CN111612823A - Robot autonomous tracking method based on vision - Google Patents

Robot autonomous tracking method based on vision Download PDF

Info

Publication number
CN111612823A
CN111612823A CN202010436066.0A CN202010436066A CN111612823A CN 111612823 A CN111612823 A CN 111612823A CN 202010436066 A CN202010436066 A CN 202010436066A CN 111612823 A CN111612823 A CN 111612823A
Authority
CN
China
Prior art keywords
grid
robot
target
coordinate
coordinates
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010436066.0A
Other languages
Chinese (zh)
Inventor
范江波
郑昆
徐云水
赵泽彪
邱平
李锐
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhaotong Power Supply Bureau of Yunnan Power Grid Co Ltd
Original Assignee
Zhaotong Power Supply Bureau of Yunnan Power Grid Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhaotong Power Supply Bureau of Yunnan Power Grid Co Ltd filed Critical Zhaotong Power Supply Bureau of Yunnan Power Grid Co Ltd
Priority to CN202010436066.0A priority Critical patent/CN111612823A/en
Publication of CN111612823A publication Critical patent/CN111612823A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10016—Video; Image sequence
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Image Analysis (AREA)
  • Manipulator (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses a vision-based robot autonomous tracking method, and belongs to the technical field of automatic robots. The method comprises the steps of determining a target position and sensing a planning path of a surrounding environment to avoid obstacles by using a GOTURN tracking model and a SegNet-based scene segmentation model, establishing a mapping relation between a pixel coordinate and a two-dimensional plane coordinate by a system through parallel projection, constructing a local grid map under a polar coordinate by a visual positioning fusion environment mathematical model, and finally planning an optimal track between a robot and a target through an A-ray algorithm; the method can realize stronger tracking capability and longer tracking distance in indoor and corridor environments.

