CN111157007A - Indoor positioning method using cross vision - Google Patents

Abstract

The invention relates to the technical field of indoor object positioning, in particular to an indoor positioning method by utilizing cross vision; the method takes a cross overlapping area formed by two cameras as a positioning range, utilizes a depth learning algorithm to carry out target detection and key point positioning, then utilizes a machine learning method to match targets in the two cameras, and finally completes multi-target positioning in a visual area according to a target matching result and the distance between the two cameras; s1, deploying the system; s2, measuring coordinates; s3, detecting a Yolo-v3 target, firstly matching the targets if the targets are multiple, then positioning the Landmark, and directly positioning the Landmark if the targets are not multiple; s4, calculating the distance between the target object and each camera; and S5, completing the positioning of the target object by calculation.

Description

Indoor positioning method using cross vision

Technical Field

The invention relates to the technical field of indoor object positioning, in particular to an indoor positioning method by utilizing cross vision.

Background

As is well known, with the increase of data services and multimedia services, people have increasingly increased demands for positioning and navigation, and especially in complex indoor environments, such as frequent halls, exhibition halls, supermarkets, libraries, underground parking lots, and the like, the positioning system cannot accurately position the position of an object due to sudden signal reduction or signal interference caused by the traffic of pedestrians, the blocking of walls or stone columns and other large objects.

The currently common indoor wireless positioning technologies include an indoor GPS positioning technology, an infrared indoor positioning technology, an ultrasonic positioning technology, a Bluetooth technology, a Wi-Fi technology and the like, and have the following defects:

1. the indoor GPS positioning technology uses a GPS receiver to position a target location, but signals are greatly attenuated by the influence of buildings, positioning accuracy is low, and the cost of a locator terminal is high.

2. The infrared indoor positioning technology utilizes an infrared IR mark to emit modulated infrared rays, and the modulated infrared rays are received by an optical sensor installed indoors for positioning. However, infrared rays cannot penetrate through barriers, so that infrared rays can only be transmitted at a sight distance, and the straight sight distance and the transmission distance are short, so that the infrared ray transmission device is easily interfered by fluorescent lamps or light in rooms, and cannot be suitable for the situation of complicated people flow in airport halls.

3. The ultrasonic positioning technology mainly adopts a reflection type distance measurement method, and calculates the distance to be measured by using the time difference between an echo and a transmitted wave according to the transmission of ultrasonic waves and the reception of the echo generated by a measured object. However, the method cannot be applied to the situation of complicated people flow in an airport hall, and the positioning technology needs a large amount of investment in bottom hardware facilities, so that the cost is too high.

4. The Bluetooth technology carries out positioning by measuring signal intensity, and is a short-distance low-power-consumption wireless transmission technology. But the technology is mainly applied to small-range positioning, and the airport lounge can have interference of a plurality of Bluetooth signals, and the technology is not suitable for the complex environment of the airport lounge.

Wi-Fi technology enables positioning, monitoring and tracking tasks in complex environments to be achieved through a Wireless Local Area Network (WLAN) formed by wireless access points. However, the technology is easily interfered by other signals, so that the precision of the technology is influenced, and the energy consumption of the positioner is high.

Disclosure of Invention

In order to solve the technical problems, the invention provides an indoor positioning method utilizing cross vision, which takes a cross overlapping area formed by imaging of two cameras as a positioning range, utilizes a depth learning algorithm to carry out target detection and key point positioning, then utilizes a machine learning method to match targets in the two cameras, and finally completes multi-target positioning in a visual area according to a target matching result and the distance between the two cameras.

The invention relates to an indoor positioning method by using cross vision, which comprises the following steps:

s1, deploying the system;

s2, measuring coordinates;

s3, detecting a Yolo-v3 target, firstly matching the targets if the targets are multiple, then positioning the Landmark, and directly positioning the Landmark if the targets are not multiple;

s4, calculating the distance between the target object and each camera;

and S5, completing the positioning of the target object by calculation.

The invention relates to an indoor positioning method by utilizing cross vision, which is based on a positioning system, and the positioning system comprises

Two high-definition cameras;

the spacing distance between the two cameras, the installation height and the angle between the cameras and the ground visual angle are known;

and one computer host is used for target detection, matching and positioning.

Compared with the prior art, the invention has the beneficial effects that:

1) indoor positioning is carried out by using a visual method, the equipment is simple, and the cost is low;

2) the cameras of the indoor positioning method are arranged at high positions, only two cameras are needed, the indoor environment does not need to be changed when the cameras are installed and deployed, any equipment does not need to be additionally arranged on the target needing to be positioned, and the indoor positioning method has the characteristics of flexible and convenient deployment and multiple environment adaptation;

3) the target in the visual region is detected by adopting a deep learning method, so that the interference of indoor pedestrians, seats, machine equipment and the like can be effectively removed, the robustness is strong, and the method is suitable for complex environments;

4) by the target detection and matching method, a plurality of targets can be positioned at the same time, the multi-target positioning function is realized, and the method has the characteristic of strong practicability;

5) the indoor positioning method of the invention utilizes the vision cross area of the two cameras to carry out multi-target indoor positioning, and utilizes deep learning to detect, position and match targets, thus being capable of positioning a plurality of targets and various targets at the same time, and having small positioning error and high precision.

