A fast scene position updating method based on sequential UAV images

Abstract

The application discloses a kind of based on sequence unmanned plane image fast scene position update method, belong to unmanned plane positioning navigation field. With the text of shop sign in image and the POI name in electronic map match and carry out scene positioning, and with the matching range of unmanned plane flight distance limit POI, improve the matching precision and efficiency;In the case where GNSS is refused in the long endurance of unmanned plane, fast scene update can be realized, so as to avoid the positioning error accumulation of inertial navigation system.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of unmanned aerial vehicle positioning and navigation, and particularly relates to a fast scene position updating method based on sequence unmanned aerial vehicle images. BACKGROUND

[0002] Under the condition of ultra-low altitude flight of unmanned aerial vehicles in cities, the visual positioning method based on image matching and the autonomous positioning of unmanned aerial vehicles based on open source maps will both fail due to scale differences, and the SLAM and inertial navigation system will also accumulate errors over time. POI generally refers to point data in an Internet electronic map, which basically includes four attributes of name, address, coordinate and category. The store signs widely distributed in cities correspond to POI data, so POI is a good data set that can be used for autonomous positioning of unmanned aerial vehicles. However, POI data has problems such as update lag (such as the POI data of some areas at present is collected in 2016), inconsistency with sign text (such as the sign text is "general traffic school", but the name in the POI data is "Hebei general school for special training"), and a large number of similar or identical POI names, which directly affect the text matching results of unmanned aerial vehicle images and POI library. At the same time, when facing a large area of predetermined scene, the POI data reaches tens of thousands or even hundreds of thousands, which will lead to very low matching efficiency and does not meet the practical requirements.

[0003] In order to realize the scene position positioning of sequence unmanned aerial vehicle images in a wide range of urban scenes, a fast and reliable unmanned aerial vehicle image scene position updating method needs to be researched. SUMMARY

[0004] The purpose of the present application is to solve the problem of inaccuracy and low efficiency when using POI for unmanned aerial vehicle image scene positioning, and a fast scene position updating method based on sequence unmanned aerial vehicle images is proposed, which realizes the fast scene position updating of sequence unmanned aerial vehicle images through POI data spatial retrieval and fuzzy text matching under the support of unmanned aerial vehicle flight parameters.

[0005] In order to achieve the above purpose, the technical scheme adopted by the present application is:

[0006] A fast scene position updating method based on sequence unmanned aerial vehicle images, comprising the following steps:

[0007] S1 Assuming that the position of the i-th frame unmanned aerial vehicle image is known, the coordinates (xi, yi) are set as the latest scene position, and then the text information recognition is performed in the subsequent unmanned aerial vehicle images from the i+1-th frame;

[0008] S2If the jth frame can recognize the text, first, the recognized text is preprocessed, and the text with a length less than 3 characters is deleted; wherein, j≥i+1;

[0009] S3With (xi, yi) as the center and V×(j-i) as the radius, spatial retrieval is performed from the POI library, and the retrieval result is P; wherein, V is the maximum flight speed;

[0010] S4The preprocessed text information in the jth frame and the name in P are subjected to text fuzzy matching using the TF-IDF method, and if the similarity of the text fuzzy matching is greater than a threshold value, the matching is successful;

[0011] S5If there is a POI with successful matching in step S4, the coordinates (xj, yj) of the POI are set as the latest scene position, j is set as i, and then step S1 is returned; if there is no POI with successful matching in step S4, j=j+1, and then step S2 is returned.

[0012] Compared with the background art, the present application has the following advantages:

[0013] The present application provides a kind of quick scene position updating method based on sequence unmanned aerial vehicle image, and the scene position updating of unmanned aerial vehicle image in real time is realized by identifying the label text on unmanned aerial vehicle image.

[0014] The present application provides a kind of quick scene position updating method based on sequence unmanned aerial vehicle image, and the range of POI matched is limited by using the flight distance of unmanned aerial vehicle, which can quickly narrow the retrieval space and improve the matching efficiency and accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is the principle framework diagram of the quick scene position updating method based on sequence unmanned aerial vehicle image of the present application.

[0016] Figure 2 It is the retrieval result of the POI retrieval of the quick scene position updating method based on sequence unmanned aerial vehicle image of the present application with the ith image scene position as the center and V×(j-i) as the radius. DETAILED DESCRIPTION

[0017] The specific embodiments of the present application are described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present application. It should be particularly noted that in the following description, when the detailed description of known functions and designs may dilute the main content of the present application, these descriptions will be omitted here.

