KR20220054133A - Pedestrian trajectory prediction apparatus - Google Patents

Abstract

Presented is an artificial intelligence device that updates a feature used for predicting a pedestrian path using a characteristic of the pedestrian path. According to the present invention, a pedestrian path prediction device comprises a first feature updating unit that updates a first feature of a pedestrian path. The first feature update unit includes a multiscale setting unit that provides a graph set using a distance scale between nodes as an edge weight.

Description

Pedestrian trajectory prediction apparatus

The present invention relates to a walking path prediction device for predicting a walking path of a pedestrian.

Artificial intelligence (artificial intelligence) is a field of computer engineering and information technology that studies how computers can do the thinking, learning, and self-development that can be done by human intelligence. This means that it can be imitated.

In addition, artificial intelligence does not exist by itself, but is directly or indirectly related to other fields of computer science. In particular, in modern times, attempts are being made to introduce artificial intelligence elements in various fields of information technology and use them to solve problems in that field.

On the other hand, technologies for recognizing and learning the surrounding situation using artificial intelligence, providing information desired by the user in a desired form, or performing an operation or function desired by the user are being actively researched. And, an electronic device that provides such various operations and functions may be called an artificial intelligence device.

AI devices can predict the path of pedestrians. The artificial intelligence device may be a walking path prediction device. In this case, the artificial intelligence device may be mounted on, for example, a walking robot to guide the walking of the walking robot.

As a prior art for predicting a walking path, Citation 1 has been proposed. Citation 1 presents a technique using a graph convolutional network (GCN). The cited document 1 predicts the path of a pedestrian by using each spatiotemporal graph information. However, Citation 1 does not reflect the movement of pedestrians in response to social norms, so there is a problem in that the prediction of the walking path is inaccurate.

Mohamed, A.; Qian, K.; Elhoseiny, M.; and Claudel, C. 2020. Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction. In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

The present invention proposes an artificial intelligence apparatus for updating a feature used for predicting a walking path by using the characteristics of the walking path as proposed in the background described above. The artificial intelligence device may predict a walking path.

The apparatus for predicting a walking path according to the present invention includes a first feature update unit for updating a first feature of a walking path, and the first feature update unit provides a graph set by using a distance scale between nodes as an edge weight. It includes a scale setting unit.

According to the present invention, it is possible to predict an accurate walking path.

1 is a view showing an artificial intelligence device according to an embodiment of the present invention.
2 is a diagram illustrating an artificial intelligence server according to an embodiment of the present invention.
3 is a view showing an artificial intelligence system according to an embodiment of the present invention.
4 is a view for explaining an artificial intelligence device according to an embodiment of the present invention.
5 is a view for explaining the operation of the multi-scale setting unit.
Fig. 6 is a table for explaining the effect of the present invention;

Hereinafter, the embodiments disclosed in the present specification will be described in detail with reference to the accompanying drawings, but the same or similar components are assigned the same reference numbers regardless of reference numerals, and redundant description thereof will be omitted. The suffixes "module" and "part" for components used in the following description are given or mixed in consideration of only the ease of writing the specification, and do not have distinct meanings or roles by themselves. In addition, in describing the embodiments disclosed in the present specification, if it is determined that detailed descriptions of related known technologies may obscure the gist of the embodiments disclosed in this specification, the detailed description thereof will be omitted. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in the present specification, and the technical idea disclosed herein is not limited by the accompanying drawings, and all changes included in the spirit and scope of the present invention , should be understood to include equivalents or substitutes.

Terms including an ordinal number such as 1st, 2nd, etc. may be used to describe various elements, but the elements are not limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

When an element is referred to as being “connected” or “connected” to another element, it is understood that it may be directly connected or connected to the other element, but other elements may exist in between. it should be On the other hand, when it is said that a certain element is "directly connected" or "directly connected" to another element, it should be understood that the other element does not exist in the middle.

<Artificial Intelligence (AI)>

Artificial intelligence refers to a field that studies artificial intelligence or methodologies that can make it, and machine learning refers to a field that defines various problems dealt with in the field of artificial intelligence and studies methodologies to solve them. do. Machine learning is also defined as an algorithm that improves the performance of a certain task through constant experience.

An artificial neural network (ANN) is a model used in machine learning, and may refer to an overall model having problem-solving ability, which is composed of artificial neurons (nodes) that form a network by combining synapses. An artificial neural network may be defined by a connection pattern between neurons of different layers, a learning process that updates model parameters, and an activation function that generates an output value.

The artificial neural network may include an input layer, an output layer, and optionally one or more hidden layers. Each layer includes one or more neurons, and the artificial neural network may include neurons and synapses connecting neurons. In the artificial neural network, each neuron may output a function value of an activation function for input signals, weights, and biases input through synapses.

A model parameter means a parameter determined through learning, and includes the weight of the synaptic connection and the bias of the neuron. In addition, the hyperparameter refers to a parameter that must be set before learning in a machine learning algorithm, and includes a learning rate, the number of iterations, a mini-batch size, an initialization function, and the like.

The purpose of learning the artificial neural network can be seen as determining the model parameters that minimize the loss function. The loss function may be used as an index for determining optimal model parameters in the learning process of the artificial neural network.

Machine learning can be classified into supervised learning, unsupervised learning, and reinforcement learning according to a learning method.

