A contact network imaging method and system based on image intelligent recognition

Abstract

The present invention relates to the field of contact network imaging technology, and in particular, to a contact network imaging method and system based on image intelligent recognition, comprising the following steps: obtaining an original image set taken by a camera during movement; performing target detection on the original image set, screening out a target image set containing a target imaging area; obtaining a shooting depth of field, and screening out images of interest within the shooting depth of field from the target image set according to the shooting depth of field; and selecting an image closest to the target from the images of interest as an archived image. Compared with the prior art, the present invention does not rely on a trigger device, and through continuous shooting, reduces the space size of the hardware, the volume of the entire device is more compact, and the overall structure is more lightweight, which is more conducive to the deployment and application of the contact network imaging device.

Description

Technical Field

[0001] The present invention relates to the technical field of overhead line imaging, and in particular to an overhead line imaging method and system based on intelligent image recognition. Background Art

[0002] The overhead contact network is a key device that provides power for electric trains. Since the overhead contact network is laid in the open air along the railway track, it is prone to failures in a harsh working environment. In order to ensure driving safety, timely discovering hidden dangers of overhead contact network equipment through various detection methods has long been a difficult problem faced by the railway power supply system. In recent years, with the development of image processing and various optical sensors, overhead contact network imaging technology has gradually become a research hotspot. High-definition imaging technology can timely discover and deal with safety hazards of various components of the overhead contact network, providing a strong guarantee for driving safety.

[0003] At present, the contact network imaging method mainly calculates an optimal shooting working distance based on the selected camera, lens, and imaging field of view, and then triggers the camera to shoot through hardware triggering or software triggering; among them, the common hardware trigger is to trigger the visible light camera to shoot the contact network support device through the detection and positioning of the pillar by the laser sensor, so as to capture the image at the optimal distance, but there is a fixed distance between the laser sensor and the visible light camera (generally several meters), and there are requirements for the installation space. In addition, the hardware trigger is limited by the triggering accuracy of the laser sensor and cannot be applied to the imaging of important components such as contact suspension and segmented insulators. At the same time, due to the limitation of the installation space, it can only be applied to the installation deployment on the roof of the inspection vehicle, but it is difficult to implement because the inspection vehicle has problems such as shunting difficulties for track sidings, stations, and dedicated lines; the common software trigger refers to the use of software output trigger signals to control the camera to shoot images, which has the advantages of flexible implementation and multiple applicable scenarios, but its delay size is related to the algorithm complexity and the processing power of the computing unit, and it cannot meet the requirements for high-speed image acquisition systems.

[0004] Based on this, how to achieve contact network imaging with low latency and algorithm complexity is a technical problem that needs to be solved urgently. Summary of the invention

[0005] The purpose of the present invention is to provide a contact network imaging method and system based on image intelligent recognition, which is independent of a trigger device and can effectively solve the technical problem of how to achieve contact network imaging with low latency and algorithm complexity through continuous shooting by a camera.

[0006] The present invention is implemented by the following technical solution: A contact network imaging method based on image intelligent recognition comprises the following steps:

[0007] Get the original image set taken by the camera during movement;

[0008] Performing target detection on the original image set to filter out a target image set containing a target imaging area;

[0009] Acquire a shooting depth of field, and select an image of interest within a range of the shooting depth of field from a target image set according to the shooting depth of field;

[0010] An image closest to the target is selected from the images of interest as an archived image.

[0011] According to a preferred embodiment, the step of acquiring a set of original images taken by the camera during movement specifically includes:

[0012] A first original image set taken by a first camera during movement and a second original image set taken by a second camera during movement are acquired, wherein the first camera and the second camera have different shooting targets.

[0013] According to a preferred embodiment, the shooting target of the first camera is the contact network supporting device, and the shooting target of the second camera is the contact network suspension.

[0014] According to a preferred embodiment, the imaging resolution of the first camera is not less than 65 million, and the imaging resolution of the second camera is not less than 12 million.

[0015] According to a preferred embodiment, the first camera and the second camera are mounted on the same mobile carrier for mobile shooting, and the mobile carrier moves along a track.

[0016] According to a preferred embodiment, the first camera and the second camera are mounted on a mobile carrier arranged at a preset height and take photos at equal intervals during the process of moving.