[0018] Figure 1This is a schematic diagram illustrating the principle framework of a fast scene location update method based on sequential UAV images according to the present invention. Specifically, the method includes the following steps:

[0019] S1 assumes that the position of the drone image in the i-th frame is known, sets its coordinates (xi, yi) to the latest scene position, and then performs text information recognition from the subsequent drone images starting from the (i+1)-th frame.

[0020] S2 If text can be recognized in the j-th frame (j≥i+1), the recognized text is first preprocessed: text shorter than 3 characters is deleted to exclude cases where the drone image does not completely cover the sign.

[0021] S3 uses (xi, yi) as the center and V×(ji) as the radius (V is the maximum flight speed) as the radius to perform a spatial search in the POI database, and the search result is P;

[0022] S4 performs fuzzy text matching between the preprocessed text information in frame j and the name in P using the TF-IDF method, setting the cosine similarity threshold to 0.75 (i.e., if the similarity of the fuzzy text matching is greater than 0.75, then the match is considered successful).

[0023] S5 If a POI is successfully matched in step S4, set the coordinates (xj, yj) of the POI to the latest scene position, set j to i, and then return to step S1; if no POI is successfully matched in step S4, then j = j + 1, and then return to step S2.

[0024] Below is a more specific example:

[0025] (1) The scene location of the first frame image of the UAV is (114.438480E, 38.0266602N).

[0026] (2) If the 10th frame image of the drone is recognized by the PPOCRv3 text recognition model, the following 5 shop signs can be identified: “MR Piaoyi Hair Salon”, “Zhige System Doors and Windows”, “21 21 Century Real Estate”, “Foot Repair Shop”, and “Nongfu Spring”.

[0027] (3) Using the scene location (114.438480E, 38.0266602N) of the first frame image as the center point, with a radius of 36 meters (the maximum flight speed of the UAV is 9m / s), a spatial search is performed in the POI database. The search results are as follows: Figure 2 As shown;

[0028] (4) The weight in the TF-IDF model is constructed according to the frequency of the characters in the POI name in the whole city, then the shop sign recognized in the 10th frame image is subjected to fuzzy text matching with the name in the POI searched in step (3);

[0029] (5) The final matching similarity of "CIGS system door and window" is greater than 0.75, the matching is successful, and the scene position of the 10th frame image is updated to (114.438738E, 38.026775N).

[0030] The present application realizes the matching of the text of the shop sign in the image and the POI name in the electronic map and the scene positioning, and limits the matching range of the POI by the flight distance of the unmanned aerial vehicle, improves the matching precision and efficiency, and can realize the rapid scene updating in the case of GNSS denial of the unmanned aerial vehicle with long endurance, thereby avoiding the accumulation of the positioning error of the inertial navigation system.

[0031] Although the above describes the specific embodiments of the present application in order to facilitate the understanding of the present application by the person skilled in the art, it should be clear that the present application is not limited to the scope of the specific embodiments, and for the person skilled in the art, all the changes which are obvious and within the spirit and scope of the present application defined and determined by the appended claims are included in the protection of the present application.

Claims

1. A fast scene position update method based on sequential UAV images, characterized in that, The method comprises the following steps: S1 assuming that the position of the i-th frame of UAV image is known, setting coordinates (xi, yi) as the latest scene position, and then performing text information recognition on subsequent UAV images from the i+1-th frame; S2 if the j-th frame can recognize text, firstly, pre-processing the recognized text, and deleting the text with length shorter than 3 characters; wherein, j≥i+1; S3 taking (xi, yi) as the center and V×(j-i) as the radius, performing spatial retrieval from the POI library, and the retrieval result is P; wherein, V is the maximum flight speed; S4 using the TF-IDF method to perform text fuzzy matching on the pre-processed text information in the j-th frame and the name in P, if the similarity of the text fuzzy matching is greater than a threshold value, the matching is successful; S5 if there is a POI matched successfully in step S4, setting the coordinates (xj, yj) of the POI as the latest scene position, setting j as i, and then returning to step S1; if there is no POI matched successfully in step S4, setting j=j+1, and then returning to step S2.

Images