A high-precision visual positioning method without blind spots in ultra-long continuous large intervals

By placing background reference objects and visual sensors in the detection area, and using similar triangle principles to build a visual positioning model, the blind spots and accuracy problems of object positioning in the grown area are solved, efficient and fully covered object positioning is achieved, and cost is reduced.

CN115115706BActive Publication Date: 2024-11-26GUANGZHOU SHUNRU ELECTRONIC TECH CO LTD
View PDF 7 Cites 0 Cited by

Patent Information

Application Number
CN202210763146.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2022-06-29
Publication Date
2024-11-26
Estimated Expiration
2042-06-29

AI Technical Summary

Technical Problem

The existing ranging positioning methods have problems such as blind spots when positioning objects in a grown area, degraded long-distance positioning accuracy and performance, high system impacted by the environment, poor robustness, and excessive cost of high-precision positioning equipment.

Method used

Using the high-precision visual positioning method without blind spots in the ultra-long continuous interval, by placing two background reference objects in the detection area and setting up visual sensors on them, a visual positioning model is constructed using the principle of similar triangles to calculate the position of the object to be positioned Information to achieve full coverage and high-precision object positioning.

Benefits of technology

It realizes object detection and single object positioning within the range of 1m-1000m in length and 0.3m-100m in width, eliminates detection blind spots, improves positioning accuracy and efficiency, reduces comprehensive costs, is more adaptable, and is suitable for foreign objects in important facilities. Intrusion detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115115706B_ABST
    Figure CN115115706B_ABST
Patent Text Reader

Abstract

The present invention discloses a high-precision visual positioning method without blind spots in an ultra-long continuous large interval, wherein two background reference objects with visual sensors set at the upper and lower vertices are placed in the detection area; the visual sensors on the two background reference objects are respectively arranged relatively; the center projection value of the object to be positioned and the center projection value of the object to be positioned on the two background reference objects are measured in two cases of whether there is an object to be positioned; the focus of the visual sensor lens, the center projection value of the object to be positioned and the center projection value of the object to be positioned are vertically projected to the positioning detection plane to obtain the focus of the positioning visual sensor, the positioning background reference line segment, the positioning projection value, and the length of the positioning background reference line segment and the distance between them; the positioning information equation is obtained according to the principle of similar triangles to obtain one or more of the position information and size information of a single object to be positioned on the positioning detection plane. The present invention has the advantages of comprehensive positioning, larger range, high efficiency, stronger adaptability, higher precision, and low comprehensive cost.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a computer vision-based object positioning method, and in particular to a high-precision visual positioning method without blind spots in an ultra-long continuous large interval. Background Art

[0002] Common positioning technologies can generally be divided into three categories: positioning technology based on external signal sources, positioning technology based on natural signal sources, and positioning technology based on ranging. Among them, positioning technology based on external signal sources relies on external signal sources and terminal sensors, and common ones include: WIFI positioning, cellular mobile networks, satellite positioning, Bluetooth positioning, ZigBee positioning, ultra-wideband (UWB) positioning, etc.; indoor positioning technology based on natural signal sources can achieve positioning only by relying on terminal sensors, and common ones include: inertial navigation, geomagnetic navigation, etc.; positioning technology based on ranging achieves positioning by measuring the distance of the target, and common ones include: infrared positioning, laser positioning, binocular vision positioning, ultrasonic positioning, and lidar, etc.

[0003] Due to their own characteristics, the above positioning technologies are suitable for different application fields. Therefore, there is a need to develop positioning technologies suitable for various application scenarios. Positioning technology based on external signal sources and indoor positioning technology based on natural signal sources, these two types of positioning methods need to carry corresponding sensors themselves, and are limited by specific signal sources, which will have obvious limitations in some application scenarios. In common application scenarios such as security, foreign body intrusion, and position measurement, when unknown objects or biological intrusions occur, positioning can only be performed through ranging positioning methods that do not rely on the objects to be located. The common ranging positioning methods are mainly divided into: vision, infrared beam detection, laser beam detection, lidar, etc. The positioning in general scenarios is relatively complete, but when it comes to positioning objects in large areas, most methods still have obvious defects. The following is an analysis of each item:

[0004] (1) Visual category.

[0005] Binocular measurement: Two cameras form a binocular, based on the principle of parallax and using imaging equipment to obtain two images of the object to be measured from different positions, and the position deviation between corresponding points of the image is calculated to locate the object to be measured. Binocular has the advantages of high efficiency, appropriate accuracy, simple system structure, and low cost, but its accuracy is affected by camera performance, lighting, focal length and baseline length (the distance between the two cameras), and it is difficult to measure objects at a distance. In addition, binocular measurement is very sensitive to ambient light and difficult to cope with environmental changes. At the same time, the focal length of the lens selected by the camera has a great influence on imaging. Due to the limitations of the imaging principle, when the camera focal length and camera parameters are determined, the camera can only obtain clear imaging within a certain distance range, and cannot achieve clear imaging at full distance. It is necessary to dynamically and synchronously adjust the focal length and baseline distance of the binocular camera to achieve full-distance ranging, but due to the complexity of the algorithm and the difficulty of accuracy, there are many limitations in application. Binocular is generally only suitable for positioning objects within a few dozen meters.

[0006] The invention application with application publication number CN105651258A proposes an active perspective binocular distance measurement system and method, which captures the target area from the binocular image, matches the best matching target image through a fast matching algorithm, and measures the distance of the measured target according to the active perspective distance measurement model. The system searches and matches through the left and right cameras, which solves some of the defects of lack of initiative, but the system is still affected by light too much, the perspective is too small to match when detecting a large area, and the binocular camera baseline distance is close and the measurement accuracy is not high.

[0007] (2) Infrared and laser types.

