Object recognition method and system

Abstract

The present application provides an object recognition method and system. First, a two-dimensional image and a three-dimensional image of the platform gap are obtained through multiple cameras and scanning radars. Then, each two-dimensional image is processed by a first preprocessing unit to obtain a processed two-dimensional image. Finally, the processed two-dimensional image and three-dimensional image are comprehensively calculated by an edge computing processor to identify the target object in the platform gap and determine the position of the target object in the platform gap, so as to perform corresponding processing on the target object. In the present application, the two-dimensional image is combined with the three-dimensional image and corresponding analysis is performed, which can make up for the defects of the two types of images, not only can the target object in the platform gap be accurately identified, but also the position of the target object in the platform gap can be accurately determined.

Description

Technical Field

[0001] The present application relates to the field of detection and identification technology, and in particular to an object identification method and system. Background Art

[0002] With the rapid development of high-speed rail, the number of people using high-speed rail is gradually increasing, and the frequency of use of high-speed rail is increasing. In order to increase the number of times high-speed rail is used every day, the stop time of high-speed rail at each station is shortened. Correspondingly, the transfer time of passengers is shortened, and the stop inspection time of train conductors is shortened. There is a wide gap between the train door and the safety door of the platform. When the train passes by at high speed, a huge negative pressure will be generated, which is enough to roll people or objects in the gap into the track 5. However, passengers are easily trapped in the gap in a crowded and panic situation, or due to the untimely inspection of the train conductor, some sundries are easily trapped in the gap, thereby causing major safety accidents of people and property.

[0003] Therefore, it is necessary to check whether there are target objects such as people or objects in the above-mentioned gaps before the train passes through the platform. When there are target objects, it is necessary to issue a notification in time to avoid danger. Usually, a camera can be used to capture images of the gaps in the platform, and the captured images can be processed and analyzed to determine whether there are target objects in the gaps. However, this method can only determine whether there are target objects, but cannot specifically calibrate the position of the target objects in the platform. In order to accurately calibrate the specific position of the target object, the gaps in the platform can be scanned by laser, microwave radar, etc. The scanning can obtain a three-dimensional image of the target object, thereby determining the specific position of the target object. However, the accuracy of the target object obtained by scanning is low, and it is impossible to scan objects at long distances and of small size, and the recognition accuracy of the target object is low. Summary of the invention

[0004] The present application provides an object recognition method and system to improve the recognition efficiency and recognition accuracy of a target object.

[0005] In a first aspect, the present application provides an object recognition method for object recognition in a platform gap, the method comprising:

[0006] A plurality of cameras are used to shoot along the direction of the track in the platform gap to obtain a plurality of two-dimensional images of the platform gap. The optical centers of the cameras of the plurality of cameras are vertically aligned and arranged to shoot the platform gap between the track and the platform safety door. Each camera corresponds to a different focal length, and the plurality of two-dimensional images are used to display the specific morphology of each object in the platform gap, wherein one camera corresponds to one two-dimensional image; a sub-image that meets a preset image clarity in each of the two-dimensional images is obtained by a first preprocessing unit, and each of the sub-images is spliced ​​to obtain a processed two-dimensional image, wherein the processed two-dimensional image includes each object in the platform gap; a scanning radar is used to scan along the direction of the track in the platform gap to obtain a three-dimensional image of the platform gap, wherein the three-dimensional image is used to display the position of each object in the platform gap in the platform gap; the edge computing processor identifies the target object in the platform gap according to the processed two-dimensional image and the three-dimensional image, and determines the position of the target object in the platform gap.

[0007] In a second aspect, the present application provides an object recognition system, which includes: multiple cameras, a first preprocessing unit, a scanning radar and an edge computing processor; the multiple cameras are electrically connected to the first preprocessing unit; the first preprocessing unit and the scanning radar are both electrically connected to the edge computing processor; wherein the camera optical centers of the multiple cameras are vertically aligned, and the multiple cameras correspond to different focal lengths.

[0008] The present application provides an object recognition method and system. First, a two-dimensional image and a three-dimensional image of the platform gap are obtained through multiple cameras and scanning radars. Then, each two-dimensional image is processed by a first preprocessing unit to obtain a processed two-dimensional image. Finally, the processed two-dimensional image and three-dimensional image are comprehensively calculated by an edge computing processor to identify the target object in the platform gap and determine the position of the target object in the platform gap, so as to perform corresponding processing on the target object. In the present application, the two-dimensional image is combined with the three-dimensional image and corresponding analysis is performed, which can make up for the defects of the two types of images, not only can the target object in the platform gap be accurately identified, but also the position of the target object in the platform gap can be accurately determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0010] Figure 1 A schematic diagram of a high-speed rail operation provided for this application;

[0011] Figure 2 A schematic diagram of the structure of an object recognition system provided in an embodiment of the present application;

[0012] Figure 3 A schematic diagram of the arrangement of a zoom camera provided in an embodiment of the present application;