[0017] According to a preferred embodiment, performing target detection on the original image set specifically includes:

[0018] Using one or more of the contact network locator positioning clamp and the positioning support as identification feature points, performing target detection on the first original image set, and screening out a first target image set including an imaging area of ​​the contact network support device;

[0019] Using one or more of the contact wire clamp, the suspension string, and the load-bearing cable clamp as feature recognition points, target detection is performed on the second original image set to screen out a second target image set containing the contact network suspension imaging area.

[0020] According to a preferred embodiment, before performing target detection on the original image set, the method further includes:

[0021] The images in the original image set are preprocessed, and the preprocessing includes image cropping and image rotation.

[0022] According to a preferred embodiment, selecting an image of interest within the range of the shooting depth of field from a target image set according to the shooting depth of field specifically includes:

[0023] A first image of interest within a first shooting depth of field is obtained by screening from a first target image set at a first shooting depth of field, wherein the first shooting depth of field is a first distance from the overhead line support device to the first camera in a horizontal direction, and the first distance is 6.5 to 7.5 meters;

[0024] With the second shooting depth of field, the first image of interest within the second shooting depth of field is screened from the second target image set. The second shooting depth of field is the second distance from the contact network to the second camera in the horizontal direction of the suspension, and the second distance is 6 to 7 meters.

[0025] The present invention also provides a contact network imaging system based on image intelligent recognition, which is applied to the imaging method as described above, and the system comprises:

[0026] An image acquisition module is used to acquire a set of original images taken by the camera during movement;

[0027] A target detection module, used to perform target detection on the original image set, and filter out a target image set containing a target imaging area;

[0028] A screening module, used to obtain a shooting depth of field, and to screen an image of interest within a shooting depth of field from a target image set according to the shooting depth of field;

[0029] The storage module is used to select an image closest to the target from the images of interest as an archived image.

[0030] The technical solution of a contact network imaging method and system based on image intelligent recognition provided by the present invention has at least the following advantages and beneficial effects: (1) Compared with the prior art, the present invention does not rely on a trigger device, and realizes low-latency imaging through continuous shooting, and reduces the space size of the hardware. The volume of the entire device is smaller and the overall structure is lighter, which is more conducive to the deployment and application of the contact network imaging device; (2) Continuous shooting is adopted, and then a target image set containing a target imaging area is obtained from the original image set obtained by shooting, and an image of interest is obtained from the target image set according to the shooting depth of field, and finally the image closest to the target is selected as the archived image, which reduces the algorithm complexity, improves the imaging quality, and makes the archived image more representative; (3) The supporting device adopts a camera of not less than 65 million, and the shooting distance in the horizontal direction is 6.5 to 7.5 meters. The suspension adopts a camera of not less than 12 million, and the shooting distance in the horizontal direction is 6 to 7 meters, which is conducive to ensuring the clarity of the archived image and improving the imaging quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A schematic diagram of a process flow of a contact network imaging method based on image intelligent recognition provided in Example 1 of the present invention;

[0032] Figure 2 A schematic diagram of an original image set provided in Embodiment 1 of the present invention;

[0033] Figure 3 A schematic diagram of a target image set provided by Embodiment 1 of the present invention;

[0034] Figure 4 The image of interest provided in Embodiment 1 of the present invention;

[0035] Figure 5 This is the archived image provided by Example 1 of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0037] Example 1

[0038] Figure 1 A schematic diagram of a method for imaging a contact network based on intelligent image recognition provided by an embodiment of the present invention. Figure 1 As shown, the contact network imaging method based on image intelligent recognition includes the following steps:

[0039] Step 1: Get the original image set taken by the camera during movement.

[0040] Specifically, in one implementation of this embodiment, the acquisition of the original image set taken by the camera during the movement process is as follows: Figure 2 As shown; specifically including: obtaining a first original image set taken by a first camera during movement and a second original image set taken by a second camera during movement, wherein the first camera and the second camera are mounted on the same mobile carrier for mobile shooting, and the mobile carrier moves along a track; in this embodiment, the first camera and the second camera are mounted on the mobile carrier for moving and shooting at equal intervals, the installation elevation angles of the first camera and the second camera remain unchanged, and shooting is performed with a shooting interval of no more than 0.5 meters, thereby ensuring that relevant images can be accurately acquired while reducing memory usage.

[0041] In addition, in order to meet the imaging requirements in dim light, the first camera and the second camera are also equipped with a light source. The IO outputs of the first camera and the second camera synchronously trigger the light source to achieve synchronization of light source and camera exposure to ensure that high image quality can be obtained even in dim light. Among them, the light source illumination field of view FOV l The deterministic expression is Where d represents the illumination distance of the light source, and θ represents the angle of the light source.