[0008] Infrared beam detection mainly uses infrared rays to form a warning line between the transmitter and the receiver, and uses an occlusion alarm method. The protection zone formed is a single beam, or a mesh defense zone composed of multiple beams. Since the infrared light spot will gradually increase with the increase of the detection distance, it is not suitable for long-distance detection. The general applicable range is about 0 to 50 meters. When long-distance detection is required, it can be replaced by laser beam detection. This type of system can only determine which plane the object is located on through the blocked beam, and cannot accurately determine the coordinates of the object to be located. There are large blind spots in the detection area, and positioning measurement of large areas cannot be achieved.

[0009] The utility model patent with the authorization announcement number CN211928699U proposes a comprehensive perimeter security system with infrared thermal imaging verification. The system is based on the principles of infrared counter-radiation and image processing. The system first monitors intrusion through infrared counter-radiation. When an intrusion occurs, the image acquisition device is called for secondary confirmation. If an intrusion is found, an alarm is triggered. The system solves the problem of high false alarm rate of infrared counter-radiation, but the system still has defects such as discontinuous detection area, unsuitable for long-distance detection, and inability to locate.

[0010] The utility model patent with the authorization announcement number CN203825755U proposes a laser beam intrusion alarm and positioning sensor system. The system is used for long-distance intrusion alarm and positioning sensor systems. It can monitor long-distance security lines and accurately obtain the interval location information of the intrusion. The system sets up a two-way laser beam device. When an object invades, it blocks the laser, causing the laser signal to be interrupted. The intrusion position can be calculated by the time difference of the two signal interruptions. The system can perform positioning monitoring in long-distance areas, but the system requires that the two time channels are completely aligned, with a certain accuracy error. The system can only monitor the edge of the perimeter and cannot achieve continuous area monitoring. It can only locate on the laser line, with a large monitoring and positioning blind area.

[0011] (3) Laser radar.

[0012] The laser radar emits a beam of laser through a group of laser transmitters. After the laser beam encounters an object, it is diffusely reflected and returns to the laser receiver. The radar module calculates the distance between the transmitter and the object based on the time of sending and receiving the signal. In addition to the distance information, it is also used to detect whether there is an intrusion. This solution is usually used for the detection of obstacles in driverless cars. Its advantages are long detection distance, high resolution, and little environmental interference. The disadvantage is that the target range scanned by a single-line laser radar is limited, which easily leads to the loss of the detection target. Multi-line laser radar has the problems of high cost, reliability to be improved, and large detection blind area. Multi-line laser radar emits laser beams. The farther the area, the sparser the laser beam spacing, and the larger the detection blind area. At the same time, the longer the distance, the lower the probability of the laser beam reflecting back. It usually has higher performance at 0-150m.

[0013] In summary, since many application scenarios require object detection and positioning in large areas, such as the long and wide continuous area of ​​230m to 460m and 1.2m to 2m between the high-speed rail and intercity platform doors before the train leaves the station, it is necessary to quickly determine whether there are foreign objects and locate the location of foreign objects. The positioning of unknown objects can only be achieved through ranging positioning, so ranging positioning has a high research value. However, the existing ranging positioning methods have obvious defects in locating objects in large areas, such as: the existence of detection blind spots, reduced long-distance positioning accuracy and performance, high environmental impact of the system, poor robustness, and high cost of high-precision positioning equipment. Summary of the invention

[0014] The purpose of the present invention is to overcome the above-mentioned problems and provide a high-precision visual positioning method without blind spots in an ultra-long continuous large interval. The visual positioning method can accurately, effectively and quickly determine the position of the object to be located, and at the same time has the advantages of more comprehensive positioning, larger positioning range, higher efficiency, stronger adaptability, higher positioning accuracy and low overall cost.

[0015] The purpose of the present invention is achieved through the following technical solutions:

[0016] A high-precision visual positioning method without blind spots in an ultra-long continuous large interval comprises the following steps: placing two background reference objects in a detection area, wherein visual sensors are arranged at the upper and lower vertices of the two background reference objects; the visual sensors on the two background reference objects are arranged relatively to each other; measuring the center projection value of the object to be positioned and the center projection value of the object to be positioned on the two background reference objects in two situations of whether there is an object to be positioned; obtaining the focus of the positioning visual sensor, the positioning background reference line segment, the positioning projection value, the length of the positioning background reference line segment and the distance between the two positioning background reference line segments by vertically projecting the focus of the visual sensor lens, the center projection value of the object to be positioned and the center projection value of the object to be positioned onto a positioning detection plane; the formation of the positioning detection plane needs to meet the following conditions:

[0017] At least one vertex of each of the two background reference objects is located on the positioning detection plane; the lines connecting the four vertex projection points of the vertical projections of the four vertices of the two background reference objects on the positioning detection plane form a rectangle; the lens focus of the visual sensor is vertically projected on the positioning detection plane, and the lens focus projection point is located on the straight line connecting the two vertex projection points of the vertical projection of the background reference object on the positioning detection plane, and the lines connecting the focal points of the positioning visual sensor form a rectangle; the angle θ between the lines connecting the vertices of the background reference objects and the lines connecting the vertical projection points of the vertices of the background reference objects on the positioning detection plane can be obtained, 0°≤θ<90°;

[0018] The visual positioning model is obtained according to the principle of similar triangles, and the positioning information equation is obtained according to the visual positioning model. One or more of the position information I1 and H, and the size information X and Y of the object to be positioned on the positioning detection plane can be solved. The specific operation steps are as follows:

[0019] According to actual needs, two background reference objects and four visual sensors are set up to establish a positioning detection plane;

[0020] The image information of the detection area is collected by the visual sensor, and the image of the background reference object is captured to determine the position of the background reference object in the image as the area of ​​interest; the situation in the area of ​​interest is confirmed, the image of the background reference object when there is no object to be located is obtained, and the background reference line segment is vertically projected onto the positioning detection plane to locate the background reference line segment, and its length and the distance information between them are measured;