Supervised learning refers to a method of training an artificial neural network in a state where a label for the training data is given, and the label is the correct answer (or result value) that the artificial neural network should infer when the training data is input to the artificial neural network. can mean Unsupervised learning may refer to a method of training an artificial neural network in a state where no labels are given for training data. Reinforcement learning can refer to a learning method in which an agent defined in an environment learns to select an action or sequence of actions that maximizes the cumulative reward in each state.

Among artificial neural networks, machine learning implemented as a deep neural network (DNN) including a plurality of hidden layers is also called deep learning (deep learning), and deep learning is a part of machine learning. Hereinafter, machine learning is used in a sense including deep learning.

<Robot>

A robot can mean a machine that automatically handles or operates a task given by its own capabilities. In particular, a robot having a function of recognizing an environment and performing an operation by self-judgment may be referred to as an intelligent robot.

Robots can be classified into industrial, medical, home, military, etc. depending on the purpose or field of use.

The robot may be provided with a driving unit including an actuator or a motor to perform various physical operations such as moving the robot joints. In addition, the movable robot includes a wheel, a brake, a propeller, and the like in the driving unit, and may travel on the ground or fly in the air through the driving unit.

<Self-Driving>

Autonomous driving refers to a technology that drives itself, and an autonomous driving vehicle refers to a vehicle that travels without or with minimal manipulation of a user.

For example, autonomous driving includes technology for maintaining a driving lane, technology for automatically adjusting speed such as adaptive cruise control, technology for automatically driving along a predetermined route, technology for automatically setting a route when a destination is set, etc. All of these can be included.

The vehicle includes a vehicle having only an internal combustion engine, a hybrid vehicle having both an internal combustion engine and an electric motor, and an electric vehicle having only an electric motor, and may include not only automobiles, but also trains, motorcycles, and the like.

In this case, the autonomous vehicle can be viewed as a robot having an autonomous driving function.

<Extended Reality (XR)>

The extended reality is a generic term for virtual reality (VR), augmented reality (AR), and mixed reality (MR). VR technology provides only CG images of objects or backgrounds in the real world, AR technology provides virtual CG images on top of images of real objects, and MR technology is a computer that mixes and combines virtual objects in the real world. graphic technology.

MR technology is similar to AR technology in that it shows both real and virtual objects. However, there is a difference in that in AR technology, a virtual object is used in a form that complements a real object, whereas in MR technology, a virtual object and a real object are used with equal characteristics.

XR technology can be applied to HMD (Head-Mount Display), HUD (Head-Up Display), mobile phone, tablet PC, laptop, desktop, TV, digital signage, etc. can be called

1 shows an AI device 100 according to an embodiment of the present invention.

AI device 100 is TV, projector, mobile phone, smart phone, desktop computer, notebook computer, digital broadcasting terminal, PDA (personal digital assistants), PMP (portable multimedia player), navigation, tablet PC, wearable device, set-top box (STB) ), a DMB receiver, a radio, a washing machine, a refrigerator, a desktop computer, a digital signage, a robot, a vehicle, etc., may be implemented as a stationary device or a movable device.

Referring to FIG. 1 , the terminal 100 includes a communication unit 110 , an input unit 120 , a learning processor 130 , a sensing unit 140 , an output unit 150 , a memory 170 and a processor 180 , etc. may include

The communication unit 110 may transmit/receive data to and from external devices such as other AI devices 100a to 100e or the AI server 200 using wired/wireless communication technology. For example, the communication unit 110 may transmit and receive sensor information, a user input, a learning model, a control signal, and the like with external devices.

In this case, the communication technology used by the communication unit 110 includes GSM (Global System for Mobile communication), CDMA (Code Division Multi Access), LTE (Long Term Evolution), 5G, WLAN (Wireless LAN), Wi-Fi (Wireless-Fidelity) ), Bluetooth??, RFID (Radio Frequency Identification), Infrared Data Association (IrDA), ZigBee, NFC (Near Field Communication), and the like.

The input unit 120 may acquire various types of data.

In this case, the input unit 120 may include a camera for inputting an image signal, a microphone for receiving an audio signal, a user input unit for receiving information from a user, and the like. Here, by treating the camera or the microphone as a sensor, a signal obtained from the camera or the microphone may be referred to as sensing data or sensor information.

The input unit 120 may acquire training data for model training and input data to be used when acquiring an output using the training model. The input unit 120 may acquire raw input data, and in this case, the processor 180 or the learning processor 130 may extract an input feature by preprocessing the input data.

The learning processor 130 may train a model composed of an artificial neural network by using the training data. Here, the learned artificial neural network may be referred to as a learning model. The learning model may be used to infer a result value with respect to new input data other than the training data, and the inferred value may be used as a basis for a decision to perform a certain operation.

In this case, the learning processor 130 may perform AI processing together with the learning processor 240 of the AI server 200 .

In this case, the learning processor 130 may include a memory integrated or implemented in the AI device 100 . Alternatively, the learning processor 130 may be implemented using the memory 170 , an external memory directly coupled to the AI device 100 , or a memory maintained in an external device.

The sensing unit 140 may acquire at least one of internal information of the AI device 100 , information on the surrounding environment of the AI device 100 , and user information by using various sensors.