[0042] In this embodiment, the shooting targets of the first camera and the second camera are different. Taking the common contact network fault location as an example, the shooting target of the first camera is the contact network support device, and the shooting target of the second camera is the contact network suspension. In addition, the shooting target can also be tunnel hanging columns, contact network segmented insulating paint, contact network phase insulator, etc., which will not be elaborated here.

[0043] Furthermore, in order to meet the imaging requirements of the supporting device imaging height not less than 3.3 meters, the left and right imaging width not less than 3.5 meters, and the image pixels not less than 50 million, and the contact network suspension imaging height not less than 3.3 meters, the left and right imaging width not less than 2 meters, and the image pixels not less than 5 million, the first camera and the second camera of this embodiment are configured as follows: the imaging resolution of the first camera is not less than 65 million, and the imaging resolution of the second camera is not less than 12 million.

[0044] It should be noted that compared with the prior art, the present invention does not rely on a trigger device. Through continuous shooting, it achieves low-latency imaging and reduces the spatial dimensions of the hardware. The entire device is smaller in size and the overall structure is lighter, which is more conducive to the deployment and application of contact network imaging devices.

[0045] Step 2: performing target detection on the original image set to filter out a target image set containing a target imaging area.

[0046] In this embodiment, before performing target detection on the original image set, the images in the original image set are preprocessed, and the preprocessing includes image cropping, image rotation, etc., so as to improve the contact network target recognition accuracy and detection speed.

[0047] Specifically, in one implementation of this embodiment, a deep learning algorithm YOLO v5 is used to train a target detection model. By training a large amount of overhead line image data, the data set covers target images taken at different distances in the horizontal direction, target images taken at different angles constructed by image transformation, and target images taken under different line scenes, so as to ensure the accuracy of recognition and the adaptability of the model. In this embodiment, target detection is performed on the original image set, specifically including:

[0048] Using one or more of the contact network locator positioning clamp and the positioning support as identification feature points, target detection is performed on the first original image set to screen out the first target image set containing the imaging area of ​​the contact network support device, see Figure 3 As shown; using one or more of the contact wire clamp, the suspension string, and the load-bearing cable clamp as feature recognition points, target detection is performed on the second original image set to filter out the second target image set containing the contact network suspension imaging area.

[0049] Step 3: obtaining a shooting depth of field, and filtering out images of interest within a range of the shooting depth of field from a target image set according to the shooting depth of field.

[0050] Specifically, in one implementation of this embodiment, the image of interest within the shooting depth of field range is obtained by screening from the target image set according to the shooting depth of field, specifically including: using the first shooting depth of field, screening from the first target image set to obtain a first image of interest within the first shooting depth of field range, see Figure 4 As shown, the first shooting depth of field is the first distance from the contact network support device to the first camera in the horizontal direction, and the first distance is 6.5 to 7.5 meters; with the second shooting depth of field, the first image of interest within the second shooting depth of field is screened from the second target image set, and the second shooting depth of field is the second distance from the contact network suspension to the second camera in the horizontal direction, and the second distance is 6 to 7 meters.

[0051] It should be noted that the supporting device of the present invention adopts a camera of no less than 65 million, with a shooting distance of 6.5 to 7.5 meters in the horizontal direction, and the hanging device adopts a camera of no less than 12 million, with a shooting distance of 6 to 7 meters in the horizontal direction, which is conducive to ensuring the clarity of the archived image and improving the imaging quality.

[0052] It should be noted that the calculation expression for the depth of field is ΔL = 2f 2 FδL 2 / (f 4 -F 2 δ 2 L 2 ), where F is the lens aperture value, δ is the allowable circle of confusion diameter, L is the camera focus distance, and f is the lens focal length.

[0053] Step 4: Select an image closest to the target from the images of interest as an archived image.

[0054] It should be noted that when the two conditions of including the target imaging area in the image and the image being within the corresponding shooting depth of field are satisfied at the same time, the closer to the contact network target, the higher the clarity and resolution of the imaging. Therefore, in this embodiment, the image closest to the target is selected from the image of interest as the archived image. For example, for the contact network support device, the first image of interest corresponding to the first shooting depth of field of 6.5 meters is selected as the archived image. Figure 5 As shown, for the overhead line suspension, the second image of interest corresponding to the second shooting depth of field of 6 meters is selected as the archived image.

