The object detecting apparatus can perform a smoothing process on the stopped object at the stop position. Various techniques such as a Kalman filter method can be used for smoothing processing.

In step 600, the object detection apparatus generates an action command corresponding to the motion trajectory. According to one aspect, the object detection apparatus may divide the motion locus by motion segment, extract a feature related to the order of motion segments, and generate an motion command based on the feature. According to another aspect of the present invention, the object detecting apparatus may combine objects of different types into one object, determine a motion trajectory of one object based on the motion trajectory of each type of object, Extract features relating to the locus, and generate an operation command based on the feature. The feature may include at least one of a position feature, a path feature, a movement direction feature, a velocity feature, and an acceleration feature.

The motion command may be generated based on motion of one target object, such as a hand, or may be generated based on motion of a plurality of target objects, such as head and hands. Motion of a plurality of target objects can be treated as one motion locus. For example, the motion locus of the head and the motion locus of the hand can be combined into one motion locus, and an operation command can be generated according to the combined motion locus.

6 is a flowchart showing a learning process according to an embodiment.

Referring to FIG. 6, in step 1100, the learning device generates a learning sample about a subject area and a neighboring area neighboring the subject area. The training samples may be generated based on event signals collected by the event-based sensor. For example, an event-based sensor can capture various motion processes of various types of objects and collect sample signals. The event-based sensor responds only to a pixel where an event such as a brightness change of a pixel occurs, and transmits and stores an event signal of the pixel. The event-based sensor may generate an event signal for the sample signal. The event signal generated by the event-based sensor from the sample signal is referred to as a sample event signal.

The event signal collected by the event-based sensor indicates the motion contour of the object to a certain level, and the shape of the object can be grasped based on the motion contour. Accordingly, neighboring pixels are suitable for describing the structure of the object corresponding to the target pixels, and can help determine which part of the object the target pixels correspond to, i.e., the type of the target object. For example, when a human body moves, a plurality of event signals are generated, and it is possible to determine whether the target pixels belong to the head, the hand, or the body or the like through the neighboring pixels within a certain range from the target pixels and the target pixels.

Therefore, the event-based sensor can determine the target pixels and neighboring pixels based on the sample event signal output from the sample signal, and determine the type of target object corresponding to the target pixels based on the target pixels and neighboring pixels . For example, the type of object corresponding to the sample event signal may be determined based on the location of the target pixels and the location of neighboring pixels. The type of object can be specifically the head, the hand, and the body.

In this way, the target pixels and neighboring pixels can be constituted by one learning sample. In addition, learning on the type of the moving object corresponding to the sample event signal can be performed through the learning sample. It is possible to sample neighboring pixels around the target pixels according to the set sampling range and select a predetermined number of neighboring pixels. The type of the specific object can be learned by using the target pixels and the selected neighboring pixels as one learning sample. According to one aspect, the learning samples may be generated in the order shown in FIG.

7 is a flowchart showing a process of generating a learning sample according to an embodiment. Referring to Fig. 7, in step 1110, the learner generates a sample image based on a sample event signal by an event-based sensor. In step 1120, the sample image is divided into a plurality of regions. In step 1130, the target pixels included in the target area and neighbor pixels included in the neighboring area among the areas are configured as one learning sample. The learner can learn the classifier based on the learning sample generated through this process.

Referring again to Fig. 6, at step 1200, the learner learns the classifier on the type of object corresponding to the object area based on the training samples. The learner can adjust the parameters of the learner according to whether the classification result of the classifier is appropriate or not. In other words, the learner can adjust the parameters of the classifier based on the classification of the classifier for the learning object type of the learning sample and the learning sample. At this time, the learning target type may include the type of the target object and the type of the neighboring object corresponding to the neighboring area. That is, the classifier can learn the target object and the neighboring object together with one learning sample.

The classifier can be learned based on Deep Belief Network (DBN). In learning based on DBN, a plurality of learning samples can be used as a set of learning samples. A classification model of the DBN can be obtained based on the learning sample set. Various methods can be used for concrete learning process using DBN. For example, a plurality of iterative learning for the DBN can be performed using learning samples for each type. One learning process during the iterative learning process includes inputting a set of learning samples composed of a plurality of learning samples to the DBN, comparing the learning target type of the learning sample with the output of the DBN, i.e., the classification result of the classifier, Adjusting the parameters of the DBN based on the received data, and continuing the iteration of the next order or stopping the iteration and obtaining the classifier.