Description

Robot autonomous tracking method based on vision
Technical Field
The invention relates to a vision-based robot autonomous tracking method, and belongs to the technical field of automatic robots.
Background
In recent studies, Sidenbladh et al uses skin color segmentation to detect a target person from an RGB image and track the head of the person, and the detection result is susceptible to ambient lighting, target motion and other factors, thereby affecting the system performance. Adiwahono et al propose a flexible leg feature detection method, but are sensitive to changes in human pose. Gockley et al used a standard particle filtering algorithm to segment laser scanner data and track potential targets, and authors designed two tracking schemes, direction tracking and path tracking, but the system lacked recognizable visual information and its effective range was less than 3.5 m. Wang et al uses a binocular camera based three-dimensional reconstruction to obtain the relative position between the robot and the person, and sets the distance between the person and the robot to be between 1-2.5 m in order to obtain sufficient reconstruction accuracy. The above conventional methods are more accurate tracking of targets and tracking target trajectories but do not employ path planning, which means that the robot can only track nearby targets.
Disclosure of Invention
The invention provides a vision-based robot autonomous tracking method.
The technical scheme of the invention is as follows: a robot autonomous tracking method based on vision comprises the steps of firstly, manually selecting a tracking target, and setting input of CNN as the target and a surrounding background area; dividing the environment into different categories according to the training data set; establishing a local map under a polar coordinate, taking the robot as the center of a polar coordinate system, establishing a conversion relation between a pixel coordinate point and a plane coordinate point and a grid coordinate, and calculating a pixel of a grid as a statistical grid obstacle; and setting a grid coordinate of the robot start, obtaining the target position of the robot through a target tracking algorithm, and finally completing the optimal path search by using an A-algorithm.
The method comprises the following specific steps:
step1, manually selecting destination target, setting CNN input as target area and background area around target, if atthe tracked object in the t-1 frame is a bounding box (C)x,Cy) Is w, h, then at the next time t, is centered on the center (C)x,Cy) Width and height are respectively k1w、k1h is extracted as a search area in the previous frame; wherein, Cx、CyIs a point coordinate, t is a time, k1w、k1h is the width and height of the background region, k1Is a coefficient;
step2, after searching the destination target, dividing the scene into different categories according to the training data set: segmenting the scene into different categories, such as people, places, trees or other categories, according to the training data set, thereby segmenting the scene;
step3, obtaining a scene segmentation model, and conveniently establishing a mapping relation between the image coordinates of the destination target and the plane actual coordinates through a visual positioning algorithm: setting a camera coordinate system { C } as a reference coordinate system, wherein a plane pi exists in the space, and a three-dimensional point p exists on the planei＝[xi,yi,zi,1]TN, n is the total number of three-dimensional points; deriving p from an aperture imaging modeliNormalized projection coordinates of and piHomogeneous image coordinates of (a);
step4, establishing a local map with polar coordinate center of the robot:
the polar coordinates are expressed in the form of (r, t); the pole diameter and the pole angle of each grid are obtained according to the following formula:
Figure BDA0002502296740000021
Figure BDA0002502296740000022
Rc＝Cres·r1,r1∈(1,Rdim)
Figure BDA0002502296740000023
Figure BDA0002502296740000024
Figure BDA0002502296740000025
Cconst＝Rdim·Cres
wherein x and y represent p in a reference coordinate systemiAbscissa and ordinate of coordinate point, RcRepresenting the radius length from the grid of the linear region to the origin of coordinates, also called the grid radius; corresponding RnIs the radial length, r, of the grid in the non-linear region1、r2A radius number representing a grid; cresRepresents the length of a single grid of the linear region, hcamRepresenting the distance of the optical center of the camera to the ground, hRminRepresents the distance, R, from the top of the camera's field of view to the ground at 1.8mdimDenotes the number of linear region grids, hRdimDenotes the number of non-linear region grids, TdimThe robot is divided into a plurality of parts in an angle dimension, namely each grid occupies a space of 2 degrees;
the obtained pole diameter rpSubstituting into formula to obtain grid coordinate in radial direction, i.e. polar diameter, tpI.e. the coordinates in the angular direction, i.e. the polar angle; the ratio lambda of the pixel points marked as obstacles in the grid to all the pixel points in the grid>At threshold, the grid is an obstacle, otherwise it is flat;
step5, using the a-algorithm to complete the optimal path search: and setting a starting point to be positioned at an origin of polar coordinates during path searching, setting a polar angle of 90 degrees as an initial orientation of the robot, obtaining a target position through an autonomous tracking algorithm, and finally completing the optimal path searching by using an A-x algorithm.
In Step2, SegNet is used to perform scene segmentation.
