A target identification method and device

Abstract

The present invention discloses a target recognition method and device. When image recognition is turned on, a reference image is set by collecting a current scene image, and the reference image of the scene where the camera is located is restored. In the subsequent recognition process, feature information of the image to be recognized is extracted, and a difference comparison is performed with the feature information of the saved reference image. Differential images are extracted from the image to be recognized and the reference image respectively, and target recognition processing is performed on the extracted differential images to obtain target recognition results corresponding to the image to be recognized and the reference image respectively, and the target recognition results corresponding to the image to be recognized and the reference image respectively are compared. According to the comparison result, the reference image is updated or the recognized target object is output. The present invention continuously updates the reference image, thereby improving the accuracy of image recognition in complex scenes; through the recognition of differential images, the resource consumption of image recognition is reduced, and more accurate recognition results can be obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to a target recognition method and device. Background Art

[0002] Economic development and the improvement of people's living standards have benefited from the rapid development of road traffic and vehicle transportation. However, due to the sharp increase in the number of cars, factors such as traffic congestion and severe environmental pollution have restricted the development of social economy and the improvement of people's quality of life. The declining traffic efficiency and the increasing number of traffic accidents are difficult to improve. Today, when the contradiction between environment and resources is becoming increasingly prominent, simply relying on the growth of road facilities and other basic means to solve transportation needs can no longer meet the needs of social development. How to use advanced technology to ensure the efficiency and safety of transportation is a hot issue today.

[0003] Against this background, the Intelligent Transportation System (ITS) came into being. The Intelligent Transportation System is an efficient management layer of the entire transportation management system. It effectively integrates high technologies such as electronic control technology, data communication transmission technology, information technology and computer processing technology to manage the road transportation system. As a result, the comprehensive transportation management system formed is more accurate and efficient, and can play a role in all directions and on a large scale. It has brought into play the potential of existing transportation infrastructure, improved the interaction between transportation subsystems such as vehicles, roads and people, thereby improving transportation efficiency, alleviating traffic congestion, ensuring traffic safety, reducing energy consumption and environmental pollution, and improving social and economic benefits as a whole, opening up a bright road for solving traffic problems.

[0004] In intelligent transportation systems, the identification of motor vehicles, non-motor vehicles, pedestrians and other suspicious objects on the road network plays a key guiding role in the establishment of road traffic systems, and image recognition technology is the premise and condition for realizing highway traffic automation. Therefore, image recognition technology has great practical significance for improving the degree of automation of highway traffic and promoting the development of intelligent transportation systems.

[0005] In the existing intelligent transportation field, the bayonet camera provides image recognition technology, including the recognition of motor vehicles, non-motor vehicles, pedestrians, etc. The existing technology mainly collects images through the bayonet camera, inputs a single image into the image recognition module, and the image recognition module extracts features from the content of the image and outputs information such as motor vehicles, non-motor vehicles, pedestrians, etc. in the image. However, within the visual range of the bayonet camera, if the background of the picture is complex, it will have a great impact on the image recognition technology, reducing the accuracy of the image recognition result. Summary of the invention

[0006] The purpose of the present invention is to provide a target recognition method and device to overcome the problem of low accuracy in image recognition in the prior art.

[0007] In order to achieve the above object, the technical solution of the present invention is as follows:

[0008] A target recognition method, the target recognition method comprising:

[0009] Obtain the image to be identified, extract the feature information of the image to be identified, perform difference comparison with the feature information of the saved reference image, and obtain a list of difference area coordinates;

[0010] According to the obtained difference region coordinate list, differential images are extracted from the image to be identified and the reference image respectively, and target recognition processing is performed on the extracted differential images to obtain target recognition results corresponding to the image to be identified and the reference image respectively;

[0011] The target recognition results corresponding to the image to be recognized and the reference image are compared, and according to the comparison result, the reference image is updated or the recognized target object is output.

[0012] Optionally, the target identification method further includes:

[0013] When image recognition is turned on, the current scene image is collected, and a check is made to see if there is a saved reference image. If there is no saved reference image, the current scene image is set as the reference image. If there is a saved reference image, the current scene image is compared with the saved reference image.

[0014] If the matching degree between the current scene image and the reference image is less than a preset threshold, the reference image is updated to the acquired current scene image, otherwise the reference image remains unchanged.