The output of the DBN is an estimate of the type of motion object. The performance of the final classifier can be improved by comparing the output of the DBN with the relatively accurate measurement result, which is the learning object type, and by adjusting the parameters of the DBN with the backward propagation learning technique according to the difference between the two. The structure of a specific classifier will be described with reference to Figs. 8 and 9. Fig.

8 is a view showing a structure of a classifier according to an embodiment. Referring to FIG. 8, the classifier includes limited Boltzmann machines (RBMs). The classifier generates output values from input values using RBMs. The parameters of the classifier can mean the weights of the RBMs. The parameters of the classifier can be adjusted through the weights of the RBMs. For example, the weights of the RBMs may be adjusted such that the output value of the classifier is close to a predetermined output value, that is, the object type. The input value may be a feature vector. The feature vector may be generated based on the target pixels and neighboring pixels. The output value may be the type of the target object. For example, the output value may include a head, a body, and a hand. Thus, the classifier according to one embodiment generates an output value corresponding to the type of the target object from the feature vector. That is, a classifier according to an exemplary embodiment can detect a target object based on target pixels and neighboring pixels.

9 is a diagram showing the structure of a classifier according to another embodiment. Referring to FIG. 8, the classifier generates an output value from a feature vector. The feature vector may be generated based on the target pixels and neighboring pixels. The output value may be the type of the target object and the type of the neighbor object. That is, the classifier according to another embodiment may detect the type of the target object corresponding to the target pixels and the type of the neighbor object corresponding to the neighboring pixels. For example, a classifier according to another embodiment may simultaneously detect a type of a target object and a type of a neighbor object based on a feature vector. The type of the detected neighboring object may be used to detect and verify the type of the target object. Thus, the accuracy of detection for the type of target object can be improved.

10 is a block diagram illustrating a learning device and a classifier according to an embodiment.

Referring to FIG. 10, the learning device 51 learns the classifier 52 based on the learning sample about the object area and the neighboring area neighboring the object area. A learning sample may be generated based on the target pixels and neighboring pixels. The learning device 51 can adjust the parameters of the classifier 52 based on the output value of the classifier 52 for the learning sample. The classifier 52 may include a DBN. Besides, the learning method described above can be applied to the learning device 51 and the classifier.

11 is a block diagram illustrating an object detection apparatus according to an embodiment.

Referring to FIG. 11, the object detection apparatus 60 includes an event-based sensor 61 and a classifier 62. The event-based sensor 61 includes a plurality of sensing pixels. The event-based sensor generates an event signal including identification information of an active pixel that senses an event among a plurality of sensing pixels. The classifier 62 detects the target object based on the event signal. According to one aspect, the classifier 62 can be learned in advance by the learning apparatus described above. According to another aspect, the object detecting apparatus 60 may include a learning apparatus, and the classifier 62 may be learned through a learning apparatus included in the object detecting apparatus 60. [

The object detecting apparatus 60 may include a processing unit. The processing unit may generate an event image based on the event signal, and may divide the event image into a plurality of regions. The processing unit may generate the feature vector based on the target area and the neighboring area. In addition, the feature vector may be input to the classifier 62 and the type and position of the target object may be obtained from the output of the classifier 62. The processing unit can track the target object. The processing unit may generate an operation command based on the motion of the target object.

The embodiments described above may be implemented in hardware components, software components, and / or a combination of hardware components and software components. For example, the devices, methods, and components described in the embodiments may be implemented within a computer system, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, such as an array, a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing device may execute an operating system (OS) and one or more software applications running on the operating system. The processing device may also access, store, manipulate, process, and generate data in response to execution of the software. For ease of understanding, the processing apparatus may be described as being used singly, but those skilled in the art will recognize that the processing apparatus may have a plurality of processing elements and / As shown in FIG. For example, the processing unit may comprise a plurality of processors or one processor and one controller. Other processing configurations are also possible, such as a parallel processor.

The software may include a computer program, code, instructions, or a combination of one or more of the foregoing, and may be configured to configure the processing device to operate as desired or to process it collectively or collectively Device can be commanded. The software and / or data may be in the form of any type of machine, component, physical device, virtual equipment, computer storage media, or device , Or may be permanently or temporarily embodied in a transmitted signal wave. The software may be distributed over a networked computer system and stored or executed in a distributed manner. The software and data may be stored on one or more computer readable recording media.