At this time, sensors included in the sensing unit 140 include a proximity sensor, an illuminance sensor, an acceleration sensor, a magnetic sensor, a gyro sensor, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, an optical sensor, a microphone, and a lidar. , radar, etc.

The output unit 150 may generate an output related to sight, hearing, or touch.

In this case, the output unit 150 may include a display unit that outputs visual information, a speaker that outputs auditory information, and a haptic module that outputs tactile information.

The memory 170 may store data supporting various functions of the AI device 100 . For example, the memory 170 may store input data obtained from the input unit 120 , learning data, a learning model, a learning history, and the like.

The processor 180 may determine at least one executable operation of the AI device 100 based on information determined or generated using a data analysis algorithm or a machine learning algorithm. In addition, the processor 180 may control the components of the AI device 100 to perform the determined operation.

To this end, the processor 180 may request, search, receive, or utilize the data of the learning processor 130 or the memory 170, and perform a predicted operation or an operation determined to be preferable among the at least one executable operation. It is possible to control the components of the AI device 100 to execute.

In this case, when the connection of the external device is required to perform the determined operation, the processor 180 may generate a control signal for controlling the corresponding external device and transmit the generated control signal to the corresponding external device.

The processor 180 may obtain intention information with respect to a user input, and determine a user's requirement based on the obtained intention information.

In this case, the processor 180 uses at least one of a speech to text (STT) engine for converting a voice input into a character string or a natural language processing (NLP) engine for obtaining intention information of a natural language, Intention information corresponding to the input may be obtained.

At this time, at least one of the STT engine and the NLP engine may be configured as an artificial neural network, at least a part of which is learned according to a machine learning algorithm. And, at least one or more of the STT engine or the NLP engine is learned by the learning processor 130, or learned by the learning processor 240 of the AI server 200, or learned by distributed processing thereof. it could be

The processor 180 collects history information including the user's feedback on the operation contents or operation of the AI device 100 and stores it in the memory 170 or the learning processor 130, or the AI server 200 It can be transmitted to an external device. The collected historical information may be used to update the learning model.

The processor 180 may control at least some of the components of the AI device 100 in order to drive an application program stored in the memory 170 . Furthermore, in order to drive the application program, the processor 180 may operate two or more of the components included in the AI device 100 in combination with each other.

2 shows an AI server 200 according to an embodiment of the present invention.

Referring to FIG. 2 , the AI server 200 may refer to a device that trains an artificial neural network using a machine learning algorithm or uses a learned artificial neural network. Here, the AI server 200 may be configured with a plurality of servers to perform distributed processing, and may be defined as a 5G network. In this case, the AI server 200 may be included as a part of the AI device 100 to perform at least a part of AI processing together.

The AI server 200 may include a communication unit 210 , a memory 230 , a learning processor 240 , a processor 260 , and the like.

The communication unit 210 may transmit/receive data to and from an external device such as the AI device 100 .

The memory 230 may include a model storage unit 231 . The model storage unit 231 may store a model (or artificial neural network, 231a) being trained or learned through the learning processor 240 .

The learning processor 240 may train the artificial neural network 231a using the training data. The learning model may be used while being mounted on the AI server 200 of the artificial neural network, or may be used while being mounted on an external device such as the AI device 100 .

The learning model may be implemented in hardware, software, or a combination of hardware and software. When a part or all of the learning model is implemented in software, one or more instructions constituting the learning model may be stored in the memory 230 .

The processor 260 may infer a result value with respect to new input data using the learning model, and may generate a response or a control command based on the inferred result value.

3 shows an AI system 1 according to an embodiment of the present invention.

Referring to FIG. 3 , the AI system 1 includes at least one of an AI server 200 , a robot 100a , an autonomous vehicle 100b , an XR device 100c , a smart phone 100d , or a home appliance 100e . It is connected to the cloud network 10 . Here, the robot 100a to which the AI technology is applied, the autonomous driving vehicle 100b, the XR device 100c, the smart phone 100d, or the home appliance 100e may be referred to as AI devices 100a to 100e.

The cloud network 10 may constitute a part of the cloud computing infrastructure or may refer to a network existing in the cloud computing infrastructure. Here, the cloud network 10 may be configured using a 3G network, a 4G or Long Term Evolution (LTE) network, or a 5G network.

That is, each of the devices 100a to 100e and 200 constituting the AI system 1 may be connected to each other through the cloud network 10 . In particular, each of the devices 100a to 100e and 200 may communicate with each other through the base station, but may also directly communicate with each other without passing through the base station.

The AI server 200 may include a server performing AI processing and a server performing an operation on big data.

The AI server 200 includes at least one of the AI devices constituting the AI system 1, such as a robot 100a, an autonomous vehicle 100b, an XR device 100c, a smartphone 100d, or a home appliance 100e, and It is connected through the cloud network 10 and may help at least a part of AI processing of the connected AI devices 100a to 100e.

In this case, the AI server 200 may train an artificial neural network according to a machine learning algorithm on behalf of the AI devices 100a to 100e, and directly store the learning model or transmit it to the AI devices 100a to 100e.

At this time, the AI server 200 receives input data from the AI devices 100a to 100e, infers a result value with respect to the input data received using the learning model, and provides a response or control command based on the inferred result value. It can be generated and transmitted to the AI devices 100a to 100e.