[0015] Furthermore, when the image recognition is turned on, the current scene image is collected and compared with the saved reference image, and the method further includes:

[0016] According to the current preset position of the image acquisition device, a corresponding reference image is selected from the reference images corresponding to each preset position that is saved;

[0017] Capture the current scene image and compare it with the selected reference image.

[0018] Furthermore, the comparison with the saved reference image also includes: obtaining position information of the image acquisition device, and when the position has not changed, if the current scene image is compared with the selected reference image and it is found that the scene has changed, traversing the saved reference images, and when the corresponding reference image is matched, updating the reference image corresponding to each preset position of the saved image acquisition device.

[0019] Optionally, updating the reference image according to the comparison result includes:

[0020] If the target recognition results corresponding to the image to be recognized and the reference image are not the target object, saving the differential images corresponding to the image to be recognized and the reference image, and counting each saved differential image;

[0021] According to the counting result, the difference image is superimposed on the reference image to update the reference image.

[0022] Optionally, updating the reference image according to the comparison result includes:

[0023] If the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is not the target object, the differential image corresponding to the image to be recognized is superimposed on the reference image to update the reference image.

[0024] Optionally, outputting the identified target object according to the comparison result includes:

[0025] If the target recognition result corresponding to the reference image is not the target object, and the target recognition result corresponding to the image to be recognized is the target object, the recognition result corresponding to the image to be recognized is output.

[0026] Optionally, outputting the identified target object according to the comparison result includes:

[0027] If the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is the target object, the recognition result corresponding to the image to be recognized is output.

[0028] Furthermore, the comparing of the target recognition results corresponding to the image to be recognized and the reference image, when the target recognition result corresponding to the video image to be recognized is not the target object, further includes:

[0029] If the target recognition result corresponding to the image to be recognized is not the target object, and the recognized object appears for the first time, an abnormal alarm is issued.

[0030] The present invention also proposes a target identification device, comprising a processor and a non-volatile memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the above method when executed by the processor.

[0031] The present invention provides a target recognition method and device, which continuously updates the reference image through training with a large number of samples, thereby improving the accuracy of image recognition in complex scenes. By recognizing differentiated images, the resource consumption of image recognition is reduced, and more accurate recognition results can be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 This is a flow chart of the target identification method of the present invention;

[0033] Figure 2 This is a flowchart of initializing a reference image according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0035] In one embodiment, Figure 1 As shown, a target recognition method is provided, which is applied to a camera in a video surveillance system, comprising:

[0036] Step S1, when image recognition is turned on, the current scene image is collected and compared with the saved reference image. If the matching degree between the current scene image and the reference image is less than a preset threshold, the reference image is updated to the collected current scene image, otherwise the reference image remains unchanged.

[0037] Video surveillance systems are widely used in transportation, public security, enterprises, and families. This application is applicable to any image acquisition device (ordinary camera, PTZ camera, etc.), or other devices for target recognition (such as background servers, etc.). In the embodiment, the camera in the field of intelligent transportation is used as an example for detailed description, which is also applicable to video surveillance systems in other fields. Or the background server performs target recognition, which requires the image acquisition device to provide the captured image.

[0038] In the field of intelligent transportation, bayonet cameras are often deployed at major traffic intersections to count the traffic participants on the road and provide effective data support for subsequent road traffic planning and adjustment. The technical solution of this application is mainly applicable to scenes with sparse moving objects, such as traffic intersections, and is not applicable to scenes with large traffic flow, such as shopping mall entrances.

[0039] This embodiment takes a bayonet camera as an example. When the image recognition function is enabled on the bayonet camera, Figure 2As shown, first capture an image of the current scene, and then check whether there is a reference image on the camera. If not, set the current scene image currently captured as the reference image. Otherwise, if it already exists, extract the feature information of the current scene image and the reference image respectively, and compare the two images. According to the matching degree of the comparison result, determine whether the current scene has been switched. For example, if the matching degree is greater than the set threshold (for example, 80%), it is considered that the current scene has not changed, and the reference image remains unchanged. On the contrary, if the matching degree is less than 80%, it is considered that there is a large difference in the current scene, and it can be considered that the scene has changed. The previously stored reference image is no longer applicable to the current scene. At this time, the reference image is updated to the captured current scene image.