[0021] Detect the change value between the real-time image in the area of ​​interest in the current image and the complete image of the background reference object when it is not blocked. If the change value is greater than the set threshold, it is determined that there is an object to be located in the detection area, and proceed to the next step;

[0022] Synchronously obtain four real-time images of the area of ​​interest taken by four visual sensors, in which the center of the object to be located is projected onto the background reference object; analyze the four real-time images and the corresponding complete images of the background reference object respectively; in the image in which the background reference object is blocked by the object to be located, measure the upper remaining length and the lower remaining length of the background reference object in the image after being blocked by the object to be located, and then vertically project them onto the positioning detection plane to obtain the positioning projection value;

[0023] According to the principle of similar triangles, a visual positioning model is constructed based on the positioning projection value of the object to be positioned, the length of the positioning background reference line segment and the distance between two positioning background reference line segments to obtain a positioning information equation. Based on the existence of the positioning information equation, the combined equation is solved to obtain one or more of the position information I1 and H and the size information X and Y of the object to be positioned on the positioning detection plane.

[0024] In a preferred embodiment of the present invention, the distance H between two background reference objects is 1m-1000m, and the height of the background reference object L is 0.3m-100m.

[0025] In a preferred embodiment of the present invention, the background reference object is a straight or curved object or a straight and curved object.

[0026] A preferred embodiment of the present invention, wherein the background reference object is a plate-shaped marker or a self-luminous LED light board, characterized in that it allows the visual sensor to obtain a clear central projection image of the object to be located on the background reference object, such as one or more of a luminous marker, a non-luminous marker, and a color-changing marker that contrasts sharply with the object to be located and the surrounding environment.

[0027] In a preferred embodiment of the present invention, the positioning projection value L *n The calculation method is:

[0028] Taking the lens focus of the visual sensor as the projection center, project the image center of the object to be positioned onto the background reference object, measure the length of the uppermost vertex of the projection image on the background reference object from the upper vertex of the background reference object, and the length of the lowermost vertex of the projection image on the background reference object from the lower vertex of the background reference object, and project it vertically onto the positioning detection plane through the angle θ between the connecting line of the background reference vertices and the connecting line of the vertical projection points of the background reference vertices on the positioning detection plane, so as to obtain the positioning projection value L of the object to be positioned on the positioning detection plane *n .

[0029] In a preferred embodiment of the present invention, the shape of the object to be located on the positioning detection plane is regarded as a rectangle parallel to the positioning background reference line segment, and when each visual sensor captures the object to be located and the background reference object is not completely blocked, one or two positioning information equations are obtained by the similar triangle theorem according to the spatial relationship between the object to be located, the background reference object and the visual sensor;

[0030] From top to bottom, the visual sensors on the two background reference objects are numbered 1-2 and 3-4;

[0031] The two positioning information equations of vision sensor No. 1 are:

[0032]

[0033]

[0034] The two positioning information equations of vision sensor No. 2 are:

[0035]

[0036]

[0037] The two positioning information equations of vision sensor No. 3 are:

[0038]

[0039]

[0040] The two positioning information equations of the No. 4 visual sensor are:

[0041]

[0042]

[0043] In the above formula, L *nis the positioning projection value, where * is the number 1-4 of the visual sensor of the image source, n=1 is the remaining length of the upper end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor, and n=2 is the remaining length of the lower end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor; I1 is the shortest distance between the vertical projection point of the upper vertex and the vertical projection point of the lower vertex of the background reference object corresponding to the focus of the current positioning visual sensor calculated on the positioning detection plane; H is the shortest distance between the lowest point of the positioning object and the line connecting the lower vertex projection points of the two positioning background reference objects calculated on the positioning detection plane; X is defined as the calculated length of the positioning object on the positioning detection plane; Y is defined as the calculated height of the positioning object on the positioning detection plane; L is the length of the positioning background reference line segment, that is, the length of the line connecting the upper vertex vertical projection point and the lower vertex vertical projection point of the background reference object on the positioning detection plane; I is the distance between the two positioning background reference line segments, that is, the length of the line connecting the vertical projection points of the corresponding vertices of the background reference objects on both sides on the positioning detection plane.

[0044] In a preferred embodiment of the present invention, the shape of the object to be located on the positioning detection plane is regarded as a line segment parallel to the positioning background reference line segment, and when each visual sensor captures the object to be located and the background reference object is not completely blocked, one or two positioning information equations can be obtained by the similar triangle theorem based on the spatial relationship between the object to be located, the background reference object and the visual sensor;

[0045] From top to bottom, the visual sensors on the two background reference objects are numbered 1-2 and 3-4;

[0046] The two positioning information equations of vision sensor No. 1 are:

[0047]

[0048]

[0049] The two positioning information equations of vision sensor No. 2 are:

[0050]

[0051]

[0052] The two positioning information equations of vision sensor No. 3 are:

[0053]

[0054]

[0055] The two positioning information equations of the No. 4 visual sensor are:

[0056]

[0057]

[0058] In the above formula, L *n is the positioning projection value, where * is the number 1-4 of the visual sensor of the image source, n=1 is the remaining length of the upper end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor, and n=2 is the remaining length of the lower end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor; I1 is the shortest distance between the vertical projection point of the upper vertex and the vertical projection point of the lower vertex of the background reference object corresponding to the focus of the current positioning visual sensor calculated on the positioning detection plane; H is the shortest distance between the lowest point of the positioning object and the line connecting the lower vertex projection points of the two positioning background reference objects calculated on the positioning detection plane; Y is defined as the calculated height of the positioning object on the positioning detection plane; L is the length of the positioning background reference line segment, that is, the length of the line connecting the upper vertex vertical projection point and the lower vertex vertical projection point of the background reference object on the positioning detection plane; I is the distance between the two positioning background reference line segments, that is, the length of the line connecting the vertical projection points of the corresponding vertices of the background reference objects on both sides on the positioning detection plane.