[0055] The present invention adopts continuous shooting, and then obtains a target image set containing a target imaging area from the original image set obtained by shooting, obtains an image of interest from the target image set according to the shooting depth of field, and finally selects the image closest to the target as the archived image, which reduces the algorithm complexity, improves the imaging quality, and makes the archived image more representative.

[0056] Example 2

[0057] This embodiment provides a contact network imaging system based on image intelligent recognition based on the technical solution provided in Example 1. The contact network imaging system based on image intelligent recognition is applied to the imaging method provided in Example 1. The system includes an image acquisition module, a target detection module, a screening module and a storage module; wherein the image acquisition module is used to acquire the original image set taken by the camera during movement, the target detection module is used to perform target detection on the original image set, and screen out a target image set containing a target imaging area, the screening module is used to acquire the shooting depth of field, and screen out images of interest within the shooting depth of field range from the target image set according to the shooting depth of field, and the storage module is used to select an image closest to the target from the image of interest as an archived image.

[0058] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A contact network imaging method based on image intelligent recognition, characterized in that: The steps include: Get the original image set taken by the camera during movement; Performing target detection on the original image set to filter out a target image set containing a target imaging area; Acquire a shooting depth of field, and select an image of interest within a range of the shooting depth of field from a target image set according to the shooting depth of field; An image closest to the target is selected from the images of interest as an archived image.

2. The contact network imaging method based on image intelligent recognition according to claim 1, characterized in that: The obtaining of the original image set taken by the camera during the movement specifically includes: A first original image set taken by a first camera during movement and a second original image set taken by a second camera during movement are acquired, wherein the first camera and the second camera have different shooting targets.

3. The contact network imaging method based on image intelligent recognition according to claim 2, characterized in that: The shooting target of the first camera is the contact network supporting device, and the shooting target of the second camera is the contact network suspension.

4. The contact network imaging method based on image intelligent recognition according to claim 3, characterized in that: The imaging resolution of the first camera is not less than 65 million, and the imaging resolution of the second camera is not less than 12 million.

5. The overhead line imaging method based on image intelligent recognition according to claim 4, characterized in that: The first camera and the second camera are mounted on the same mobile carrier for mobile shooting, and the mobile carrier moves along a track.

6. The contact network imaging method based on image intelligent recognition according to claim 5, characterized in that: The first camera and the second camera are mounted on a mobile carrier arranged at a preset height and take photos at equal intervals during the process of moving.

7. The contact network imaging method based on image intelligent recognition according to claim 6, characterized in that: Performing target detection on the original image set specifically includes: Using one or more of the contact network locator positioning clamp and the positioning support as identification feature points, performing target detection on the first original image set, and screening out a first target image set including an imaging area of ​​the contact network support device; Using one or more of the contact wire clamp, the suspension string, and the load-bearing cable clamp as feature recognition points, target detection is performed on the second original image set to screen out a second target image set containing the contact network suspension imaging area.

8. The contact network imaging method based on image intelligent recognition according to claim 7, characterized in that: Before performing target detection on the original image set, the method further includes: The images in the original image set are preprocessed, and the preprocessing includes image cropping and image rotation.

9. The overhead line imaging method based on image intelligent recognition according to claim 8, characterized in that: The method of selecting an image of interest within the range of the shooting depth of field from a target image set according to the shooting depth of field specifically includes: A first image of interest within a first shooting depth of field is obtained by screening from a first target image set at a first shooting depth of field, wherein the first shooting depth of field is a first distance from the overhead line support device to the first camera in a horizontal direction, and the first distance is 6.5 to 7.5 meters; With the second shooting depth of field, the first image of interest within the second shooting depth of field is screened from the second target image set. The second shooting depth of field is the second distance from the contact network to the second camera in the horizontal direction of the suspension, and the second distance is 6 to 7 meters.

10. A contact network imaging system based on image intelligent recognition, characterized in that: Applied to the imaging method according to any one of claims 1 to 9, the system comprises: An image acquisition module is used to acquire a set of original images taken by the camera during movement; A target detection module, used to perform target detection on the original image set, and filter out a target image set containing a target imaging area; A screening module, used to obtain a shooting depth of field, and to screen an image of interest within a shooting depth of field from a target image set according to the shooting depth of field; The storage module is used to select an image closest to the target from the images of interest as an archived image.

Images