Drawings

FIG. 1 is an indoor positioning deployment scenario of the present invention;

FIG. 2 is a schematic diagram of imaging of two high definition cameras;

fig. 3 is a flow chart of the indoor positioning of the present invention.

Detailed Description

The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.

The airport hall is taken as a case scene, the baggage carts in the hall are taken as indoor positioning targets for illustration, the baggage carts are taken as resources in the airport, and if the position information of the baggage carts can be accurately provided, the boarding/departing time of passengers can be saved, and the baggage cart resources can be reasonably and effectively managed and utilized by airport departments; meanwhile, the airport hall is used as an indoor environment and has the characteristics of large pedestrian flow, complex environment, large wireless signal interference and the like. In conclusion, the baggage pushcart in the airport hall is taken as a research object of the indoor positioning method, so that the complex indoor environment requirement is met, the pain point in the industry can be solved, the travel of passengers is facilitated, and the resource utilization rate is improved.

S1 deployment scheme of indoor positioning of the invention is shown in figure 1, and the positions (X) of two cameras are known₁,Y₁,Z₁)、(X₂,Y₂,Z₂) The coordinates of the baggage cart are (X, Y, Z), the following equation is given:

the airport hall is used as a horizontal plane, and the height coordinate system Z takes the ground as a reference coordinate system, then the height values of the luggage barrow are all 0, so the three-dimensional coordinate relation can be simplified into a two-dimensional relation, and the equation is as follows:

the coordinates (X, Y) of the luggage trolley can be obtained by two unknowns of the two equations;

s2, taking the image as a two-dimensional plane, and when the two cameras have the same parameters and the coordinate positions are known, as shown in fig. 2, the coordinates of the baggage cart in the two pictures are (x ', y') and (x ", y"), respectively, then the pixel distance r of the baggage cart can be solved₁And r₂Knowing the coordinate positions of the two cameras, the actual distance corresponding to each pixel point of the two cameras can be solved as S₁And S₂Then equation (2) can be expressed as:

the coordinates (± X, ± Y) of the baggage cart may be solved and finally the coordinates (X, Y) of the baggage cart may be determined from the image coordinates (X ', Y') and (X ", Y");

s3, based on the positioning principle of the luggage cart, detecting the position of the luggage cart in the image by using a Yolo-v3 target detection model, and then determining the coordinates (x ') of the two rear wheels of the luggage cart by using a Landmark key point positioning model'₁,y′₁) And (x'₂,y′₂) The middle point of the two wheels is the coordinate (x ', y') of the baggage trolley on the camera 1, and the coordinate (x ', y') of the baggage trolley on the camera 2 is positioned in the same way;

s4, the image reference coordinate of the camera 1 is known as (x)₁，y₁) The actual distance corresponding to each pixel point is S₁(ii) a Camera 2 image reference coordinate is (x)₂,y₂) The actual distance corresponding to each pixel point is S2; then according to the equation:

s5, solving for R₁And R₂Substituting the coordinate world coordinates (X, Y) of the trolley into the formula (2) to obtain the coordinate world coordinates (X, Y) of the trolley, and completing indoor positioning of the single luggage trolley;

when a plurality of luggage trolleys appear in a detection area, the luggage trolleys in two cameras need to be matched in pairs, the matching method comprises the steps of respectively setting the object resize in the two cameras to be 100 x 100, respectively carrying out surf conversion on R, G, B channels of images after resize, respectively carrying out feature extraction on R, G, B channel images after surf conversion by using a kernel with the size of 5 x 5 and the step length of 5, finally extracting 1200 features from the images by each object, describing each object by 1203 features together with the pixel coordinates and the distance of the object in the images, combining the object ① in the camera 1 and all the objects in the camera 2 in pairs after the feature extraction is finished, comparing the similarity of each combination by using a random algorithm (RF), taking one group with the highest similarity as the same object, and analogizing to finish the matching of the same object in the two cameras, and finally, respectively carrying out steps S3 and S4 calculation on the luggage trolleys after the matching is finished, so that the indoor positioning of the plurality of luggage trolleys can be finished.

The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, several modifications and variations can be made without departing from the technical principle of the present invention, and these modifications and variations should also be regarded as the protection scope of the present invention.

Claims (2)

1. An indoor positioning method using cross-vision, comprising the steps of:

s1, deploying the system;

s2, measuring coordinates;

s3, detecting a Yolo-v3 target, firstly matching the targets if the targets are multiple, then positioning the Landmark, and directly positioning the Landmark if the targets are not multiple;

s4, calculating the distance between the target object and each camera;

and S5, completing the positioning of the target object by calculation.

2. An indoor positioning method using cross-vision as claimed in claim 1, characterized in that based on a positioning system, the positioning system includes

Two high-definition cameras;

the spacing distance between the two cameras, the installation height and the angle between the cameras and the ground visual angle are known;

and one computer host is used for target detection, matching and positioning.

Images