The invention has the beneficial effects that: the vision robot tracking technology is one of the key technologies of the intelligent mobile robot, and has wide application prospects in related fields. The robot has wide application value in the fields of museum navigation, pedestrian tracking, hospital assistance and the like: the method comprises the steps of determining a target position and sensing a planning path of a surrounding environment to avoid obstacles by using a GOTURN tracking model and a SegNet-based scene segmentation model, establishing a mapping relation between a pixel coordinate and a two-dimensional plane coordinate by a system through parallel projection, constructing a local grid map under a polar coordinate by a visual positioning fusion environment mathematical model, and finally planning an optimal track between a robot and a target through an A-ray algorithm; the method can realize stronger tracking capability and longer tracking distance in indoor and corridor environments.
Drawings
FIG. 1 is a schematic diagram of a CNN target tracking model;
FIG. 2 is a diagram of a scene segmentation network architecture;
FIG. 3 is a perspective projection view;
FIG. 4 is a schematic diagram of dividing a vertical grid;
FIG. 5 is a block diagram of a grid map;
FIG. 6 is a target occlusion test chart;
fig. 7 is a robot indoor follow-up experiment.
Detailed Description
Example 1: as shown in fig. 1-7, a vision-based robot autonomous tracking method, first, a tracking target is manually selected, and input of CNN is set as the target and a surrounding background area; dividing the environment into different categories according to the training data set; establishing a local map under a polar coordinate, taking the robot as the center of a polar coordinate system, establishing a conversion relation between a pixel coordinate point and a plane coordinate point and a grid coordinate, and calculating a pixel of a grid as a statistical grid obstacle; and setting a grid coordinate of the robot start, obtaining the target position of the robot through a target tracking algorithm, and finally completing the optimal path search by using an A-algorithm. According to the method, a similar GOTURN tracking model is selected, as shown in figure 1, a CNN model is used for extracting basic features of a current frame and a previous frame of image respectively, then the image features are fused and input into a 3-layer full-connection layer, the full-connection layer calculates the relative motion trend of a target by comparing the target features of the previous frame with the current image features, and the last layer outputs coordinates of a bounding Box of the tracked target.
Further, the specific steps of the method may be set as follows:
step1, manually selecting destination target, setting CNN input as target area and background area around target, if the tracked object in t-1 frame is bounding box (C)x,Cy) Is w, h, then at the next time t, is centered on the center (C)x,Cy) Width and height are respectively k1w、k1h is extracted as a search area in the previous frame; wherein, Cx、CyIs a point coordinate, t is a time, k1w、k1h is the width and height of the background region, k1Is a coefficient;
step2, after searching the destination target, dividing the scene into different categories according to the training data set: segmenting the scene into different categories, such as people, places, trees or other categories, according to the training data set, thereby segmenting the scene; to improve model performance and robustness, 2200 additional images were captured in the test environment, which were extracted from a video sequence lasting about 1 hour. These images were annotated using LabelMe and their labels were obtained. The floor is marked 1 and the obstacle is marked 0. 2000 images in the data set were randomly selected as a training set, and the remaining 200 images were used as a test set. The model was initially learned at a rate of 0.0001 and trained on a PyTorch platform.
Step3, obtaining a scene segmentation model, and conveniently establishing a mapping relation between the image coordinates of the destination target and the plane actual coordinates through a visual positioning algorithm: setting a camera coordinate system { C } as a reference coordinate system, wherein a plane pi exists in the space, and a three-dimensional point p exists on the planei＝[xi,yi,zi,1]TN, n is the total number of three-dimensional points; deriving p from an aperture imaging modeliToNormalized projection coordinates and piHomogeneous image coordinates of (a); specifically, the method comprises the following steps:
the normalized projection coordinates of a point in space and p can be determined by camera-internal parametersiThe homogeneous image coordinates of (a) are converted to each other.
The mathematical expression for the viewing plane of the coordinate system { a } may be expressed as Z ═ 0 and pi in the form of projective geometryA＝[nATdA]T＝[0 0 1 0]T，nADenotes the plane piANormal vector in coordinate system { A }, dADenotes the origin of coordinates of the coordinate system { A } to the plane piAThe distance of (c). If the camera coordinate system { C } is taken as the reference coordinate system, the view plane Z ═ 0 can be expressed as pic＝[ncTdc]TWherein:
Figure BDA0002502296740000041
Figure BDA0002502296740000042
in FIG. 3, the camera optical center is defined
Figure BDA0002502296740000043
And normalizing the projection coordinates m to a space line L
Figure BDA0002502296740000044
And m is written in homogeneous coordinate form:
Figure BDA0002502296740000051
B＝[mT1]T
the Plucker matrix of the line L is L ═ ABT-BATBy calculating the line L and the plane πcThe intersection point coordinates of the two-dimensional projection coordinate m are obtainedcProjection coordinate P in (1):
Figure BDA0002502296740000052
and (3) converting the coordinates under the camera coordinate system into a ground coordinate system { A }, and performing homogeneous coordinate transformation:
Figure BDA0002502296740000053
after the coordinates of point P are found, P can be calculatedA(ii) a Establishing a mapping relation between the image coordinate and the plane actual coordinate through the step;
step4, the grid chart is the most widely applied form in the field of path planning, but the grid chart is not suitable for robot navigation, because the angle and the distance are usually discrete, a local map with a polar coordinate center of a robot is selected to be established.
As in fig. 4, an area within 1.8 meters from the camera is divided into 9 cells, each of which has a length of 0.2 meters. For grids exceeding 1.8 meters, the length increases in a non-linear fashion, so that linear and non-linear intervals can be divided in the grid map, respectively. The radius of each grid is determined by the following system of equations:
Rc＝Cres·r1,r1∈(1,Rdim)
Figure BDA0002502296740000054
Figure BDA0002502296740000055
Figure BDA0002502296740000056
Cconst＝Rdim·Cres
wherein R iscRepresenting the radius length from the grid of the linear region to the origin of coordinates, also called the grid radius; corresponding RnIs half of the grid of the non-linear regionRadial length, r1、r2A radius number representing a grid; cres0.2m represents the length of a single grid of the linear region, hcam0.72m denotes the distance between the optical center of the camera and the ground, hRminRepresents the distance, R, from the top of the camera's field of view to the ground at 1.8mdim9 denotes the number of linear region grids, hRdimDenotes the number of non-linear region grids, T dim180 denotes dividing the robot in angular dimension 180 times a turn, i.e. each grid occupies a space of 2 degrees in size; cconstIndicating the linear zone length, △ h indicates 1/10 of the distance from the camera's optical center to the ground.
The polar coordinates are expressed in the form of (r, t); the pole diameter and the pole angle of each grid are obtained according to the following formula:
Figure BDA0002502296740000061
Figure BDA0002502296740000062
wherein x and y represent p in a reference coordinate systemiThe abscissa and ordinate of the coordinate point. The obtained pole diameter rpSubstituting into equation (2) to obtain grid coordinate in radial direction, i.e. polar diameter, tpI.e. the coordinates in the angular direction, i.e. the polar angle; thus, a pixel coordinate point is established>Plane coordinate point->The conversion relation of the grid coordinates, and the mapping relation between the grid coordinates and the pixel coordinates is stored; the ratio lambda of the pixel points marked as obstacles in the grid to all the pixel points in the grid>At 0.1, the grid is an obstacle, otherwise it is flat; therefore, when the robot receives one frame of image, the construction of the local map can be completed through scene segmentation and coordinate mapping.
Step5, using the a-algorithm to complete the optimal path search: as in fig. 5, a polar grid map is constructed with dimensions 16 × 90. When path searching is carried out, a starting point is set to be located at the origin of polar coordinates, a polar angle of 90 degrees is set to be the initial orientation of the robot, the grid coordinates of the robot start are (0,45), the target position can be obtained through an autonomous tracking algorithm, and finally the optimal path searching is completed through an A-x algorithm.
In Step2, SegNet is used to perform scene segmentation.
In order to test the robustness of the robot to the target occlusion processing, in the process of normal forward movement of the robot, a pedestrian walks back and forth before tracking a target, so that a scene that the target is occluded in an actual situation is simulated. Fig. 6 shows that when the target is temporarily blocked, the tracker is not easily interfered by the blocking object, and after the blocking disappears, the robot can still stably track the target pedestrian. Since the target is located right in front of the robot, the planned path is a straight line trajectory directed directly to the target. When the target is lost or is shielded for a long time, the robot enters an emergency stop state, so that the robot is protected.
The robot target following task is to move towards the target's position until a certain safe distance is maintained from the target. The laboratory area is about 8m 10m, and facilities such as pedestrians, tables and chairs, garbage cans, sofas and the like are arranged indoors, so that the indoor environment is relatively complex. The experimental results are shown in fig. 7, which shows path generation, scene segmentation and grid maps, respectively. Considering the volume of the robot, the planned path needs to be manually fine-tuned. We have 2 methods to deal with this problem, one is to artificially enlarge the boundary of the obstacle so that the segmentation map includes the size of the robot, which can effectively avoid collision with the obstacle while keeping the model of the robot itself unchanged. And secondly, manually adding an offset term on a planned path, wherein the number of the offset grids is determined by the size of the grids and the distance of the grids. We take a second strategy that is easier to implement, where the planned path is the path with the bias added. Experimental results also prove that the robot can stably move to a target position, and meanwhile, obstacles can be effectively avoided, and the indoor experimental effect initially meets the system requirements.
While the present invention has been described in detail with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, and various changes can be made without departing from the spirit of the present invention within the knowledge of those skilled in the art.