[0040] In this embodiment, it is necessary to first check whether a reference image already exists. If the camera is deployed for the first time and no reference image has been saved, the captured current scene image is directly saved as the reference image.

[0041] In one embodiment, if the bayonet camera is a pan-tilt camera, since the pan-tilt camera is provided with a plurality of preset positions, each preset position corresponds to a different scene, and a corresponding reference image needs to be set for each preset position. In other words, different reference images are set for different scenes. In this embodiment, the current preset position can be determined according to the rotation angle of the pan-tilt camera or according to the position information of the pan-tilt camera, and the corresponding reference image can be obtained.

[0042] That is, in this embodiment, when image recognition is enabled, the current scene image is captured and compared with the saved reference image, and the following steps are also included:

[0043] Save the reference image corresponding to each preset position of the camera;

[0044] When image recognition is turned on, a reference image corresponding to each preset position is selected from the reference images corresponding to each preset position saved according to the current preset position of the camera;

[0045] Capture the current scene image and compare it with the selected reference image.

[0046] Specifically, when the user adjusts the position of the camera, the scene will also change, and it is necessary to synchronously update the reference image corresponding to the current position. In order to reduce the impact of the pan / tilt rotation on subsequent image recognition, this embodiment can save the original reference image and the corresponding extracted feature information each time the pan / tilt rotates to the preset position, and at the same time, it is necessary to associate the current pan / tilt rotation angle or preset position information. Since the rotation angle corresponds to the preset position one by one, the following description is taken as an example.

[0047] For example, the reference image of scene one corresponds to a pan / tilt rotation angle of 30 degrees, and the reference image of scene two corresponds to a pan / tilt rotation angle of 60 degrees. This information is saved in the camera, as shown in Table 1. If the pan / tilt rotates later, first search Table 1 for a corresponding reference image. If no corresponding reference image is found, re-capture the current scene image as the reference image. If a corresponding reference image is found, directly use the original reference image to avoid the camera re-training the reference image from scratch, which leads to the problem of low initial result accuracy.

[0048] The information stored in the camera is as follows:

[0049] Camera latitude and longitude information N30.17788,E120.22137 PTZ Angle Scene Information 30 Scene 1 60 Scene 2

[0050] Table 1

[0051] After the user has adjusted the camera position and started the image recognition function, the following operations need to be performed. If the camera is a PTZ camera, it is necessary to first determine the current rotation angle of the PTZ, and then find the reference image corresponding to the current position based on the rotation angle. Then, combined with the steps in step S1, it is determined whether the scene has changed and whether the reference image corresponding to the current scene needs to be retrained.

[0052] In one embodiment, the comparing with the saved reference image further comprises:

[0053] Get the camera's position information. When the position has not changed, if the current scene image is compared with the selected reference image and it is found that the scene has changed, traverse the saved reference images. When the corresponding reference image is matched, update the reference image corresponding to each preset position of the camera.

[0054] This embodiment selects the reference image more accurately. If the camera has a GPS positioning function, the latitude and longitude information of the camera can be bound to the scene. For example, if the latitude and longitude information of the camera has not changed, but the matching degree between the reference image obtained under the corresponding pan-tilt angle and the current scene image is quite different, it can be considered that the installation angle has shifted. At this time, all the previously stored reference images can be traversed and compared. If the pan-tilt rotation angle of the camera is 30 at this time, but the pan-tilt rotation angle of the matched reference image saved in the camera is 60. It can be considered that the camera installation angle has shifted horizontally by 30 degrees, and the original reference image can be modified in turn. For example, the scene information saved in the above table is updated as shown below:

[0055]

[0056]

[0057] Table 2

[0058] In this embodiment, step S1 completes the setting of the reference image, so that when the subsequent recognition starts, the reference image is already saved. It is easy to understand that even if the reference image is not set in step S1, that is, the saved reference image is empty, the reference image can be updated in subsequent steps.

[0059] In this embodiment, when image recognition is turned on, the reference image is set by capturing the current scene image, and the reference image of the scene where the camera is located is saved. Therefore, when target recognition is performed on the image to be recognized later, the saved reference image can be directly used for comparison and recognition, so that the target can be recognized quickly after the camera is turned on.

[0060] The following describes in detail the process of processing the image to be identified, updating the reference image and identifying the target.