[0013] Figure 4 A schematic diagram of a two-dimensional image provided in an embodiment of the present application;

[0014] Figure 5 A schematic diagram of the arrangement of fixed-focus cameras provided in an embodiment of the present application;

[0015] FIG6 (1) is a schematic diagram of a two-dimensional image of a platform gap provided in an embodiment of the present application;

[0016] FIG6(2) is a schematic diagram of a two-dimensional image of a platform gap provided in an embodiment of the present application;

[0017] FIG6(3) is a schematic diagram of a two-dimensional image of a platform gap provided in an embodiment of the present application;

[0018] Figure 7 A schematic diagram of a processed two-dimensional image provided in an embodiment of the present application;

[0019] Figure 8 A schematic diagram of a three-dimensional image provided in an embodiment of the present application;

[0020] Fig. 9 A schematic diagram of a multi-object recognition system configuration provided in an embodiment of the present application.

[0021] Illustration Description:

[0022] Among them, 1-train, 2-platform, 3-tunnel, 4-platform gap, 5-track, 6-safety door, 100-object recognition system, 101-main frame, 102-camera, 1021-first preprocessing unit, 103-scanning radar, 1031-second preprocessing unit, 104-edge computing processor, 105-public auxiliary system, 1051-power module, 1052-communication module, 106-pan-tilt device, 107-cloud server. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Figure 1 A schematic diagram of the operation of a high-speed railway is provided for the present application. Train 1 is running in a tunnel 3. When train 1 is running at high speed, a huge negative pressure is generated. This negative pressure is enough to draw people or objects within a certain distance around train 1 into track 5. In order to prevent such incidents from happening, a safety door 6 is provided at an appropriate position on platform 2 to prevent people or objects from getting too close to train 1. With the rapid development of high-speed railways, the number of people using high-speed railways has gradually increased, and the frequency of use of high-speed railways has increased. In order to increase the number of times high-speed railways are used every day, the stop time of high-speed railways at each station is shortened. Correspondingly, the transfer time of passengers is shortened, and the stop inspection time of conductors is shortened. There is a wide gap between the safety door 6 of train 1 and platform 2, such as Figure 1 The shaded part in the middle may be called the platform gap 4. Once passengers are trapped in the platform gap 4 in a crowded and chaotic situation, or if some debris is trapped in the platform gap 4 due to the conductor's failure to check in time, it is very easy to cause major safety accidents involving people and property.

[0025] Therefore, it is necessary to check whether there are target objects such as people or objects at the platform gap 4 before the train 1 passes through the platform 2. When there are target objects, it is necessary to issue a notification in time to avoid danger. Usually, a camera can be used to capture images at the platform gap 4, and the captured images can be processed and analyzed to determine whether there are target objects at the gap. However, this method can only determine whether there are target objects, but cannot specifically calibrate the position of the target object in the platform 2. In order to accurately calibrate the specific position of the target object, the platform gap 4 can be scanned by laser, microwave radar, etc. The scanning can obtain a three-dimensional image of the target object, thereby determining the specific position of the target object. However, the accuracy of the target object obtained by scanning is low, and it is impossible to scan objects at a long distance or of small size, and the recognition accuracy of the target object is low.

[0026] In order to solve the above problems and improve the recognition efficiency and recognition accuracy of the target object, the embodiment of the present application provides an object recognition system, such as Figure 2 As shown, the object recognition system 100 includes: a plurality of cameras 102, a first preprocessing unit 1021, a scanning radar 103 and an edge computing processor 104. Specifically, the object recognition system 100 can be set at an appropriate position in the tunnel 3, such as the ceiling, side wall or ground of the tunnel 3. Since the present application is mainly aimed at identifying target objects on the platform 2, it is necessary to set the object recognition system 100 at a position higher than the platform 2, and try to set the object recognition system 100 at a height higher than the average height of the person to avoid human damage. Preferably, the object recognition system 100 is set at a higher position, so that when shooting an object, it is not easy to have the problem of the object being blocked along the direction (longitudinal) of the track 5.

[0027] like Figure 2 As shown, in order to ensure the integrity of the object recognition system 100, each device in the object recognition system 100 can be installed on a main frame 101, and the main frame 101 serves as a support. The main frame 101 can be made of stainless steel frame, lightweight aluminum frame, titanium and other materials, which can not only ensure the strength, but also delay surface corrosion and extend the service life when the main frame 101 needs to be set in an open-air environment.

[0028] One object recognition system 100 is equivalent to one detection point. In order to ensure the detection quality, multiple detection points can be set in the same area, that is, multiple object recognition systems 100 can be set.