[0059] Compared with the prior art, the present invention has the following characteristics and beneficial effects:

[0060] 1. The visual positioning method of the present invention can accurately, effectively and quickly locate the position information and shape and size information of a single object, and the information obtained is more comprehensive.

[0061] 2. The present invention can realize object detection and precise positioning in ultra-long continuous large areas, and can realize object detection and single object positioning in the range of 1m-1000m in length and 0.3m-100m in width, or in a longer and larger range. The size of the detection area, positioning accuracy, minimum size of the object to be positioned, and the clarity of the background reference obtained by the visual sensor and the accuracy of measuring the length of the background reference are related, and are determined by the image resolution of the visual sensor, the focal length of the lens, and the design quality of the background reference.

[0062] 3. The present invention utilizes four visual sensors for monitoring within the positioning detection area. When an object exists within the detection area, it ensures that at least two cameras in each area can capture the object information. Compared with previous positioning methods, it is more efficient and completely eliminates the detection blind spots within the detection area under ideal conditions.

[0063] 4. The present invention has stronger adaptability. When the information is insufficient to infer the length and height of the object, by simplifying the model, the object position information, the height of the highest point of the object or the height of the lowest point of the object can be solved with almost no decrease in accuracy, thereby realizing the positioning of the object on the positioning detection plane.

[0064] 5. When the present invention is used for long-distance monitoring, the positioning accuracy and positioning performance will not decrease as the distance increases, and a high level can be maintained throughout the entire detection area.

[0065] 6. The positioning method of the present invention has its own background reference, is less affected by environmental interference, has good robustness and low overall cost.

[0066] 7. Compared with the binocular positioning method, the present invention adopts a fixed-focus lens and a fixed camera position, which can realize continuous single object positioning without blind spots at low cost and high precision within two background reference objects. The positioning accuracy is theoretically not affected by the distance of the object. It is particularly suitable for foreign body intrusion detection and rapid removal in important ultra-long continuous large-area facilities, such as airport runways, high-speed railways, and intercity platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] Figure 1 It is a visual positioning model and a schematic diagram of its structure when the shape of the object to be positioned in the ultra-long continuous large interval of the present invention is regarded as a rectangle on the positioning detection plane.

[0068] Figure 2 It is a schematic diagram of the positioning detection plane of a high-precision visual positioning method without blind spots in an ultra-long continuous large interval of the present invention.

[0069] Figure 3 The background reference of the present invention is a schematic diagram of a bar-shaped straight line structure.

[0070] Figure 4 The background reference object of the present invention is a schematic diagram of a strip-shaped curved structure.

[0071] Figure 5 This is a schematic diagram of a standard single-sided machine vision-based structure.

[0072] Figure 6 It is a schematic diagram of a visual positioning model in which the positioning vision sensor 201 of the present invention captures an object and has two positioning information equations.

[0073] Figure 7 It is a schematic diagram of a visual positioning model in which the positioning vision sensor 202 of the present invention captures a situation in which an object is captured and has two positioning information equations.

[0074] Figure 8It is a schematic diagram of a visual positioning model in which the positioning vision sensor 203 of the present invention captures a situation in which an object is captured and has two positioning information equations.

[0075] Fig. 9 It is a schematic diagram of a visual positioning model in which the positioning vision sensor 204 of the present invention captures a situation in which an object is captured and has two positioning information equations.

[0076] Fig.10 The visual positioning model and its structural diagram when the shape of the object to be positioned is regarded as a line segment in the positioning detection plane.

[0077] Fig.11 for Figure 1 Schematic diagram of background reference object images obtained by visual sensors, wherein from left to right are background reference object images obtained by visual sensors 201-204, and the missing area in the figure is the area of ​​object projection.

[0078] Fig.12 It is a schematic diagram of a visual positioning model of the present invention in which an object to be positioned is photographed by four visual sensors and has two positioning information equations.

[0079] The numbers in the figure are explained as follows: 101-the vertical projection of the left background reference object on the positioning detection plane, which is the positioning background reference line segment 101; 102-the vertical projection of the right background reference object on the positioning detection plane, which is the positioning background reference line segment 102; 201-the vertical projection position of the lens focus of visual sensor No. 1 on the positioning detection plane, which is the positioning visual sensor 201; f1 is the lens focus of positioning visual sensor 201; 202-the vertical projection position of visual sensor No. 2 on the positioning detection plane, which is the positioning visual sensor 202; f2 is the lens focus of positioning visual sensor 202; 203-the vertical projection position of visual sensor No. 3 on the positioning detection plane, which is the positioning visual sensor 203; f3 is the lens focus of positioning visual sensor 203; 204-the vertical projection position of the lens focus of visual sensor No. 4 on the positioning detection plane, which is the position of positioning visual sensor 4; f4 is the lens focus of positioning visual sensor 204; 301-the positioning object calculated on the positioning detection plane of the object to be positioned.