[0061] Step S2: Obtain the image to be identified for target identification, extract feature information of the image to be identified, perform difference comparison with the feature information of the saved reference image, and obtain a difference area coordinate list.

[0062] After the reference image has been determined, the camera starts to work normally and performs target recognition on the image to be recognized. At this time, the camera collects images, and the collected images are used as the image to be recognized for feature extraction operations to extract the feature information of the picture. feature .

[0063] The feature information O of the reference image in the scene in the camera feature With P feature Perform difference comparison and return the difference area coordinate list L region .

[0064] Step S3: extracting differential images from the image to be identified and the reference image respectively according to the obtained difference region coordinate list, performing target recognition processing on the extracted differential images, and obtaining target recognition results corresponding to the image to be identified and the reference image respectively.

[0065] For L region Analyze and extract the difference image from the image and the reference image respectively. That is, according to the obtained difference area coordinate list L region The difference area coordinate list L is placed in the reference image and image respectively region The corresponding areas are cropped out to obtain L in the two images. region The image L corresponding to the coordinates picture 1 and L picture 2. Here we assume that L picture 1 is the difference image extracted from the reference image, L picture 2 is the difference image extracted from the image to be identified.picture 1 and L picture 2. Perform feature extraction respectively and perform target recognition processing on the extracted features.

[0066] This embodiment only performs target recognition processing on the differentiated image, which reduces the amount of calculation for target recognition. The targets generally include moving targets such as motor vehicles, non-motor vehicles, and people, and this application does not limit this. Regarding specific target recognition methods, there are many mature technologies in the field that can be used, which will not be repeated here.

[0067] Step S4: compare the target recognition results corresponding to the image to be recognized and the reference image, and update the reference image or output the recognized target object according to the comparison result.

[0068] This application compares the target recognition results of the image to be recognized and the reference image. The following situations may occur, and different processing is performed respectively:

[0069] Case 1: If the target recognition results corresponding to the image to be identified and the reference image are not the target object, the differential images corresponding to the image to be identified and the reference image are saved, and each saved differential image is counted; according to the counting result, the differential image is superimposed on the reference image to complete the update of the reference image.

[0070] In this case, L picture 1 The result of recognition is not any of the target objects (such as motor vehicles, non-motor vehicles, people), that is, other objects, while L picture 2 The recognized result is not any of the target objects, that is, it is also another object.

[0071] It is easy to understand that L picture 1 or L picture 2 are the differential images extracted separately, and the objects recognized by the two are different. picture 1 and L picture 2 and the coordinate data of the corresponding reference image are saved and counted, and the initial count value is 1.

[0072] Subsequently, the new image to be recognized is processed. If the recognition result shows the situation 1 again, it is necessary to first determine the current L picture 1 or L picture 2. Whether it is consistent with the previously saved differential image (the judgment condition for consistency can be that the matching degree between the two is greater than 90%). If it is consistent, the count is directly added to the original data by 1, otherwise a new image information is created and saved.

[0073] In order to reduce the number of judgments, only 3 images are saved by default. In fact, at least 3 images need to be kept. For example, if only 2 differential images are saved, if the first 2 scenes appear repeatedly many times, a new scene will appear at this time. It will not be updated to the baseline image because the count is less than the previous 2. However, the new unexpected scene may be kept for a long time. If the original scene is not updated, it will result in the need to perform feature comparison on the area every time, wasting camera resources. Each time you save, the differential image data with fewer counts and older time will be discarded first.

[0074] In this case, the differential image is superimposed on the reference image according to the counting result, thereby completing the updating of the reference image.

[0075] Among them, the statistics of the counting results can be counted once according to a preset time interval, for example, once every 5 minutes or 10 minutes, and then the differential image with the highest count is superimposed on the benchmark image, and the update of the benchmark image is completed. The number of processing times of the image to be identified can also be set, for example, setting the processing of 50 frames of images to be counted once, or setting the processing of 100 frames of images to be counted once, and this application does not limit this. After the number of processing times is reached, the differential image with the highest count is superimposed on the benchmark image after the threshold, and the update of the benchmark image is completed. It is also possible to directly superimpose the differential image on the benchmark image based on whether the counting result exceeds the set threshold, for example, as long as the differential image count exceeds 3 times. The present application is not limited to the above specific methods. Counting personnel in this field can superimpose the differential image on the benchmark image according to the counting results to complete the update of the benchmark image.