[0029] like Figure 2 As shown, the camera 102, the scanning radar 103, and the edge computing processor 104 are all arranged on the main support 101. In order to ensure the normal operation of the object recognition system 100, as Figure 2 As shown, the object recognition system 100 also includes a public auxiliary system 105, which mainly includes a power module 1051 and a communication module 1052. Specifically, in this embodiment, the power module 1051 can use a local AC / DC power supply or environmental energy (such as a solar cell), etc. According to actual use needs, the public auxiliary system 105 can also include an AC / DC or DC power conversion module to convert the electric energy provided by the power module 1051 into the required type of electric energy. At the same time, the public auxiliary system 105 can also include a protective component to cooperate with the on-site implementation environment to do a good job of anti-interference and insulation protection. In this embodiment, the communication module 1052 can be a 4G or 5G communication module, so that the object recognition system 100 can interact with the cloud server through the communication module 1052, such as receiving a device adjustment signal from the cloud server, or sending a collected data signal to the cloud server. In order to ensure the speed and quality of signal transmission, the communication module 1052 preferably has a high reliability and high communication capacity communication system.

[0030] The public auxiliary system 105 can ensure the normal power supply of each electrical component in the object recognition system 100 (camera 102, scanning radar 103, edge computing processor 104, etc.). The public auxiliary system 105 can be turned on by an external power switch to power other electrical components, or a power supply instruction can be sent to the communication module 1052 through a control device such as a mobile phone to instruct the public auxiliary system 105 to power each electrical component.

[0031] After the public auxiliary system 105 supplies power to the object recognition system 100, other devices in the object recognition system 100 begin to work normally, as follows:

[0032] The distance that the station 2 needs to detect is usually long. The camera has its corresponding focal length, that is, the distance or range that the camera can clearly shoot. Objects within the focal length of the camera or close to the focal length of the camera are relatively clear after imaging, and objects farther away from the focal length of the camera are relatively blurred after imaging. Therefore, in order to ensure the clarity of the captured image, it is necessary to ensure that the identified distance is within the focal length range of the camera.

[0033] In one implementation, the multiple cameras 102 may include at least one zoom camera, which can adjust the focal length of the camera according to the distance between the object to be photographed and the optical center of the camera, so that the object to be photographed is always located at the focus of the camera, ensuring the clarity of the captured image. The zoom camera usually corresponds to a focal length range, such as 2 to 20 meters. If the object to be photographed is within this distance range, the zoom camera can capture the image by adjusting the focal length. Furthermore, if the distance to be identified exceeds the focal length range of the zoom camera, for example, the detection distance is 10 to 100 meters, then the distance of 20 to 100 meters also needs to be captured. At this time, a zoom camera corresponding to other focal length ranges can be used for shooting, such as a zoom range of 20 to 200 meters, or multiple zoom cameras, the focal length ranges of these zoom cameras can be combined to cover 20 to 200 meters. In this way, the overall focal length range corresponding to the combination of multiple zoom cameras can meet the requirements of the recognition distance. It should be noted that these zoom cameras are set in a way that the optical center of the camera is vertically aligned. In this way, the starting point position for calculating the focal length of these zoom cameras is the same, which can facilitate subsequent calculations and avoid the process of field of view correction, which can speed up the calculation speed of the overall object recognition process. Figure 3As shown, if the multiple cameras 102 include three zoom cameras: camera A, camera B and camera C, the focal length range corresponding to camera A is 2-10m, the focal length range corresponding to camera B is 10-50m, and the focal length range corresponding to camera C is 50-100m. It can be seen that the overall focal length range corresponding to the multiple cameras 102 is 2-100m. If the recognition distance is 5-100m, then the overall focal length range corresponding to the multiple cameras 102 can cover the recognition distance. For each camera, during the shooting process, it is necessary to focus on each object to be photographed within its corresponding focal length range, and take a two-dimensional image of the platform gap 4 after each focusing is completed. Taking camera A as an example, if there is object a at a position of 3m, camera A needs to be focused to 3m, and then take the first 2D image of platform gap 4; if there is object b at a position of 5m, camera A needs to be focused to 5m, and then take the second 2D image of platform gap 4; if there is object c at a position of 8m, camera A needs to be focused to 8m, and then take the third 2D image of platform gap 4. It can be seen that for a zoom camera, when taking the platform image of platform gap 4, it is necessary to continuously focus according to the position of the object and take pictures in multiple times. The shooting time of the 2D image of platform gap 4 is the cumulative value of multiple focusing and shooting times. Although the clarity of the 2D image of platform gap 4 is very high, the overall shooting time is relatively long.