Claims (3)

1. A robot autonomous tracking method based on vision is characterized in that: firstly, manually selecting a tracking target, and setting the input of CNN as the target and a surrounding background area; dividing the environment into different categories according to the training data set; establishing a local map under a polar coordinate, taking the robot as the center of a polar coordinate system, establishing a conversion relation between a pixel coordinate point and a plane coordinate point and a grid coordinate, and calculating a pixel of a grid as a statistical grid obstacle; and setting a grid coordinate of the robot start, obtaining the target position of the robot through a target tracking algorithm, and finally completing the optimal path search by using an A-algorithm.
2. The vision-based autonomous robot tracking method according to claim 1, characterized in that: the method comprises the following specific steps:
step1, manually selecting destination target, setting CNN input as target area and background area around target, if the tracked object in t-1 frame is bounding box (C)x,Cy) Is a BoundingBox of w, h, respectively, then at the next time t, one is extracted at the t-1 th frame to be (c)x,cy) Is a center, and has a width and a height of k1w、k1h as a search area; wherein, Cx、CyIs a point coordinate, t is a time, k1w、k1h is the width and height of the background region, k1Is a coefficient;
step2, after searching the destination target, dividing the scene into different categories according to the training data set: segmenting the scene into different categories, such as people, places, trees or other categories, according to the training data set, thereby segmenting the scene;
step3, obtaining a scene segmentation model, and conveniently establishing a mapping relation between the image coordinates of the destination target and the plane actual coordinates through a visual positioning algorithm: let the camera coordinate system { C } be the reference coordinate system, in spaceThere is a plane pi on which there is a three-dimensional point pi＝[xi,yi,zi,1]TN, n is the total number of three-dimensional points; deriving p from an aperture imaging modeliNormalized projection coordinates of and piHomogeneous image coordinates of (a);
step4, establishing a local map with polar coordinate center of the robot:
the polar coordinates are expressed in the form of (r, t); the pole diameter and the pole angle of each grid are obtained according to the following formula:
Figure FDA0002502296730000011
Figure FDA0002502296730000012
Rc＝Cres·r1,r1∈(1,Rdim)
Figure FDA0002502296730000021
Figure FDA0002502296730000022
Figure FDA0002502296730000023
Cconst＝Rdim·Cres
wherein x and y represent p in a reference coordinate systemiAbscissa and ordinate of coordinate point, RcRepresenting the radius length from the grid of the linear region to the origin of coordinates, also called the grid radius; corresponding RnIs the radial length, r, of the grid in the non-linear region1、r2A radius number representing a grid; cresRepresents the length of a single grid of the linear region, hcamRepresenting the distance of the optical center of the camera to the ground, hRminRepresents the distance, R, from the top of the camera's field of view to the ground at 1.8mdimDenotes the number of linear region grids, hRdimDenotes the number of non-linear region grids, TdimThe robot is divided into a plurality of parts in an angle dimension, namely each grid occupies a space of 2 degrees;
the obtained pole diameter rpSubstituting into formula to obtain grid coordinate in radial direction, i.e. polar diameter, tpI.e. the coordinates in the angular direction, i.e. the polar angle; the ratio lambda of the pixel points marked as obstacles in the grid to all the pixel points in the grid>At threshold, the grid is an obstacle, otherwise it is flat;
step5, using the a-algorithm to complete the optimal path search: and setting a starting point to be positioned at an origin of polar coordinates during path searching, setting a polar angle of 90 degrees as an initial orientation of the robot, obtaining a target position through a target tracking algorithm, and finally completing the optimal path searching by using an A-x algorithm.
3. The vision-based autonomous robot tracking method according to claim 2, characterized in that: in Step2, SegNet is used to perform scene segmentation.
CN202010436066.0A 2020-05-21 2020-05-21 Robot autonomous tracking method based on vision Pending CN111612823A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010436066.0A CN111612823A (en) 2020-05-21 2020-05-21 Robot autonomous tracking method based on vision