The method according to an embodiment may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions to be recorded on the medium may be those specially designed and configured for the embodiments or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

Although the embodiments have been described with reference to the drawings, various technical modifications and variations may be applied to those skilled in the art. For example, it is to be understood that the techniques described may be performed in a different order than the described methods, and / or that components of the described systems, structures, devices, circuits, Lt; / RTI > or equivalents, even if it is replaced or replaced.

Claims (20)

Generating an event image based on an event signal by an event-based sensor;
Determining a feature vector based on the target pixels and neighboring pixels included in the event image; And
Determining a target object corresponding to the target pixels based on the feature vector
The object detection method comprising: The method according to claim 1,
Wherein the step of determining the target object based on the feature vector comprises:
Determining a target object based on a result value obtained by inputting the feature vector to a classifier learned by a learning sample related to a target area and a neighboring area neighboring the target area,
Object detection method. The method according to claim 1,
Wherein the step of determining the target object based on the feature vector comprises:
Determining a type of the target object and a position of the target object;
Object detection method. The method according to claim 1,
Wherein the step of determining the target object based on the feature vector comprises:
Determining a position of the target object based on a position of pixels corresponding to the target object;
Object detection method. The method according to claim 1,
Wherein the step of determining the feature vector comprises:
Dividing the event image into a plurality of regions; And
Sampling the neighboring pixels in neighboring regions neighboring the object region to which the target pixels belong in the regions
The object detection method comprising: 6. The method of claim 5,
Wherein sampling the neighboring pixels comprises:
And randomly sampling a predetermined number of pixels in the neighboring region.
Object detection method. The method according to claim 1,
Verifying the type of the target object
The method further comprising: 8. The method of claim 7,
The step of verifying the type of the target object
And verifying the type of the target object based on a relationship between the target object and a neighboring object corresponding to the neighboring pixels.
Object detection method. 8. The method of claim 7,
Wherein the verifying the type of the target object comprises:
And verifying the type of the target object based on an effective range concerning the position of the target object,
Wherein the validity range is determined based on a previous location of the target object and an expected range of operation of the type,
Object detection method. The method according to claim 1,
Determining a motion trajectory of the target object based on the position of the target object; And
Generating an action command corresponding to the motion locus
The method further comprising: 11. The method of claim 10,
Wherein generating the action command comprises:
Dividing the motion locus by motion segment;
Extracting features relating to the order of the action segments; And
Generating the motion command based on the feature
Lt; / RTI >
The feature may include at least one of a position feature, a path feature, a movement direction feature, a velocity feature, and an acceleration feature.
Object detection method. 11. The method of claim 10,
Wherein generating the action command comprises:
Determining a motion trajectory of the one object based on a motion trajectory of each type of object after combining different types of objects into one object;
Extracting a feature relating to a motion locus of the one object; And
Generating the motion command based on the feature
Lt; / RTI >
The feature may include at least one of a position feature, a path feature, a movement direction feature, a velocity feature, and an acceleration feature.
Object detection method. Generating a learning sample about a subject area and a neighboring area neighboring the subject area; And
Learning the classifier on the type of the target object corresponding to the target area based on the learning sample
≪ / RTI > 14. The method of claim 13,
Wherein the step of generating the learning sample comprises:
Generating a sample image based on an event signal by an event based sensor;
Dividing the sample image into a plurality of regions; And
Constructing one learning sample among target pixels included in the target area and neighboring pixels included in the neighboring area among the areas;
≪ / RTI > 14. The method of claim 13,
The step of learning the classifier comprises:
And learning the classifier based on a Deep Belief Network (DBN).
Learning method. 14. The method of claim 13,
The step of learning the classifier comprises:
Adjusting a parameter of the classifier based on a classification result of the classifier for the learning object type of the learning sample and the learning sample,
Learning method. 17. The method of claim 16,
The learning object type may include:
The type of the target object and the type of the neighbor object corresponding to the neighboring area,
Learning method. A processor for generating an event image based on an event signal of the event-based sensor; And
A classifier for determining a feature vector based on the target pixels and neighboring pixels included in the event image and detecting a target object corresponding to the target pixels based on the feature vector,
And an object detection device. 19. The method of claim 18,
Wherein the classifier comprises:
Detecting a target object based on a result value obtained by inputting the feature vector to a classifier learned by a learning sample related to a target area and a neighboring area neighboring the target area,
Object detection device 19. The method of claim 18,
A verification unit for verifying the type of the target object based on a relationship between the target object and a neighboring object corresponding to the neighboring pixels;
The object detection apparatus comprising:

Images