[0034] In another implementation, the multiple cameras 102 may include at least two fixed-focus cameras. The focal length of the fixed-focus camera is a fixed value and cannot be adjusted according to the location of the object. For example, if the focal length of the fixed-focus camera is 10m, then the image of the object located at 10m is the clearest when the fixed-focus camera is used. Figure 4As shown, object d is located at 10m, object e is located at 9m, and object f is located at 3m. The image clarity of object d is the highest, while the image clarity of objects e and f is relatively poor. However, if the image clarity of objects within 5m from the focus does not affect the subsequent analysis and processing of the image, then it can be considered that the focal length range is 5 to 15m, and the image clarity of objects within this focal length range can meet the requirements, that is, the image clarity of objects d and e both meet the requirements. If the distance to be identified exceeds the focal length range of the fixed-focus camera, for example, if the identification distance is 10 to 100 m, then images need to be captured at a distance of 15 to 100 m. In this case, multiple fixed-focus cameras corresponding to different focal lengths can be used for combined shooting. For example, the multiple cameras 102 include a fixed-focus camera with a focal length of 10 m; a fixed-focus camera with a focal length of 20 m, and its corresponding focal length range is 15 to 25 m; a fixed-focus camera with a focal length of 30 m, and its corresponding focal length range is 25 to 35 m; a fixed-focus camera with a focal length of 40 m, and its corresponding focal length range is 50 to 60 m. The focal length range is 35-45m; the fixed-focus camera with a focal length of 50m has a corresponding focal length range of 45-55m; the fixed-focus camera with a focal length of 60m has a corresponding focal length range of 55-65m; the fixed-focus camera with a focal length of 70m has a corresponding focal length range of 65-75m; the fixed-focus camera with a focal length of 80m has a corresponding focal length range of 75-85m; the fixed-focus camera with a focal length of 90m has a corresponding focal length range of 85-95m; the fixed-focus camera with a focal length of 100m has a corresponding focal length range of 95-105m. It can be seen that the focal length range corresponding to these fixed-focus cameras is 5-105m, which can cover the recognition distance. Therefore, a clear planar image of the platform gap 4 at the recognition distance can be captured. It should be noted that, if Figure 5 As shown, these fixed-focus cameras also need to be set in a manner that the optical center of the camera is vertically aligned. In this way, the starting position of the focal length calculation of these fixed-focus cameras is the same, which can facilitate subsequent calculations, and can avoid the process of field correction, and can speed up the calculation speed of the overall object recognition process. Further, when shooting a two-dimensional image at the recognition distance of the platform gap 4, each camera in the camera device 102 can shoot at the same time, and two-dimensional images of different resolutions can be obtained through one shooting. Compared with a zoom camera, there is no need for a focusing process, and only one shooting is required, which can greatly shorten the overall shooting time.

[0035] In this embodiment, the camera may be a visible light camera, an infrared camera, a special spectrum camera, a multi-spectral camera, etc., and each camera in the camera device 102 includes a wide-angle, medium-range, and beyond-viewing-range camera.

[0036] Combining the above two implementation methods, if multiple cameras 102 are arranged from top to bottom in the order of focal length from small to large, correspondingly, as shown in Figure 6 (1)(2)(3), a two-dimensional image of the platform gap 4 at each focal length is obtained. In this way, these two-dimensional images can be used for subsequent image calculations.

[0037] Furthermore, when calculating a plane image, the main reference parameter is the clarity of the image. The higher the clarity of the image, the more reference value it has. As can be seen from the above, there are some parts with lower clarity in the two-dimensional image captured by the camera 102. These parts have less reference value for subsequent calculations and may even affect the accuracy and speed of subsequent calculations. Therefore, a first preprocessing unit 1021 can be set in the object recognition system 100, the camera 102 is electrically connected to the first preprocessing unit 1021, and the first preprocessing unit 1021 is electrically connected to the edge computing processor 104. The first preprocessing unit 1021 has image cropping and splicing functions. After each camera 102 captures a two-dimensional image, the first preprocessing unit 1021 can crop the part with higher image clarity, so that the part with lower clarity can be eliminated, thereby replacing each two-dimensional image with a cropped two-dimensional image with a relatively small data volume, as shown in the dotted part of each two-dimensional image in Figure 6 (1) (2) (3). It can be seen that the image obtained after being cropped by the first preprocessing unit 1021 is only an image of a part of the objects in the platform gap 4. Therefore, in order to facilitate the analysis of all objects in the platform gap 4, it is necessary to splice the cropped two-dimensional images together through the first preprocessing unit 1021 to form a Figure 7 The processed two-dimensional image shown in the figure can clearly show each object in the platform gap 4 on one processed two-dimensional image. In order to avoid missing objects in the platform gap 4, it is necessary to ensure that the size of each cropped two-dimensional image can be equal to the size of the two-dimensional image before cropping when spliced ​​together, or the edge overlap splicing method is used between each cropped two-dimensional image, and the objects in the overlapped part can be based on any cropped two-dimensional image, or based on the cropped two-dimensional image with higher definition.

[0038] Furthermore, the first preprocessing unit 1021 may include functions such as color correction, light enhancement, and effective data extraction. In this way, through color correction, the color contrast of the image can be brighter, the outline can be clearer, and it can be closer to the actual appearance of the object; through light enhancement, the image with poor lighting can be made brighter, thereby further improving the clarity of the image; through effective data extraction, the effective data of specified objects in the image, such as faces, eyes, etc., can be extracted, and then these effective data can be used to identify the identity of the person, etc.