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010436066.0A CN111612823A (en) 2020-05-21 2020-05-21 Robot autonomous tracking method based on vision

Publications (1)

Publication Number Publication Date
CN111612823A true CN111612823A (en) 2020-09-01

Family

ID=72199836

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010436066.0A Pending CN111612823A (en) 2020-05-21 2020-05-21 Robot autonomous tracking method based on vision

Country Status (1)

Country Link
CN (1) CN111612823A (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112364693A (en) * 2020-10-12 2021-02-12 星火科技技术(深圳)有限责任公司 Barrier identification method, device and equipment based on binocular vision and storage medium
CN112530267A (en) * 2020-12-17 2021-03-19 河北工业大学 Intelligent mechanical arm teaching method based on computer vision and application
CN113096104A (en) * 2021-04-15 2021-07-09 北京达佳互联信息技术有限公司 Training method and device of target segmentation model and target segmentation method and device
CN113361519A (en) * 2021-05-21 2021-09-07 北京百度网讯科技有限公司 Target processing method, training method of target processing model and device thereof
CN113421275A (en) * 2021-05-13 2021-09-21 影石创新科技股份有限公司 Image processing method, image processing device, computer equipment and storage medium
CN114353818A (en) * 2021-12-31 2022-04-15 上海洛轲智能科技有限公司 Target object following method, apparatus, device, medium, and computer program product

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101860729A (en) * 2010-04-16 2010-10-13 天津理工大学 Target tracking method for omnidirectional vision
US20100316287A1 (en) * 2009-06-11 2010-12-16 Duong Tuan A Method and system for object recognition search
CN102848388A (en) * 2012-04-05 2013-01-02 上海大学 Service robot locating and grabbing method based on multiple sensors
CN108460790A (en) * 2018-03-29 2018-08-28 西南科技大学 A kind of visual tracking method based on consistency fallout predictor model
CN108765450A (en) * 2018-05-21 2018-11-06 南京邮电大学 intelligent robot visual tracking method based on deep learning
CN109190635A (en) * 2018-07-25 2019-01-11 北京飞搜科技有限公司 Target tracking method, device and electronic equipment based on classification CNN
CN109682381A (en) * 2019-02-22 2019-04-26 山东大学 Big visual field scene perception method, system, medium and equipment based on omnidirectional vision
CN109949375A (en) * 2019-02-02 2019-06-28 浙江工业大学 A kind of mobile robot method for tracking target based on depth map area-of-interest
CN110751670A (en) * 2018-07-23 2020-02-04 中国科学院长春光学精密机械与物理研究所 Target tracking method based on fusion
CN111062328A (en) * 2019-12-18 2020-04-24 中新智擎科技有限公司 Image processing method and device and intelligent robot

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100316287A1 (en) * 2009-06-11 2010-12-16 Duong Tuan A Method and system for object recognition search
CN101860729A (en) * 2010-04-16 2010-10-13 天津理工大学 Target tracking method for omnidirectional vision
CN102848388A (en) * 2012-04-05 2013-01-02 上海大学 Service robot locating and grabbing method based on multiple sensors
CN108460790A (en) * 2018-03-29 2018-08-28 西南科技大学 A kind of visual tracking method based on consistency fallout predictor model
CN108765450A (en) * 2018-05-21 2018-11-06 南京邮电大学 intelligent robot visual tracking method based on deep learning
CN110751670A (en) * 2018-07-23 2020-02-04 中国科学院长春光学精密机械与物理研究所 Target tracking method based on fusion
CN109190635A (en) * 2018-07-25 2019-01-11 北京飞搜科技有限公司 Target tracking method, device and electronic equipment based on classification CNN
CN109949375A (en) * 2019-02-02 2019-06-28 浙江工业大学 A kind of mobile robot method for tracking target based on depth map area-of-interest
CN109682381A (en) * 2019-02-22 2019-04-26 山东大学 Big visual field scene perception method, system, medium and equipment based on omnidirectional vision
CN111062328A (en) * 2019-12-18 2020-04-24 中新智擎科技有限公司 Image processing method and device and intelligent robot