[0080] H-the shortest distance from the lowest point of the positioning object to the line connecting the lower vertices of the two positioning background reference line segments calculated on the positioning detection plane; I-the distance between the positioning background reference line segments 101 and 102 on the positioning detection plane; I1-the shortest distance from the positioning object to the positioning background reference line segment 101 calculated on the positioning detection plane; L-the length of the positioning background reference line segment 101 on the positioning detection plane; X-is defined as the calculated length of the positioning object on the positioning detection plane; Y-is defined as the calculated height of the positioning object on the positioning detection plane, that is, the distance from the highest point to the lowest point of the positioning object calculated on the positioning detection plane; L 11 - On the positioning detection plane, the length from the highest point of the center projection of the positioning object calculated by the positioning vision sensor 201 on the positioning background reference line segment 102 to the upper vertex; L 12 - On the positioning detection plane, the length from the lowest point to the lower vertex of the center projection of the positioning object calculated by the positioning vision sensor 201 on the positioning background reference line segment 102; L 21 - On the positioning detection plane, the length from the highest point of the center projection of the positioning object calculated by the positioning vision sensor 202 on the positioning background reference line segment 102 to the upper vertex; L 22 - On the positioning detection plane, the length from the lowest point to the lower vertex of the center projection of the positioning object calculated by the positioning vision sensor 202 on the positioning background reference line segment 102; L 31 - On the positioning detection plane, the length from the highest point of the center projection of the positioning object calculated by the positioning vision sensor 203 on the positioning background reference line segment 101 to the upper vertex; L 32 - On the positioning detection plane, the length from the lowest point to the lower vertex of the center projection of the positioning object on the positioning background reference line segment 101 calculated and photographed by the positioning vision sensor 203; L 41 - On the positioning detection plane, the length from the highest point of the center projection of the positioning object calculated by the positioning vision sensor 204 on the positioning background reference line segment 101 to the upper vertex; L 42 - On the positioning detection plane, the length from the lowest point to the lower vertex of the center projection of the positioning object on the positioning background reference line segment 101 calculated by the positioning vision sensor 204 . DETAILED DESCRIPTION

[0081] In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention is further described below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.

[0082] See also Figure 1The present embodiment provides a high-precision visual positioning method without blind spots in an ultra-long continuous large interval. The method can be achieved by setting two strip background reference objects separated by a certain distance and four visual sensors set at both ends of the background reference objects.

[0083] Furthermore, the visual positioning model of this embodiment is established based on two dimensions. The background reference object, the object to be positioned, and the focus of the visual sensor lens are all projected vertically onto the positioning detection plane to establish a positioning system model, and the positioning background reference line segments 101 and 102, the positioning visual sensors 201, 202, 203, 204, and the positioning projection values ​​are obtained. The positioning information equation is obtained based on similar triangles.

[0084] The positioning detection plane is a plane where the visual positioning model is located for easy measurement. The construction of the positioning detection plane needs to meet the following conditions: 1) There is at least one vertex on each of the background reference objects on both sides located on the positioning detection plane; 2) The four vertex projection points of the vertical projection of the background reference objects on both sides on the positioning detection plane (such as the attached Figure 2 In the figure, points f1, f2, f3, and f4 can form a rectangle; 3) the lens focus of the visual sensor is vertically projected on the positioning detection plane, and the lens focus projection point is located on the straight line connecting the two vertex projection points of the background reference object vertically projected on the positioning detection plane; 4) the angle θ (0°≤θ<90°) between the line between the vertices of the background reference object and the line between the vertex projection points of the background reference object can be obtained.

[0085] The background reference object spacing may be 1m to 1000m, and the background reference object length may be 0.3m to 100m, or greater; the edge points at both ends of the background reference object are called vertices of the background reference object; the positioning background reference line segment is the line 101, 102 between the projection points of the background reference object vertices, and the length is L. The length L of the positioning background reference line segment is calculated by measuring the length L′ of the line connecting the background reference object vertices and the known angle θ and projecting it vertically to the positioning detection plane. The calculation formula is:

[0086] L = L'cos(θ).

[0087] Among them, the positioning projection value (where * is the visual sensor 1-4, n=1 is the upper projection number, and n=2 is the lower projection number) is calculated as follows:

[0088] Take the lens focus of the visual sensor as the projection center, project the center of the image of the object to be located onto the background reference, measure the length of the top vertex of the projected image on the background reference from the top vertex of the reference background reference, and the length of the bottom vertex of the projected image on the background reference from the bottom vertex of the reference background reference, and get the center projection length Through the angle θ between the known background reference object vertex connection line and the background reference object vertex projection point connection line on the positioning detection plane, it is vertically projected to the positioning detection plane, thereby obtaining the positioning projection value of the object to be positioned on the positioning detection plane. The calculation method is:

[0089]

[0090] The schematic diagram of its visual positioning model is shown in the attached Figure 2 As shown, the positioning result calculated at the same time is also located on the positioning detection plane. The following embodiments have been converted to the positioning detection plane and will not be specifically explained again.

[0091] Among them, the visual sensor can obtain a clear central projection image of the object to be located on the background reference object; among them, the visual sensor's viewing angle can obtain a clear image of the complete background reference object when there is no object to be located, and at the same time can distinguish the object to be located with the minimum size requirement within the background reference object spacing.

[0092] See also Figure 2 The vertical projections of the four visual sensor lens focal points on the positioning detection plane are respectively located at the four vertices of the positioning background reference line segments on both sides, such as points f1, f2, f3, and f4; its characteristic is that the monitoring area within the rectangle connecting the four points of the vertical projections of the four visual sensor lens focal points on the positioning detection plane is a non-positioning blind area.

[0093] Furthermore, the focal points of the four vision sensor lenses may be on the positioning detection plane, or may not be on the same plane or may be partially on the same plane.

[0094] See also Figure 3-4 The two background references are straight line background references, or curved line background references, or curved+straight background references. Of course, the background reference is not limited to a strip, and the background reference can be a background reference of various shapes according to the actual scene.

[0095] Furthermore, the background reference object is a plate-shaped marker or a self-luminous LED light board with a sharp contrast in color with the surrounding environment, and its characteristic is that it allows the visual sensor to obtain a clear central projection image of the object to be located on the background reference object, such as one or more of a luminous marker, a non-luminous marker, and a color-changing marker with a sharp contrast with the object to be located and the surrounding environment.