[0039] Only a two-dimensional image of the platform gap 4 can be obtained through the camera 102. If a three-dimensional coordinate system is established with the horizontal direction perpendicular to the track 5 as the x-axis, the vertical direction perpendicular to the track 5 as the y-axis, and the direction along the track 5 as the z-axis, then the two-dimensional image of the platform gap 4 can only reflect the x and y coordinates of each object in the platform gap 4. It can be seen that the two-dimensional image of the platform gap 4 can only show the specific morphological features of the object, such as facial features, body features, patterns on the object and other specific details, as well as the lateral position of the object (x and y coordinates). However, the specific position of the object in the platform gap 4 (including the z coordinate) cannot be displayed. Since the width of the platform gap 4 is relatively small and the longitudinal depth is large, after knowing the z coordinate, it is equivalent to determining the specific position of the object in the platform gap 4.

[0040] In order to obtain the longitudinal position of the object in the platform gap 4, the platform gap 4 can be scanned by the scanning radar 103, and 3D modeling can be performed for the scanned object to obtain a three-dimensional image of the platform gap 4. In this embodiment, a single-line or multi-line scanning radar can be used, including but not limited to laser radar, microwave radar, millimeter wave radar, ultrasonic radar and other radars that measure based on the TOF principle. The scanned image reflects the three-dimensional image of the platform gap 4. At this time, the three-dimensional image can reflect the longitudinal distance of each object in the platform gap 4 from the radar scanning device 103, such as Figure 8 As shown, the three-dimensional image of the platform gap 4 can display the three-dimensional images of the objects a, b and c in the platform gap, and can directly calibrate the longitudinal depths of the objects a, b and c.

[0041] The object recognition system 100 also includes a second preprocessing unit 1031, which has distance calculation and calibration functions. There are some objects corresponding to the basic environment in the platform gap 4, such as the platform 2, the safety door 6, etc. These are preset fixed facilities of the platform 2 and will not affect the travel of the train 1. In addition, the setting positions (intervals, sizes, etc.) of these objects have relatively fixed and strict regulations and have certain reference value. Therefore, they can be used as a reference for judging the positions of other objects. For example, Figure 8In the figure, object a is the platform, object b is the safety door, and object c is the target object. At this time, the second preprocessing unit 1031 can further calibrate the position of object c according to the position of object b, for example, along the direction of track 5, object c is 1m away from object b. Reference objects generally include landmark objects such as safety doors, which are convenient for relevant personnel to identify. Because after determining the position of the target object, relevant personnel usually need to go to its location for processing. If the longitudinal depth of the target object in the platform gap 4 is directly informed, for example, 88m, then the relevant personnel need to measure from the radar scanning device 103 to the position of 88m to find the target object, which is slow and has low accuracy. However, if the relevant personnel are informed of the reference object corresponding to the target object, since the reference object is easier to identify, the relevant personnel can quickly find the reference object. At this time, the relevant personnel can further find the target object based on the reference object and the calibrated distance between the target object and the reference object, which is fast and has high accuracy.

[0042] The three-dimensional image of the platform gap 4 shows a relatively rough body outline, which does not have specific morphological features. Therefore, it is only possible to roughly identify the corresponding object category through the outline, such as people, dogs, boxes, etc., but it is impossible to determine the corresponding specific object. This is not convenient for formulating countermeasures according to specific objects. For example, if the target object is a person, dog, suitcase, etc., relevant personnel need to take evacuation measures immediately; if the target object is a bottle-shaped object, it can be a plastic bottle or a bottle with high hardness (a material that will affect the normal operation of the train 1). The specific information of this type of object is an important basis for determining the countermeasures, but it is not possible to accurately determine the corresponding specific object only through the outline, and it is impossible to accurately formulate countermeasures. Since the outline of each object in the scanned image is too rough, only objects with larger sizes and clearer outlines are easier to identify their corresponding categories. This type of object whose object category can be determined by the outline can be called the first type of object, but this type of object cannot be accurately corresponded to a specific person or object. For objects with smaller sizes and longer vertical distances, it is almost impossible to distinguish their corresponding object categories by scanning images. This type of object can be called the second type of object.

[0043] The second preprocessing unit 1031 can also have an effective data screening function. As can be seen from the above, for the first type of objects, the corresponding category of the object can be determined by the outline. In this way, if a more accurate response can be formulated based on the category alone, there is no need to identify the specific morphology of the first type of objects, and there is no need to continue to perform subsequent processing and calculations on the first type of objects. The first type of objects can be accurately found directly based on the calibrated distance. At this time, the second preprocessing unit 1031 can remove the first type of objects from the scanned image to obtain a processed three-dimensional image. In this way, the amount of data contained in the three-dimensional image can be effectively reduced, which is convenient for subsequent processing and calculation of the three-dimensional image. It should be noted that in the case where the specific morphology of the first type of objects needs to be further determined, the second preprocessing unit 1031 does not need to perform effective data screening processing.