Alternatively, the AI devices 100a to 100e may infer a result value with respect to input data using a direct learning model, and generate a response or a control command based on the inferred result value.

Hereinafter, various embodiments of the AI devices 100a to 100e to which the above-described technology is applied will be described. Here, the AI devices 100a to 100e shown in FIG. 3 can be viewed as specific examples of the AI device 100 shown in FIG. 1 .

<AI+Robot>

The robot 100a may be implemented as a guide robot, a transport robot, a cleaning robot, a wearable robot, an entertainment robot, a pet robot, an unmanned flying robot, etc. to which AI technology is applied.

The robot 100a may include a robot control module for controlling an operation, and the robot control module may mean a software module or a chip implemented as hardware.

The robot 100a acquires state information of the robot 100a by using sensor information obtained from various types of sensors, detects (recognizes) surrounding environments and objects, generates map data, moves path and travels A plan may be determined, a response to a user interaction may be determined, or an action may be determined.

Here, the robot 100a may use sensor information obtained from at least one sensor among LiDAR, radar, and camera in order to determine a movement route and a travel plan.

The robot 100a may perform the above-described operations using a learning model composed of at least one artificial neural network. For example, the robot 100a may recognize a surrounding environment and an object using a learning model, and may determine an operation using the recognized surrounding environment information or object information. Here, the learning model may be directly learned from the robot 100a or learned from an external device such as the AI server 200 .

In this case, the robot 100a may perform an operation by generating a result using a direct learning model, but transmits sensor information to an external device such as the AI server 200 and receives the result generated accordingly to perform the operation You may.

The robot 100a determines a movement path and travel plan using at least one of map data, object information detected from sensor information, or object information obtained from an external device, and controls the driving unit to apply the determined movement path and travel plan. Accordingly, the robot 100a may be driven.

The map data may include object identification information for various objects disposed in a space in which the robot 100a moves. For example, the map data may include object identification information for fixed objects such as walls and doors and movable objects such as flowerpots and desks. In addition, the object identification information may include a name, a type, a distance, a location, and the like.

In addition, the robot 100a may perform an operation or drive by controlling the driving unit based on the user's control/interaction. In this case, the robot 100a may acquire intention information of an interaction according to a user's motion or voice utterance, determine a response based on the acquired intention information, and perform the operation.

<AI + Autonomous Driving>

The autonomous driving vehicle 100b may be implemented as a mobile robot, a vehicle, an unmanned aerial vehicle, etc. by applying AI technology.

The autonomous driving vehicle 100b may include an autonomous driving control module for controlling an autonomous driving function, and the autonomous driving control module may mean a software module or a chip implemented as hardware. The autonomous driving control module may be included as a component of the autonomous driving vehicle 100b, or may be configured and connected to the outside of the autonomous driving vehicle 100b as separate hardware.

The autonomous vehicle 100b obtains state information of the autonomous vehicle 100b using sensor information obtained from various types of sensors, detects (recognizes) surrounding environments and objects, generates map data, A movement route and a driving plan may be determined, or an operation may be determined.

Here, the autonomous driving vehicle 100b may use sensor information obtained from at least one sensor among a lidar, a radar, and a camera, similarly to the robot 100a, in order to determine a moving route and a driving plan.

In particular, the autonomous vehicle 100b may receive sensor information from external devices to recognize an environment or object for an area where the field of view is obscured or an area over a certain distance, or receive information recognized directly from external devices. .

The autonomous vehicle 100b may perform the above-described operations by using a learning model composed of at least one artificial neural network. For example, the autonomous driving vehicle 100b may recognize a surrounding environment and an object using a learning model, and may determine a driving route using the recognized surrounding environment information or object information. Here, the learning model may be directly learned from the autonomous vehicle 100b or learned from an external device such as the AI server 200 .

In this case, the autonomous vehicle 100b may generate a result by using the direct learning model and perform the operation, but it operates by transmitting sensor information to an external device such as the AI server 200 and receiving the result generated accordingly. can also be performed.

The autonomous vehicle 100b determines a movement path and a driving plan by using at least one of map data, object information detected from sensor information, or object information acquired from an external device, and controls the driving unit to determine the movement path and driving The autonomous vehicle 100b may be driven according to a plan.

The map data may include object identification information for various objects disposed in a space (eg, a road) in which the autonomous vehicle 100b travels. For example, the map data may include object identification information for fixed objects such as street lights, rocks, and buildings, and movable objects such as vehicles and pedestrians. In addition, the object identification information may include a name, a type, a distance, a location, and the like.

Also, the autonomous vehicle 100b may perform an operation or drive by controlling the driving unit based on the user's control/interaction. In this case, the autonomous vehicle 100b may acquire intention information of an interaction according to a user's motion or voice utterance, determine a response based on the obtained intention information, and perform the operation.

<AI+XR>

The XR device 100c is AI technology applied, so a head-mount display (HMD), a head-up display (HUD) provided in a vehicle, a television, a mobile phone, a smart phone, a computer, a wearable device, a home appliance, and a digital signage , a vehicle, a stationary robot, or a mobile robot.