[0096] See also Figure 5, a standard single-sided monitoring system based on machine vision can be used, the lens focal points of the visual sensors 201 and 202 are respectively parallel to the upper vertex and the lower vertex of the background reference line segment 101, and the camera is positioned towards the background reference line segment 101. The visual positioning model of the system can realize object monitoring in approximately 75% of the area and object positioning in 25% of the area.

[0097] Furthermore, the positioning method of this embodiment can freely use multiple sets of standard machine vision-based single-side monitoring systems for combination according to actual conditions, but the monitoring range and positioning range will change accordingly, resulting in blind spots. It is generally recommended to use two sets of combined beam-type monitoring systems to achieve long-range object monitoring and object positioning without blind spots.

[0098] Specifically, the complete object positioning process of this embodiment includes the following steps:

[0099] (1) According to actual needs, set up two background reference objects and four visual sensors to establish a positioning detection plane.

[0100] (2) The image information of the detection area is collected through the visual sensor, and the image of the background reference object is captured to determine the position of the background reference object in the image as the area of ​​interest; the situation in the area of ​​interest is confirmed, and the image of the background reference object when there is no object to be located is obtained, and the background reference line segment is vertically projected onto the positioning detection plane to measure its length and the distance information between them.

[0101] (3) Detect the change value between the real-time image in the area of ​​interest in the current image and the complete image of the background reference object when it is not blocked. If the change value is greater than the set threshold, it is determined that there is an object to be located in the detection area, and enter step 4).

[0102] (4) Synchronously obtain four real-time images of the area of ​​interest taken by four visual sensors, where the center of the object to be located is projected onto the background reference object; analyze the four real-time images and the corresponding complete images of the background reference object respectively; in the image where the background reference object is blocked by the object to be located, measure the upper remaining length and the lower remaining length of the background reference object in the image after being blocked by the object to be located, and then vertically project them onto the positioning detection plane to obtain the positioning projection value. If the upper vertex or the lower vertex of the background reference object is blocked, the corresponding remaining length measurement result is zero. If the measurement result is not zero, it means that the positioning information equation containing the positioning projection value exists.

[0103] (5) According to the principle of similar triangles, a visual positioning model is constructed based on the positioning projection value of the object to be positioned, the length of the positioning background reference line segment, and the distance between two positioning background reference line segments to obtain the positioning information equation.

[0104] (6) According to the measured L*n , determine the existence of the positioning information equation, infer the position of the object to be positioned on the positioning detection plane, combine the corresponding positioning information equations for solution, and obtain one or more of the position information I1 and H, and the size information X and Y of the object to be positioned on the positioning detection plane.

[0105] Example 1:

[0106] This example 1 mainly provides four unknown parameters that can realize object positioning and size estimation: I1 、 Details of the H, X, and Y positioning information equations.

[0107] The visual positioning model is established as described above, see Figure 2 , all parameters are converted to the positioning detection plane according to the above method, and no further description is given. In this embodiment, in order to simplify the unknown quantity, the object to be positioned calculated by the following method is simplified to a rectangle parallel to the positioning background reference line segment on the positioning detection plane. According to the visual positioning model established above, each visual sensor can obtain up to two positioning information equations when capturing an object. The specific formula is as follows:

[0108] Visual sensor 201, visual positioning model such as Figure 6 As shown, we have:

[0109]

[0110]

[0111] Visual sensor 202, visual positioning model such as Figure 7 As shown, we have:

[0112]

[0113]

[0114] Visual sensor 203, visual positioning model such as Figure 8 As shown, we have:

[0115]

[0116]

[0117] Visual sensor 204, visual positioning model such as Fig. 9 As shown, we have:

[0118]

[0119]

[0120] In the above formula, L *1 The equation is defined as the upper triangle positioning information equation, including L *2 The equation is defined as the lower triangle positioning information equation. The unknown parameters are: I1, H, X, Y. The known parameters are: L, I, and L *n It can be determined by measuring the image and then converting it to the positioning detection plane. The method has been described above and will not be repeated here.

[0121] The method of object positioning and size estimation is to measure the positioning projection value L of the four images *n , determine the existing positioning information equations, and combine the corresponding equations to solve them.

[0122] Example 2:

[0123] This example 2 mainly gives Figure 1 The solution formula for object positioning and size estimation in .

[0124] See also Figure 1 , the object to be positioned appears in the detection area, only the positioning background reference line segment photographed by the positioning vision sensors 201 and 202 is blocked, and the two ends of the positioning background reference line segment are not blocked. For the background reference object blocking situation, see Fig.11 , measure four images simultaneously to obtain the positioning projection value L *n . Due to L 11 , L 12 , L 21 , L *22 If the measurement result is not zero, there are four corresponding positioning information equations, namely formula (1), (2), (3), and (4). Solving the equations together, we get the solution formula:

[0125]

[0126]

[0127]

[0128]

[0129] According to the solution formula, the position information I1 and H of the object simplified into a rectangle and the object size information X and Y can be calculated.

[0130] Example 3:

[0131] This Example 3 mainly provides a method for solving the position information of the upper or lower vertex of a positioning object that can be calculated on a positioning detection plane, including details of the positioning information equation for unknown parameters: I1, H, (H+Y).

[0132] In Example 1, the positioning information equation contains 4 unknown parameters, and 4 positioning information equations are required to solve it. There are some inapplicable situations. For example, see Fig.12 At this time, the system can only obtain two positioning information equations and cannot be solved. It is necessary to further simplify the positioning object to be calculated.