[0044] The radar scanning device 103 can periodically scan the platform gap 4 to obtain multiple three-dimensional images. Once a moving object appears, the moving object corresponds to different longitudinal depths in different three-dimensional images. At this time, the second preprocessing unit 1031 can calibrate the distance of the moving object in each three-dimensional image.

[0045] Furthermore, in order to improve the shooting quality and shooting range of two-dimensional and three-dimensional images, a pan-tilt device 106 can be set on the main support 101. The pan-tilt device 106 can adopt a six-axis pan-tilt. The camera 102 and the scanning radar 103 are both set on the pan-tilt device 106, and the vertical, up and down pitch and left and right rotation are achieved through the drive of the pan-tilt device 106. The pan-tilt device 106 preferably has a high-speed, silent servo motor system to achieve the purpose of fast and precise movement. Since a 1° deviation will result in a 1.72-meter offset at a distance of 100 meters, this error will directly lead to the loss of the target tracked by the scanning radar 103. Therefore, the angular rotation deviation of the pan-tilt device 106 should be less than 1°. The pan-tilt device 106 ensures the breadth coverage capability of the camera 102 and the scanning radar 103, and provides a basis for the subsequent long-distance tracking and discrimination of moving targets. From the above analysis, it can be seen that the images captured by the camera 102 and the scanning radar 103 have their own problems that need to be improved. At this time, the edge computing processor 104 can be used to integrate the processed two-dimensional image and three-dimensional image.

[0046] A correspondence between each object in the processed two-dimensional image and each object in the three-dimensional image is established, wherein if the three-dimensional image needs to be preprocessed, a correspondence between each object in the processed two-dimensional image and each second-category object in the processed three-dimensional image is established.

[0047] In one implementation, the edge computing processor 104 may establish a correspondence between objects based on the lateral position and outer contour of the objects in the processed two-dimensional image and the lateral position and outer contour of each second-category object in the processed three-dimensional image.

[0048] Specifically, the lateral position (x coordinate and y coordinate) of the object in the processed two-dimensional image can be matched with the lateral position (x coordinate and y coordinate) of the object in the scanned image, wherein objects with the same lateral position are the same object in the platform gap 4. Based on the above, if there are multiple objects with the same lateral position, they can be matched according to the outer contour of the object. Specifically, the category of the object can be known through the outer contour, and objects with the same category are the same object in the platform gap 4.

[0049] In another implementation, the first pre-processing unit 1021 may calculate the first outline size of each object according to the focal length of the camera corresponding to the sub-image where each object is located and the size of each object in the sub-image.

[0050] When processing the two-dimensional images captured by each camera 102, the first pre-processing unit 1021 will crop and splice each two-dimensional image, wherein each cropped image obtained by cropping is a sub-image, and each sub-image is captured by a camera 102, so each sub-image corresponds to a focal length. The real size of the object can be calculated based on the size of the object on the two-dimensional image and the focal length corresponding to the two-dimensional image. Therefore, the first pre-processing unit 1021 can calculate the real size (first outline size) of each object based on the size of each object on the sub-image and the focal length corresponding to the sub-image.

[0051] The second preprocessing unit 1031 calculates the second outline size of the second type of object according to the size of the second type of object in the processed three-dimensional image and the longitudinal distance calibrated in the processed three-dimensional image.

[0052] There is a corresponding relationship between the real size of the second type of object, the size in the processed three-dimensional image, and the corresponding longitudinal distance of the second type of object in the platform gap 4. Therefore, the second preprocessing unit 1031 can calculate the real size (second contour size) of the second type of object based on the size of the second type of object in the processed three-dimensional image and the longitudinal distance calibrated in the processed three-dimensional image.

[0053] In this way, the edge computing processor 104 can establish a correspondence between each object in the processed two-dimensional image and each second-category object in the processed three-dimensional image according to the first outline size and the second outline size.

[0054] Specifically, the edge computing processor 104 matches each first outline size with each second outline size, and establishes a corresponding relationship between objects whose first outline size matches the second outline size.

[0055] In summary, when the same object in the processed two-dimensional image and the processed three-dimensional image is determined, the corresponding relationship between the object in the processed two-dimensional image and the object in the processed three-dimensional image is established. Based on this corresponding relationship, the target object can be identified and the specific position of the target object in the platform gap 4 can be determined. For example, the target object A corresponds to a in the processed two-dimensional image and a' in the processed three-dimensional image. A corresponding relationship is established between a and a'. In this way, the specific shape (a) of the target object A and the specific position (a') of the target object A in the platform gap 4 can be determined based on the corresponding relationship.