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112364693A (en) * 2020-10-12 2021-02-12 星火科技技术(深圳)有限责任公司 Barrier identification method, device and equipment based on binocular vision and storage medium
CN112364693B (en) * 2020-10-12 2024-04-16 星火科技技术(深圳)有限责任公司 Binocular vision-based obstacle recognition method, device, equipment and storage medium
CN112530267A (en) * 2020-12-17 2021-03-19 河北工业大学 Intelligent mechanical arm teaching method based on computer vision and application
CN113096104A (en) * 2021-04-15 2021-07-09 北京达佳互联信息技术有限公司 Training method and device of target segmentation model and target segmentation method and device
CN113421275A (en) * 2021-05-13 2021-09-21 影石创新科技股份有限公司 Image processing method, image processing device, computer equipment and storage medium
CN113361519A (en) * 2021-05-21 2021-09-07 北京百度网讯科技有限公司 Target processing method, training method of target processing model and device thereof
CN113361519B (en) * 2021-05-21 2023-07-28 北京百度网讯科技有限公司 Target processing method, training method of target processing model and device thereof
CN114353818A (en) * 2021-12-31 2022-04-15 上海洛轲智能科技有限公司 Target object following method, apparatus, device, medium, and computer program product
CN114353818B (en) * 2021-12-31 2024-05-14 上海洛轲智能科技有限公司 Target object following method, apparatus, device, medium and computer program product

Similar Documents

Publication Publication Date Title
CN111612823A (en) Robot autonomous tracking method based on vision
CN110349250B (en) RGBD camera-based three-dimensional reconstruction method for indoor dynamic scene
CN113269098B (en) Multi-target tracking positioning and motion state estimation method based on unmanned aerial vehicle
WO2021196294A1 (en) Cross-video person location tracking method and system, and device
Xu et al. SLAM of Robot based on the Fusion of Vision and LIDAR
Zhang et al. Robust appearance based visual route following for navigation in large-scale outdoor environments
Yue et al. Fast 3D modeling in complex environments using a single Kinect sensor
CN110163963B (en) Mapping device and mapping method based on SLAM
CN106650701A (en) Binocular vision-based method and apparatus for detecting barrier in indoor shadow environment
CN111161334B (en) Semantic map construction method based on deep learning
CN112258553B (en) All-time target tracking method based on multi-source image fusion
CN112541938A (en) Pedestrian speed measuring method, system, medium and computing device
CN115205339A (en) Mobile robot target following method based on improved YOLOX
CN111489392A (en) Single target human motion posture capturing method and system in multi-person environment
CN116879870A (en) Dynamic obstacle removing method suitable for low-wire-harness 3D laser radar
Pauls et al. Automatic mapping of tailored landmark representations for automated driving and map learning
EP4053801A1 (en) Landmark learning and localization without labels
Ristić-Durrant et al. Low-level sensor fusion-based human tracking for mobile robot
Uzawa et al. Dataset Generation for Deep Visual Navigation in Unstructured Environments
Ortigosa et al. Fuzzy free path detection from disparity maps by using least-squares fitting to a plane
Choi et al. Autonomous homing based on laser-camera fusion system
Einhorn et al. Monocular detection and estimation of moving obstacles for robot navigation
Xue Analysis of simultaneous localization and mapping technology for mobile robot based on binocular vision
Zhao et al. Pedestrian detection and tracking based on 2D Lidar and RGB-D camera
CN114326695B (en) Self-propelled vehicle following system and self-propelled vehicle following method

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20200901

RJ01 Rejection of invention patent application after publication