The XR device 100c analyzes 3D point cloud data or image data acquired through various sensors or from an external device to generate location data and attribute data for 3D points, thereby providing information on surrounding space or real objects. It can be obtained and output by rendering the XR object to be output. For example, the XR apparatus 100c may output an XR object including additional information on the recognized object to correspond to the recognized object.

The XR apparatus 100c may perform the above-described operations using a learning model composed of at least one artificial neural network. For example, the XR apparatus 100c may recognize a real object from 3D point cloud data or image data using a learning model, and may provide information corresponding to the recognized real object. Here, the learning model may be directly learned from the XR device 100c or learned from an external device such as the AI server 200 .

In this case, the XR device 100c may perform an operation by generating a result using the direct learning model, but it transmits sensor information to an external device such as the AI server 200 and receives the result generated accordingly to perform the operation. can also be done

<AI+Robot+Autonomous Driving>

The robot 100a may be implemented as a guide robot, a transport robot, a cleaning robot, a wearable robot, an entertainment robot, a pet robot, an unmanned flying robot, etc. to which AI technology and autonomous driving technology are applied.

The robot 100a to which AI technology and autonomous driving technology are applied may mean a robot having an autonomous driving function or a robot 100a that interacts with the autonomous driving vehicle 100b.

The robot 100a having an autonomous driving function may collectively refer to devices that move by themselves according to a given movement line without user's control or by determining a movement line by themselves.

The robot 100a with the autonomous driving function and the autonomous driving vehicle 100b may use a common sensing method to determine one or more of a moving route or a driving plan. For example, the robot 100a having an autonomous driving function and the autonomous driving vehicle 100b may determine one or more of a movement route or a driving plan by using information sensed through lidar, radar, and camera.

The robot 100a interacting with the autonomous driving vehicle 100b exists separately from the autonomous driving vehicle 100b and is linked to an autonomous driving function inside the autonomous driving vehicle 100b or connected to the autonomous driving vehicle 100b. It is possible to perform an operation associated with the user on board.

At this time, the robot 100a interacting with the autonomous driving vehicle 100b acquires sensor information on behalf of the autonomous driving vehicle 100b and provides it to the autonomous driving vehicle 100b, or obtains sensor information and obtains information about the surrounding environment or By generating object information and providing it to the autonomous driving vehicle 100b, the autonomous driving function of the autonomous driving vehicle 100b may be controlled or supported.

Alternatively, the robot 100a interacting with the autonomous driving vehicle 100b may monitor a user riding in the autonomous driving vehicle 100b or control a function of the autonomous driving vehicle 100b through interaction with the user. . For example, when it is determined that the driver is in a drowsy state, the robot 100a may activate an autonomous driving function of the autonomous driving vehicle 100b or assist in controlling a driving unit of the autonomous driving vehicle 100b. Here, the function of the autonomous driving vehicle 100b controlled by the robot 100a may include not only an autonomous driving function, but also a function provided by a navigation system or an audio system provided in the autonomous driving vehicle 100b.

Alternatively, the robot 100a interacting with the autonomous driving vehicle 100b may provide information or assist a function to the autonomous driving vehicle 100b from the outside of the autonomous driving vehicle 100b. For example, the robot 100a may provide traffic information including signal information to the autonomous driving vehicle 100b, such as a smart traffic light, or interact with the autonomous driving vehicle 100b, such as an automatic electric charger for an electric vehicle. You can also automatically connect an electric charger to the charging port.

<AI+Robot+XR>

The robot 100a may be implemented as a guide robot, a transport robot, a cleaning robot, a wearable robot, an entertainment robot, a pet robot, an unmanned flying robot, a drone, etc. to which AI technology and XR technology are applied.

The robot 100a to which the XR technology is applied may mean a robot that is a target of control/interaction within an XR image. In this case, the robot 100a is distinguished from the XR device 100c and may be interlocked with each other.

When the robot 100a, which is the target of control/interaction within the XR image, obtains sensor information from sensors including a camera, the robot 100a or the XR device 100c generates an XR image based on the sensor information. and the XR apparatus 100c may output the generated XR image. In addition, the robot 100a may operate based on a control signal input through the XR device 100c or a user's interaction.

For example, the user can check the XR image corresponding to the viewpoint of the remotely linked robot 100a through an external device such as the XR device 100c, and adjust the autonomous driving path of the robot 100a through interaction or , control motion or driving, or check information of surrounding objects.

<AI+Autonomous Driving+XR>

The autonomous vehicle 100b may be implemented as a mobile robot, a vehicle, an unmanned aerial vehicle, etc. by applying AI technology and XR technology.

The autonomous driving vehicle 100b to which the XR technology is applied may mean an autonomous driving vehicle equipped with a means for providing an XR image or an autonomous driving vehicle subject to control/interaction within the XR image. In particular, the autonomous driving vehicle 100b, which is the target of control/interaction within the XR image, is distinguished from the XR device 100c and may be interlocked with each other.

The autonomous driving vehicle 100b having means for providing an XR image may obtain sensor information from sensors including a camera, and output an XR image generated based on the acquired sensor information. For example, the autonomous vehicle 100b may provide an XR object corresponding to a real object or an object in a screen to the passenger by outputting an XR image with a HUD.