[0056] If the scanning radar 103 periodically captures three-dimensional images, the edge computing processor 104 needs to establish a corresponding relationship between the objects in each three-dimensional image, for example, based on the outer contour and true size of the object. Please refer to the above description and will not be repeated here. After the corresponding relationship between the objects in each scanned image is established, the target object to send the displacement, that is, the moving object, can be determined. For the detection of such objects, it is usually necessary to determine the moving speed of the moving object. The longitudinal depth of the moving object in each scanned image is calibrated according to the calibration method of the object position provided above. According to the change value of the longitudinal depth and the scanning period corresponding to the change value, the moving speed of the moving object can be calculated using the speed calculation formula. At the same time, the corresponding running trajectory of the moving object can be depicted according to the position of the moving object in each three-dimensional image. It can be seen that the edge computing processor 104 has completed most of the calculation tasks for the plane image and the scanned image, so that the calculation result obtained is smaller than the image data volume before the calculation.

[0057] The edge computing processor 104 can determine the position of each object in the platform gap 4, and at the same time, the specific shape corresponding to each object can be determined. In this way, it can be judged whether relevant personnel need to handle it based on the specific shape of the object, and moving objects can also be determined to facilitate relevant personnel to pursue them.

[0058] In this application, the camera 102 is combined with the scanning radar 103, that is, the two-dimensional image is combined with the three-dimensional image and corresponding analysis is performed, which can make up for the defects of the two types of images, not only can the position of the target object in the platform gap 4 be accurately determined, but also the specific shape of the target object in the platform gap 4 can be accurately determined.

[0059] Furthermore, the object recognition system 100 also includes a cloud server 107. After the edge computing processor 104 calculates various parameters of the target object, these parameters are transmitted to the cloud server 107 through the communication module 1052 in the public auxiliary system 105, and the cloud server 107 further processes these parameters. For example, the specific person information corresponding to the face can be determined based on big data comparison, deep learning and multi-network data fusion, the specific information of dangerous goods such as explosives can be determined, the expected movement trajectory of moving objects can be predicted, and the connection with security systems such as the public security system can be achieved.

[0060] In summary, the object recognition system 100 involves three image data processing processes in the process of analyzing image data, namely, the preprocessing process of the first preprocessing unit 1021 and the second preprocessing unit 1031 for two-dimensional images and three-dimensional images, the integration processing process of the edge computing processor 104 for two-dimensional images and three-dimensional images, and the deep learning and extensive comparison processing process of the image data by the cloud server 107. After each level of data processing, the image data will greatly reduce the amount of data transmitted downward, which can effectively reduce the bandwidth pressure, increase the transmission speed, and facilitate subsequent data calculations, making the calculation process closer to the perception layer, more in line with the concept of edge computing, thereby reducing the calculation pressure of the cloud server 107, and allowing the cloud server 107 to serve more object recognition systems 100 at the same time to provide cloud computing services.

[0061] Furthermore, the object recognition system 100 can be used to identify the situation along the entire track 5. At this time, since the main support 101 is equipped with a camera 102, a scanning radar 103, an edge computing processor 104 and a public auxiliary system, it mainly relies on the camera 102 and the scanning radar 103 to shoot and scan objects in the platform gap 4. The camera 102 shoots images based on the propagation of light, and the scanning radar 103 scans objects based on the emission and reception of electromagnetic waves. When the light or electromagnetic waves are blocked, the normal shooting of the camera 102 and the scanning radar 103 will be affected. Therefore, in order to ensure the normal operation of the camera 102 and the scanning radar 103, that is, to ensure the image quality of the two shots or scans, it is necessary to avoid large obstacles within the working range of the two. Corresponding to the operation scenario of the train 1 in the tunnel 3, the turning point of the tunnel is the obstacle that is not recommended to appear. Therefore, it can be as follows Fig. 9 As shown, the object recognition system 100 is arranged on each straight section along the track 5 to segmentally recognize the target object on each straight section.

[0062] As can be seen from the above, the present application provides an object recognition method and system. First, two-dimensional images and three-dimensional images of the platform gap are obtained through multiple cameras and scanning radars. Then, each two-dimensional image is processed by a first preprocessing unit to obtain a processed two-dimensional image. Finally, the processed two-dimensional image and three-dimensional image are comprehensively calculated by an edge computing processor to identify the target object in the platform gap and determine the position of the target object in the platform gap, so as to perform corresponding processing on the target object. In the present application, the two-dimensional image is combined with the three-dimensional image and a corresponding analysis is performed, which can make up for the defects of each of the two types of images, and can not only accurately identify the target object in the platform gap, but also accurately determine the position of the target object in the platform gap.