[0133] Since in its application scenario, more attention is paid to the positioning of the object, the parameters of the size of the positioning object can be simplified. Considering that in the monitoring of super-long detection areas, the size of the object to be positioned is usually much smaller than the length of the detection area. In most cases, the object is a few centimeters to a few meters. Compared with the length of the detection area, ignoring the object length X, the positioning accuracy of the system will not be significantly reduced. When solving the coordinates of the upper vertex position of the positioning object, Y and H can be combined as an unknown parameter. The unknown parameters are I1 and (H+Y), and when solving the coordinates of the lower vertex position of the positioning object, the formula only contains I1 and H, both of which contain only 2 unknown parameters. Since this embodiment performs dual redundant detection of the detection area through four visual sensors, when the visual sensor captures the object and does not completely block all the positioning background reference line segments, there must be another camera visual sensor that also captures the object, and they will have the upper triangle positioning information equation at the same time or the lower triangle positioning information equation at the same time, which can be combined for solution. By simplifying the unknown quantities, the positioning algorithm calculation is easier to implement and the accuracy will not be significantly reduced. At this time, only two information equations are needed to solve the problem, and the scope of application is more comprehensive.

[0134] The visual positioning model is established as described above, see Figure 2 In this embodiment, in order to simplify the unknown quantity, the object calculated by the following method is simplified to a line segment parallel to the background reference line segment, in which one case is as follows Fig.10 According to the visual positioning model established above, the method for obtaining the positioning information equation is the same as in Example 1, and will not be repeated here. The following directly gives the formula for obtaining two positioning information equations when each visual sensor captures an object:

[0135] Vision Sensor 201:

[0136]

[0137]

[0138] Vision Sensor 202:

[0139]

[0140]

[0141] Vision Sensor 203:

[0142]

[0143]

[0144] Vision Sensor 204:

[0145]

[0146]

[0147] The method of object positioning is to measure the positioning projection value L of the four images *n , determine the existing positioning information equations, and combine the corresponding upper triangle or lower triangle positioning information equations to solve them.

[0148] Example 4:

[0149] This example 4 mainly gives Fig.10 The solution formula for object positioning in .

[0150] See also Fig.10 , the object to be positioned appears in the detection area, only the positioning background reference line segment photographed by the positioning vision sensors 201 and 202 is blocked, and the two ends of the positioning background reference line segment are not blocked. For the background reference object blocking situation, see Fig.11 , measure four images simultaneously to obtain the positioning projection value L *n , due to L 11 , L 12 , L 21 , L *22 If the measurement result is not zero, there are four corresponding positioning information equations, namely formulas (13), (14), (15), and (16).

[0151] When solving the position information of the upper vertex, the upper triangle positioning information equations (13) and (15) of the combined visual sensors 201 and 202 are solved to obtain the solution formula:

[0152]

[0153]

[0154] When solving the position information of the lower vertex, the lower triangle positioning information equations (14) and (16) of the combined visual sensors 201 and 202 are solved to obtain the solution formula:

[0155]

[0156]

[0157] According to the existence of the positioning information equation, the corresponding upper triangular equation or lower triangular equation can be called for solution, and the upper vertex position information I1 and (H+Y) or the lower vertex position information I1 and H of the object simplified as a line segment can be calculated to achieve the positioning of the object.

[0158] The above example is only one of the cases, which is demonstrated as an example. Other cases can be calculated by referring to the above example, which is not listed one by one. The key point of the present invention is the idea of ​​obtaining the position and size information of the object by solving the combination of positioning information equations in the visual positioning model of the present invention, which should be protected.

[0159] The above is a preferred embodiment of the present invention, but the embodiment of the present invention is not limited to the above content. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A high-precision visual positioning method without blind spots in an ultra-long continuous large interval, characterized in that: The following steps are involved: Two background reference objects are placed in the detection area, and visual sensors are set at the upper and lower vertices of the two background reference objects; the visual sensors on the two background reference objects are respectively set relatively; the center projection value of the object without the object and the center projection value of the object to be located on the two background reference objects are measured in two cases of whether there is an object to be located; the focus of the visual sensor lens, the center projection value of the object without the object and the center projection value of the object to be located are vertically projected onto the positioning detection plane to obtain the focus of the positioning visual sensor, the positioning background reference line segment, the positioning projection value, the length of the positioning background reference line segment and the distance between the two positioning background reference line segments; The formation of the positioning detection plane needs to meet the following conditions: At least one vertex of each of the two background reference objects is located on the positioning detection plane; the lines connecting the four vertex projection points of the vertical projections of the four vertices of the two background reference objects on the positioning detection plane form a rectangle; the lens focus of the visual sensor is vertically projected on the positioning detection plane, and the lens focus projection point is located on the straight line connecting the two vertex projection points of the vertical projection of the background reference object on the positioning detection plane, and the line connecting the focus of the positioning visual sensor forms a rectangle; the angle θ between the line connecting the vertices of the background reference object and the line connecting the vertical projection points of the vertices of the background reference object on the positioning detection plane can be obtained, 0°≤θ<90°; The visual positioning model is obtained according to the principle of similar triangles, and the positioning information equation is obtained according to the visual positioning model. One or more of the position information I1 and H, and the size information X and Y of the object to be positioned on the positioning detection plane can be solved. The specific operation steps are as follows: According to actual needs, two background reference objects and four visual sensors are set up to establish a positioning detection plane; The image information of the detection area is collected by the visual sensor, and the image of the background reference object is captured to determine the position of the background reference object in the image as the area of ​​interest; the situation in the area of ​​interest is confirmed, and the image of the background reference object when there is no object to be positioned and the positioning background reference line segment vertically projected onto the positioning detection plane are obtained, and their length and the distance information between them are measured; Detect the change value between the real-time image in the area of ​​interest in the current image and the complete image of the background reference object when it is not blocked. If the change value is greater than the set threshold, it is determined that there is an object to be located in the detection area, and proceed to the next step; Synchronously obtain four real-time images of the area of ​​interest taken by four visual sensors, in which the center of the object to be located is projected onto the background reference object; analyze the four real-time images and the corresponding complete images of the background reference object respectively; in the image in which the background reference object is blocked by the object to be located, measure the upper remaining length and the lower remaining length of the background reference object in the image after being blocked by the object to be located, and then vertically project them onto the positioning detection plane to obtain the positioning projection value; According to the principle of similar triangles, a visual positioning model is constructed based on the positioning projection value of the object to be positioned, the length of the positioning background reference line segment and the distance between two positioning background reference line segments to obtain a positioning information equation. Based on the existence of the positioning information equation, the combined equation is solved to obtain one or more of the position information I1 and H and the size information X and Y of the object to be positioned on the positioning detection plane.