In this case, when the XR object is output to the HUD, at least a portion of the XR object may be output to overlap the real object to which the passenger's gaze is directed. On the other hand, when the XR object is output to the display provided inside the autonomous vehicle 100b, at least a portion of the XR object may be output to overlap the object in the screen. For example, the autonomous vehicle 100b may output XR objects corresponding to objects such as a lane, other vehicles, traffic lights, traffic signs, two-wheeled vehicles, pedestrians, and buildings.

When the autonomous driving vehicle 100b, which is the subject of control/interaction in the XR image, acquires sensor information from sensors including a camera, the autonomous driving vehicle 100b or the XR device 100c performs An XR image is generated, and the XR apparatus 100c may output the generated XR image. In addition, the autonomous vehicle 100b may operate based on a control signal input through an external device such as the XR device 100c or a user's interaction.

The artificial intelligence server 200 may include the configuration of the AI device 100 described with reference to FIGS. 1 to 3 and perform functions of the AI device 100 .

The electronic device 300 may include the configuration of the AI device 100 described with reference to FIGS. 1 to 3 and perform functions of the AI device 100 .

The artificial intelligence device described in this specification may be the artificial intelligence server 200 or the electronic device 300 .

Hereinafter, the artificial intelligence device may act as a gait prediction device.

The gait prediction device may act as a resource for predicting a gait path of a person in a plurality of electronic devices including robots and autonomous driving devices.

There may be many representative behavioral patterns of pedestrians. For example, it may include a pattern according to interaction between person and person and between person and group. The behavior pattern may include three types. [1] Collision avoidance: avoiding collision when there is a person or group on one's walking path, [2] Joining group: moving toward the group to which one belongs and joining; [3] Interaction: may involve stopping and acting in groups.

As another example, it may include a pattern according to the interaction between people and the whole crowd. For example, [4] Walking at the same pace: This may involve a pedestrian walking with a large crowd in the periphery. This interaction between pedestrians can be understood as following social norms and actions. It is important that the GCN reflects the behavioral pattern of the above pedestrian.

On the other hand, in GCN, what kind of edge is configured and connected between each pedestrian node is a problem. As a simple example, if the distance between two pedestrians is close, when a high-edge connection is provided, the person closest to the pedestrian is determined as the main object of interest. In this case, the person in question can be avoided. This can be solved for the situation corresponding to [1] among the above four behavior patterns. However, it leads to a prediction result that is not restored to the original destination direction after evasion. Also, in [2] and [3] above, prediction results that join with another person and avoid it without interacting with the target may appear frequently. Even in the case of [4], the connection to other pedestrians farther away is weak, which may lead to prediction results of not recognizing people around them, and walking at different speeds.

<GCN: graph convolutional network>

The graph convolutional network may update features for each layer using Equation (1).

where H is the features of all nodes. σ is the nonlinear activation function. In the example, PReLU was used. A is an adjacency matrix. W is a learnable weight matrix for each layer. l is the index of the layer. The node may refer to all pedestrians. The connection of at least two nodes may be referred to as a graph. The end of the graph may be referred to as an edge. The English characters in the specification may have the same meaning.

For other information related to the graph convolutional network, reference may be made to the contents of Citation 1. For the understanding of the present invention, the description of Citation 1 is included in the description of the present invention.

Using Equation 1 above, it is possible to update features for gait path prediction using GCN.

According to Equation 1, [1] Collision avoidance, [2] Joining group, [3] Interaction, and [4] Walking such as surrounding flow at the same pace) cannot be updated correctly.

This embodiment proposes an artificial intelligence device for predicting a walking path by reflecting the walking characteristics of the pedestrian.

4 is a view for explaining an artificial intelligence device according to an embodiment of the present invention.

Referring to FIG. 4 , a first feature updater 1000 for updating a first feature and a second feature updater 2000 for updating a second feature may be included. The first feature may be a set of at least two features. The second feature may be included in the first feature.

A configuration of the first feature updater 1000 will be described. The first feature updater 1000 may include a process of updating a plurality of complementary spatiotemporal features by separating them into time and space axes. The process of updating the feature on the spatial axis may include a multi-relational GCN 1010 . An embodiment may include two features. The two characteristics may include relative displacement information between pedestrians and distance information between pedestrians. Here, the distance information may be Euclidean distance information. In the drawing, it can be seen that the distance information 1011 and the relative displacement information 1012 are separated.

The multi-relationship sharing unit can be expressed by Equation (2).

Here, R may refer to the multi-relationship. The R element may include the relative displacement information and the distance information. The R may include more information. The multi-relationship sharing unit 1010 is different from the cited document 1 in which only the relative displacement information is handled. According to this technique, [2] joining group, and [3] interaction could be predicted well. For example, it can be well predicted that pedestrians will join and interact with a group.

The first feature updater 1000 may include a multi-relationship merging unit 1040 . The features separated by the multi-relationship sharing unit 1010 and updated in the spatial axis may be merged in the time axis by the multi-relationship merging unit 1040 . The multi-relational merging unit 1040 may be implemented as a 2D convolutional layer having a 3*1 filter along a channel axis. Of course, this does not exclude other means of merging features.

The first feature updater 1000 may include a graph processing unit 1020 that supplements insufficient learning materials. The graph processing unit 1020 may utilize dropedge technology. The graph processing unit 1020 may create a subgraph by randomly removing edges from the input graph.