[0076] Case 2: if the target recognition result corresponding to the reference image is not the target object, and the target recognition result corresponding to the image to be recognized is the target object, then the recognition result corresponding to the image to be recognized is output.

[0077] Specifically, L picture 1 The result of recognition is not any of the target objects, that is, other objects, while L picture 2The identified result is one of the target objects.

[0078] This indicates that L in the image to be identified picture 2. If there is a target object in the area, the area is marked and displayed to the user, and the identified target is output. There is no need to update the reference image at this time.

[0079] Case 3: if the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is not the target object, then the differential image corresponding to the image to be recognized is superimposed on the reference image to update the reference image.

[0080] Specifically, Lpicture 1 The result of recognition is one of the target objects, that is, other objects, while L picture 2 The recognized result is not any of the target objects, that is, it is another object.

[0081] At this time, according to the strategy, L picture 2 is closer to the original scene (that is, there is no target object information in the scene, only static objects). In this case, L picture 2 is superimposed on the corresponding coordinate area of ​​the reference image to generate the latest reference image, and the feature information of the reference image is extracted and saved. The subsequent collected images are compared with the latest reference image for difference. At this time, there is no need to save L picture 1 target identification information.

[0082] Case 4: If the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is the target object, then the recognition result corresponding to the image to be recognized is output.

[0083] Specifically, L picture 1 The result of recognition is one of the target objects, while L picture 2The identified result is also one of the target objects.

[0084] At this time, it can be considered that there is a target object in this area in both images, and the area in the captured image is marked and displayed to the user, and the recognized target is output, and there is no need to update the reference image.

[0085] As the number of images collected increases, the application continuously learns and trains the collected images and continuously repeats the above steps, so that the baseline image of the current scene becomes more realistic, and the comparison difference between the subsequent collected images and the baseline image of the current scene will be closer and closer to the required result, that is, only the recognition results of target objects such as motor vehicles, non-motor vehicles, pedestrians, etc. are concerned, thereby improving the camera's recognition accuracy of target objects such as motor vehicles, non-motor vehicles, pedestrians, etc. This application focuses on continuous training and optimization of the environment in which the object is located, so that the scene is closer to the original scene. Therefore, by comparing the difference with the original scene, the recognition accuracy of the camera is greatly improved. In summary, when the target recognition result corresponding to the image to be recognized is not the target object, the baseline image is updated; when the target recognition result corresponding to the image to be recognized is the target object, the recognized target is output.

[0086] It should be understood that although Figure 1The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0087] In one embodiment, in the above situation 1, if the target recognition result corresponding to the image to be recognized is not the target object, and the recognized object appears for the first time, an abnormal alarm is issued.

[0088] Specifically, when an object identified in Lpicture2 appears for the first time, an alarm can be selected to alert the user that an abnormal object has appeared in the current scene. In particular, for cameras deployed at traffic intersections, if a suspicious object suddenly appears on the road, if it can be discovered in time, a huge safety hazard will be eliminated in time.

[0089] Finally, when the camera turns off the image recognition function, the final reference image and the corresponding pan / tilt rotation angle are saved in the camera as the reference image when the image recognition function is started next time. At the same time, the intermediate data in the recognition process is cleared to release related resources.

[0090] The present application first captures the picture of the current scene when starting the image recognition function, and uses the first captured image as the reference image of the current scene and saves it in the camera. At the same time, feature extraction is performed on the reference image, and the extracted feature information is saved in the memory for comparison with the image collected by the subsequent camera. Image recognition is performed on the difference parts of the comparison results of the two images, and the information of motor vehicles, non-motor vehicles, pedestrians, etc. is identified. If the difference part of the reference image is identified as information such as motor vehicles, non-motor vehicles, pedestrians, etc., and the subsequent collected image is not information such as motor vehicles, non-motor vehicles, pedestrians, etc., but other objects, it can be considered that the subsequent collected image is more likely to be close to the original scene (i.e., it does not contain moving objects such as motor vehicles, non-motor vehicles, pedestrians, etc.), then this part of the difference of the subsequent collected image is superimposed on the reference image, that is, the information of the reference image is updated, and the superimposed image is saved at the same time, and the features of the new reference image are extracted at the same time. By analogy, as the number of recognitions increases, the original scene information will be continuously trained and optimized, and the subsequent collected images will be compared with the reference image, and the difference part is closer to the moving object, i.e., the object to be identified. At this time, extracting the image data of the difference part and performing image recognition can obtain more accurate recognition results.