2. According to claim 1, a high-precision visual positioning method without blind spots in a super-long continuous large interval, characterized in that: The ultra-long continuous large section is a continuous section consisting of two positioning background reference line segments with a spacing H of 1m-1000m and a height L of 0.3m-100m.

3. The high-precision visual positioning method without blind spots in an ultra-long continuous large interval according to claim 1 is characterized in that: The background reference object is a straight or curved object or a straight and curved object.

4. The high-precision visual positioning method without blind spots in an ultra-long continuous large interval according to claim 1 is characterized in that: The background reference object is one or more of a luminous marker, a non-luminous marker, and a color-changing marker that allows the visual sensor to obtain a clear central projection image of the object to be located on the background reference object.

5. The high-precision visual positioning method without blind spots in an ultra-long continuous large interval according to claim 1 is characterized in that: The positioning projection value L *n The calculation method is: Taking the lens focus of the visual sensor as the projection center, project the image center of the object to be positioned onto the background reference object, measure the length of the uppermost vertex of the projection image on the background reference object from the upper vertex of the background reference object, and the length of the lowermost vertex of the projection image on the background reference object from the lower vertex of the background reference object, and project it vertically onto the positioning detection plane through the angle θ between the connecting line of the background reference vertices and the connecting line of the vertical projection points of the background reference vertices on the positioning detection plane, so as to obtain the positioning projection value L of the object to be positioned on the positioning detection plane *n .

6. The high-precision visual positioning method without blind spots in an ultra-long continuous large interval according to claim 1 is characterized in that: The shape of the object to be located on the positioning detection plane is regarded as a rectangle parallel to the positioning background reference line segment. When each visual sensor captures the object to be located and the background reference object is not completely blocked, one or two positioning information equations are obtained by the similar triangle theorem according to the spatial relationship between the object to be located, the background reference object and the visual sensor. From top to bottom, the visual sensors on the two background reference objects are numbered 1-2 and 3-4; The two positioning information equations of vision sensor No. 1 are: The two positioning information equations of vision sensor No. 2 are: The two positioning information equations of vision sensor No. 3 are: The two positioning information equations of the No. 4 visual sensor are: In the above formula, L *n is the positioning projection value, where * is the number 1-4 of the visual sensor of the image source, n=1 is the remaining length of the upper end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor, and n=2 is the remaining length of the lower end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor; I1 is the shortest distance between the vertical projection point of the upper vertex and the vertical projection point of the lower vertex of the background reference object corresponding to the focus of the current positioning visual sensor calculated on the positioning detection plane; H is the shortest distance between the lowest point of the positioning object and the line connecting the lower vertex projection points of the two positioning background reference objects calculated on the positioning detection plane; X is defined as the calculated length of the positioning object on the positioning detection plane; Y is defined as the calculated height of the positioning object on the positioning detection plane; L is the length of the positioning background reference line segment, that is, the length of the line connecting the upper vertex vertical projection point and the lower vertex vertical projection point of the background reference object on the positioning detection plane; I is the distance between the two positioning background reference line segments, that is, the length of the line connecting the vertical projection points of the corresponding vertices of the background reference objects on both sides on the positioning detection plane.

7. The high-precision visual positioning method without blind spots in an ultra-long continuous large interval according to claim 1 is characterized in that: The calculated shape of the positioning object on the positioning detection plane is regarded as a line segment parallel to the positioning background reference line segment. When each visual sensor captures the object to be positioned and the background reference object is not completely blocked, one or two positioning information equations can be obtained by the similar triangle theorem based on the spatial relationship between the object to be positioned, the background reference object and the visual sensor. From top to bottom, the visual sensors on the two background reference objects are numbered 1-2 and 3-4; The two positioning information equations of vision sensor No. 1 are: The two positioning information equations of vision sensor No. 2 are: The two positioning information equations of vision sensor No. 3 are: The two positioning information equations of the No. 4 visual sensor are: In the above formula, L *n is the positioning projection value, where * is the number 1-4 of the visual sensor of the image source, n=1 is the remaining length of the upper end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor, and n=2 is the remaining length of the lower end of the positioning background reference line segment projected by the object to be positioned relative to the current visual sensor; I1 is the shortest distance between the vertical projection point of the upper vertex and the vertical projection point of the lower vertex of the background reference object corresponding to the focus of the current positioning visual sensor calculated on the positioning detection plane; H is the shortest distance between the lowest point of the positioning object and the line connecting the lower vertex projection points of the two positioning background reference objects calculated on the positioning detection plane; Y is defined as the calculated height of the positioning object on the positioning detection plane; L is the length of the positioning background reference line segment, that is, the length of the line connecting the upper vertex vertical projection point and the lower vertex vertical projection point of the background reference object on the positioning detection plane; I is the distance between the two positioning background reference line segments, that is, the length of the line connecting the vertical projection points of the corresponding vertices of the background reference objects on both sides on the positioning detection plane.

Citation Information

Patent Citations

  • Initiative-view-angle binocular vision ranging system and initiative-view-angle binocular vision ranging method

    CN105651258A

  • Laser bijection invasive alarming and positioning sensing system

    CN203825755U

  • Comprehensive perimeter security and protection system with infrared thermal image verification

    CN211928699U

  • Method and apparatus for monitoring a spatial volume and a calibration method

    CN101601048A

  • Projection interaction method based on pure machine vision positioning

    CN111354007A