The operation of the graph processing unit 1020 may be expressed by Equation (3).

은 상대적인 인접 메트릭스이고, 인접 매트릭스(A)의 정규화된 텀이다.

은 오리지널 노드에 대하여 랜덤하게 연결된 에지이다. p는 드랍율(dropping rate)이다.

는 원소들 간의 곱 연산자이다. 상기 그래프 가공부에 의해서 그래프는 늘어날 수 있다. here,

is the relative adjacency matrix, and is the normalized term of the adjacency matrix (A).

is a randomly connected edge with respect to the original node. p is the dropping rate.

is the multiplication operator between elements. The graph may be stretched by the graph processing unit.

If the learning material is sufficient, the graph processing unit may not be included. However, preferably, it may be a more accurate feature update including learning materials.

The first feature update unit 1000 includes a multi-scale setting unit 1030 . The multi-scale setting unit 1030 uses k-hop neighborhoods to collect features. The operation of the multi-scale setting unit 1030 can be expressed by Equation (4).

Here, k may mean a hop. It can be seen that the adjacency matrix (A) and the weight matrix (W) are given for each hop. Comparing Equation 4 with Equation 1, there is a difference in that learning is performed for each hop by the multi-scale setting unit 1030 . Comparing Equation 4 and Equation 2, Equation 2 shows that Equation 4 is performed with respect to the distance information and the relative displacement information, respectively.

By the multi-scale setting unit 1030 , it is possible to learn the relationship between distant nodes as well as close nodes from any one node. Of course, a separate weight may be applied to the learning. Rich relationships can be used for learning by the multi-scale setting unit 1030 .

When Equation 3 is applied, there may be a problem in that the edge weight increases in an exponential function. When a node belonging to a certain group has a strong connection, there may be a problem that the weight is strongly biased only to the node belonging to the group. It will be described with reference to the drawings.

5 is a view for explaining the operation of the multi-scale setting unit. 4 and 5, the operation of the multi-scale setting unit will be described.

Referring to FIG. 5( a ), there is a group 1 in which nodes 6 - 9 are strongly connected to each other, and there is a group 2 in which nodes 3 - node 4 are connected. In this case, the reference group must consider the group 1 and the group 2. Nevertheless, the weight is biased only between the nodes in the group 1.

Referring to the upper four adjacent matrices of FIG. 5B , it can be seen that a strong bias is applied between the adjacent nodes of the group 1.

In order to solve this problem and unravel the strong bond between nodes (disentangle), a graph set can be provided by using the distance scale between nodes (pedestrians) as an edge weight. The distance scale may correspond to the hop.

Using the distance scale, the neighbor matrix can be newly defined as in Equation (4).

이면, 강한 관계가 설정될 수 있다. 다시 말하면, 상기 멀티스케일설정부는, 상기 노드 간의 거리가 일정한 범위 내이면, 인접 메트릭스의 요소를 동일하게 처리할 수 있다. 이때의 강한 관계는 한 쌍의 노드가 멀리 떨어져 있는 경우에도 마찬가지이다. 그러므로, 노드 간의 거리가 멀거나 가깝거나 관계없이 노드 간의 상호작용이 학습될 수 있다. 도 5(b)의 오른쪽 아래쪽 네 개의 인접 메트릭스를 보면, 모든 노드도 강하게 묶여있는 것을 볼 수 있다. 특히, 인접노드 간의 바이어스가 없어진 것을 볼 수 있다. 결국, 인접 노드 간의 결속이 풀려있는 것을 볼 수 있다. 도 5(b)는 k에 따라서 결속이 되는 것을 나타낸다.Here, Υ is a set of scales and can be set to 0, 0.5, 1, 2, 4, and of course other cases are possible. i and j represent nodes. According to this, the distance between nodes is

, a strong relationship can be established. In other words, if the distance between the nodes is within a certain range, the multi-scale setting unit may equally process the elements of the adjacent matrix. The strong relationship at this time is the same even when a pair of nodes are far apart. Therefore, the interaction between nodes can be learned regardless of whether the distance between the nodes is far or close. Looking at the four adjacent matrices in the lower right of FIG. 5(b), it can be seen that all nodes are also strongly bound. In particular, it can be seen that the bias between adjacent nodes is eliminated. Eventually, it can be seen that the bond between adjacent nodes is released. Figure 5 (b) shows the binding according to k.

는 가중치가 적용되지 않은 서브 그래프 들의 세트로 주어질 수 있다. As another aspect, according to Equation 5,

can be given as a set of unweighted subgraphs.

는 수학식 5의

로 변경되어 치환되는 것에 의해서, 결국, 수학식 2로 구현될 수 있다. 상기 수학식 2는 수학식 3으로 제공되는 모든 서브 그래프에 대하여 수행될 수 있다. 상기 그래프 가공부(1020)이 제공되지 않는 경우에는, 입력되는 그래프에 대하여 수행될 수 있다. of Equation 4

is in Equation 5

By being changed to and substituted, it can be finally implemented as Equation (2). Equation 2 may be performed for all subgraphs provided by Equation 3 above. When the graph processing unit 1020 is not provided, it may be performed on an input graph.

According to the above description, the first feature updater 1000 may use both absolute distance information 1011 and relative distance information 1012 between pedestrians. Accordingly, group joining and interaction can be predicted well.