[0091] In one embodiment, the present application further provides a target identification device, which includes a processor and a non-volatile memory storing a plurality of computer instructions, wherein the computer instructions implement the steps of the above target identification method when executed by the processor.

[0092] For the specific definition of the target recognition device, please refer to the definition of the target recognition method above, which will not be repeated here. Each module in the above target recognition device can be implemented in whole or in part by software, hardware and a combination thereof. The above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0093] The memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, these elements can be electrically connected to each other through one or more communication buses or signal lines. The memory stores a computer program that can be run on the processor, and the processor implements the network topology layout method in the embodiment of the present invention by running the computer program stored in the memory.

[0094] The memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc. The memory is used to store a program, and the processor executes the program after receiving an execution instruction.

[0095] The processor may be an integrated circuit chip with data processing capabilities. The above-mentioned processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention may be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0096] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A target recognition method, characterized in that: The target recognition method comprises: Obtain the image to be identified, extract the feature information of the image to be identified, perform difference comparison with the feature information of the saved reference image, and obtain a list of difference area coordinates; According to the obtained difference region coordinate list, differential images are extracted from the image to be identified and the reference image respectively, and target recognition processing is performed on the extracted differential images to obtain target recognition results corresponding to the image to be identified and the reference image respectively; Comparing the target recognition results corresponding to the image to be recognized and the reference image, and according to the comparison result, superimposing the difference image onto the reference image to update the reference image or outputting the recognized target object; Wherein, superimposing the differential image onto the reference image to update the reference image according to the comparison result includes: If the target recognition results corresponding to the image to be recognized and the reference image are not the target object, saving the differential images corresponding to the image to be recognized and the reference image, and counting each saved differential image; According to the counting result, the differential image is superimposed on the reference image to update the reference image; Or, if the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is not the target object, then superimposing the differential image corresponding to the image to be recognized onto the reference image to update the reference image; And, outputting the identified target object according to the comparison result includes: If the target recognition result corresponding to the reference image is not the target object, and the target recognition result corresponding to the image to be recognized is the target object, then outputting the recognition result corresponding to the image to be recognized; Alternatively, if the target recognition result corresponding to the reference image is the target object, and the target recognition result corresponding to the image to be recognized is the target object, the recognition result corresponding to the image to be recognized is output.

2. The target recognition method according to claim 1, characterized in that: The target recognition method further includes: When image recognition is turned on, the current scene image is collected, and a check is made to see if there is a saved reference image. If there is no saved reference image, the current scene image is set as the reference image. If there is a saved reference image, the current scene image is compared with the saved reference image. If the matching degree between the current scene image and the reference image is less than a preset threshold, the reference image is updated to the acquired current scene image, otherwise the reference image remains unchanged.

3. The target recognition method according to claim 2, characterized in that: When the image recognition is turned on, the current scene image is collected and compared with the saved reference image, and further includes: According to the current preset position of the image acquisition device, a corresponding reference image is selected from the reference images corresponding to each preset position that is saved; Capture the current scene image and compare it with the selected reference image.

4. The target recognition method according to claim 3, characterized in that: The comparison with the saved reference image also includes: obtaining the position information of the image acquisition device, and when the position has not changed, if the current scene image is compared with the selected reference image and it is found that the scene has changed, traversing the saved reference images, and when the corresponding reference image is matched, updating the reference image corresponding to each preset position of the saved image acquisition device.

5. The target recognition method according to claim 1, characterized in that: The comparing of the target recognition results corresponding to the image to be recognized and the reference image, when the target recognition result corresponding to the image to be recognized is not the target object, further includes: If the target recognition result corresponding to the image to be recognized is not the target object, and the recognized object appears for the first time, an abnormal alarm is issued.

6. A target recognition device, comprising a processor and a non-volatile memory storing a plurality of computer instructions, characterized in that: When the computer instructions are executed by a processor, the steps of the method described in any one of claims 1 to 5 are implemented.

Images