[0063] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

[0064] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. An object recognition method, characterized in that: Applied to an object recognition system, the object recognition system is used for object recognition in platform gaps, the object recognition system includes multiple cameras, a first preprocessing unit, a scanning radar and an edge computing processor; The method comprises: The multiple cameras are used to shoot along the direction of the track in the platform gap to obtain multiple two-dimensional images of the platform gap, the optical centers of the multiple cameras are vertically aligned and used to shoot the platform gap between the track and the platform safety door, each camera corresponds to a different focal length, and the multiple two-dimensional images are used to show the specific appearance of each object in the platform gap, wherein one camera corresponds to one two-dimensional image; The first preprocessing unit obtains a sub-image of each of the two-dimensional images that meets the preset image definition, and splices the sub-images to obtain a processed two-dimensional image, wherein the processed two-dimensional image includes each object in the platform gap; Scanning along the track direction in the platform gap by the scanning radar to obtain a three-dimensional image of the platform gap, wherein the three-dimensional image is used to display the position of each object in the platform gap in the platform gap; The object recognition system further includes a second preprocessing unit, the scanning radar is electrically connected to the second preprocessing unit, and the second preprocessing unit is electrically connected to the edge computing processor, and the scanning radar is used to scan along the track direction in the platform gap to obtain a three-dimensional image of the platform gap, and further includes: Determine, by the second preprocessing unit, a first type of object in the three-dimensional image, wherein the first type of object is an object other than a reference object and the type of the object can be determined by the outline without determining the specific shape, and the reference object is an object corresponding to a preset fixed facility; Eliminating the first type of objects from the three-dimensional image by the second pre-processing unit to obtain a processed three-dimensional image; identifying, by the edge computing processor, a target object in the platform gap according to the processed two-dimensional image and the three-dimensional image, and determining a position of the target object in the platform gap; Identifying, by the edge computing processor, a target object in the platform gap according to the processed two-dimensional image and the three-dimensional image, and determining a position of the target object in the platform gap includes: determining, by the second preprocessing unit, a second type of object in the processed three-dimensional image, wherein the second type of object is an object other than the reference object in the processed three-dimensional image; The edge computing processor establishes a correspondence between each object in the processed two-dimensional image and the second-category objects based on the lateral position and outer contour of each object in the processed two-dimensional image and the lateral position and outer contour of each second-category object in the processed three-dimensional image, so as to determine the specific morphology of the target object and its position in the platform gap based on the corresponding relationship, and the target object is the second-category object.

2. The method according to claim 1, characterized in that The object recognition system further includes a pan-tilt device, the plurality of cameras and the scanning radar are both arranged on the pan-tilt device, and the method further includes: The shooting positions and shooting angles of the multiple cameras and the scanning radar are adjusted by the pan-tilt device so that the multiple cameras and the scanning radar can obtain two-dimensional images and three-dimensional images of the platform gap at the adjusted shooting positions and shooting angles.

3. The method according to claim 1, characterized in that The obtaining of sub-images that meet the preset image definition in each of the two-dimensional images by the first pre-processing unit, and splicing the sub-images to obtain the processed two-dimensional image further includes: A valid image is extracted from the processed two-dimensional image by the first preprocessing unit, and the valid image is an image corresponding to a preset image.

4. The method according to claim 1, characterized in that: The scanning along the track direction in the platform gap by the scanning radar to obtain a three-dimensional image of the platform gap further includes: The longitudinal distance corresponding to each object in the three-dimensional image in the platform gap is calibrated by the second preprocessing unit, and the longitudinal distance is the distance between the object and the scanning radar.

5. The method according to claim 4, characterized in that The scanning along the track direction in the platform gap by the scanning radar to obtain a three-dimensional image of the platform gap further includes: Determine, by the second preprocessing unit, a reference object and a target object in the three-dimensional image, wherein the target object is an object in the platform gap other than the reference object; The distance between the target object and the reference object is calibrated by the second preprocessing unit according to the corresponding longitudinal distance of each object in the platform gap.

6. The method according to claim 4, characterized in that The step of identifying the target object in the platform gap according to the processed two-dimensional image and the three-dimensional image by the edge computing processor, and determining the position of the target object in the platform gap includes: Calculating, by the first preprocessing unit, a first outline size of each object according to the focal length of the camera corresponding to the sub-image where each object is located and the size of each object in the sub-image; Calculating, by the second preprocessing unit, a second outline size of the second type of object according to a size of the second type of object in the processed three-dimensional image and a longitudinal distance calibrated by the second type of object in the processed three-dimensional image; The edge computing processor establishes a correspondence between each object in the processed two-dimensional image and each second-category object in the processed three-dimensional image according to the first contour size and the second contour size, so as to determine the specific shape of the target object and its position in the platform gap according to the correspondence, and the target object is the second-category object.

7. The method according to claim 1, characterized in that The object recognition system further includes a cloud server, the cloud server being communicatively connected to the edge computing processor, and the method further includes: The cloud server combines the cloud data and the two-dimensional image corresponding to the target object to determine whether the target object is a preset search object.

8. An object recognition system, characterized in that: Used to perform the method according to any one of claims 1 to 7, the object recognition system comprising: a plurality of cameras, a first pre-processing unit, a scanning radar, and an edge computing processor; The multiple cameras are electrically connected to the first pre-processing unit; The first pre-processing unit and the scanning radar are both electrically connected to the edge computing processor; The optical centers of the cameras of the plurality of cameras are vertically aligned, and the plurality of cameras correspond to different focal lengths; The object recognition system also includes a second preprocessing unit, the scanning radar is electrically connected to the second preprocessing unit, and the second preprocessing unit is electrically connected to the edge computing processor.

Images