The first feature updater 1000 acts as a multi-scale setting unit 1030 that classifies different reference groups according to distance. Objects to avoid and objects to interact with are now separated into different groups and can be distinguished. Accordingly, distant nearby crowds that are not avoidable may also be grouped into other groups to act. In the end, it is possible to predict walking at a similar speed by checking the walking speed of people around, as well as group joining and interaction.

The second feature updater 2000 may update the second feature.

An operation of the second feature updater will be described with reference to FIG. 4 .

The second feature updater 2000 may compensate for the accumulated error. The second feature may be described as a feature of a global temporal aggregation (GTA), that is, a GTA feature. The second feature may act to reflect the first walking path prior to the pedestrian avoiding the obstruction. The second feature may act to reference the first walking path to the second walking path after the pedestrian avoids the obstacle. The second feature may act to add a vector component of the first walking path to the second walking path after the pedestrian avoids the obstacle.

As a conventional method for predicting the avoidance path of pedestrians, a TCN (temporal co-volutional network) technique has been proposed. The TCN technique is 'Bai, S.; Kolter, J. Z.; and Koltun, V. 2018. An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. The technology introduced in 'arXiv preprint arXiv:1803.01271' can be applied.

The TCN technique may compensate by using the second feature to compensate for an increase in the accumulated error. In the embodiment, it can be seen that the second feature is reflected in the results of the two TCN techniques. For this purpose, it can be seen that the second feature adjusting unit 2001 is provided.

The second feature for a predetermined time can be updated by Equation (6).

Here, h is the second feature, and w and b are the learnable kernel weights and biases. The second feature may be an element constituting the first feature indicated by H. The second feature may provide a portion of the first feature.

By reflecting the second feature, it can be seen that the accumulated error is reduced when the pedestrian avoids the avoidance target 2002 . For example, the second walking path C2 when the second feature is applied is the first walking path, which is the original walking path, compared to the second walking path C1 when only the conventional TCN technique is applied. It can be confirmed that the predicted return is faster.

According to the second feature updater, it is possible to predict a vector capable of recovering a walking direction through a window viewing the avoidance pattern of the entire walking path at a time. By adding the vector, overshooting of the path by the avoidance target 2002 can be prevented and the path can be restored back to the original destination direction.

6 is a table for explaining the effect.

6, Linear regression, Social-LSTM (Alahi et al. 2016) nition (CVPR), Social-GAN-P (Gupta et al. 2018), SoPhie (Sadeghian et al. 2019), PIF (Liang et al.) al. 2019), Reciprocal Learning (Sun, Zhao, and He 2020), STGAT (Huang et al. 2019), Social-BiGAT (Kosaraju et al. 2019), Social-STGCNN (Mohamed et al. 2020), and Bonn The inventor, DMRGCN, was tested in various environments. The results of the experiment were evaluated as average displacement error (ADE) and final displacement error (FDE).

Bold type is the device with the best effect, and the underline is the device with the second best effect. According to the experiment, it can be confirmed that the artificial intelligence device of the present invention predicts the walking path well in various environments.

The pedestrian path according to the present invention can be predicted more accurately.

Claims (11)

It is an artificial intelligence device that learns features of a graph convolutional network,
A first feature update unit for updating a first feature of the walking path,
The first feature update unit includes a multi-scale setting unit that provides a graph set by using a distance scale between nodes as an edge weight
Pedestrian path prediction device. The method of claim 1,
The graph convolutional network is

update the first feature for each layer by
Here, H is the features of all nodes, σ is the nonlinear activation function, A is the adjacency matrix, W is the learnable weight matrix for each layer, and l is the index of the layer,
Pedestrian path prediction device. 3. The method of claim 2,
The multi-scale setting unit,
The adjacency matrix

defined as,
Υ is the scale set,
Pedestrian path prediction device. 4. The method of claim 3,
Υ comprises at least two of 0, 0.5, 1, 2, 4,
Pedestrian path prediction device The method of claim 1,
The multi-scale setting unit, if the distance between the nodes is within a certain range, processing the elements of the adjacent matrix as the same value,
Pedestrian path prediction device. The method of claim 1,
The first feature update unit,
a multi-relationship sharing unit that separates and updates at least two complementary features; and
Including a multi-relationship merging unit for merging the features
Pedestrian path prediction device. 7. The method of claim 6,
The complementary feature includes the relative displacement information of the pedestrian and the distance information of the pedestrian.
Pedestrian path prediction device. The method of claim 1,
The first feature update unit,
Comprising a graph processing unit for additionally providing a subgraph by removing an edge from the graph
Pedestrian path prediction device. 9. The method of claim 8,
The graph processing unit,

works by

is the normalized term of the adjacency matrix (A),

is a randomly connected edge with respect to the original node, and p is the dropping rate.

is the multiplication operator between elements.
Pedestrian path prediction device. The method of claim 1,
A second feature update unit for updating the second feature is included;
The second feature is a feature that ensures the cumulative error of the walking path, and is a component of the first feature,
Pedestrian path prediction device. 11. The method of claim 10,
The second feature acts to add a vector component of the first walking path before the pedestrian avoids the obstacle to the second walking path after the pedestrian avoids the obstacle
Pedestrian path prediction device.

Images