Target image deduplication method and apparatus

Abstract

The application provides a target image deduplication method and device, and relates to the technical field of image processing. The method comprises the following steps: traversing data corresponding to each target in a primary deduplication set at a previous snapshot moment, determining time intervals based on a target snapshot moment and each snapshot moment included in the data corresponding to the target; splitting the primary deduplication set at the previous snapshot moment based on the time intervals to obtain sub-sets corresponding to the time intervals; sorting all the sub-sets in the order of the time intervals from small to large; traversing each sub-set after sorting, matching a target full set at the target snapshot moment with the sub-set currently traversed, and performing target small image deduplication based on the matching result. The size of each time interval reflects the degree of proximity of the snapshot moment, and each sub-set in the sub-sets after sorting is gradually matched and deduplicated with the target full set from near to far, thereby reducing the consumption of computing resources in the entire deduplication process.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and apparatus for deduplicating target images. Background Technology

[0002] With the widespread application of artificial intelligence (AI) technology in the security industry, AI-based target analysis technology has become the industry mainstream.

[0003] In related technologies, camera devices typically perform target detection on captured scene images to obtain raw data for each target. The raw data for each target is then sent to a backend device. To prevent the camera device from missing any targets or losing raw data, the backend device needs to perform full target analysis on all scene images included in each raw data set. Then, it needs to match all target images obtained from the full analysis with the target images in all the raw data sent by the camera device to remove duplicates.

[0004] However, in the aforementioned technologies, during the deduplication process, the backend equipment needs to match all target images obtained from the full analysis with all target images sent by the camera equipment one by one, resulting in a large amount of computing resources being consumed in the entire deduplication process. Summary of the Invention

[0005] This invention provides a method and apparatus for deduplicating target images, which solves the problem that the existing technology consumes a large amount of computing resources in the entire deduplication process.

[0006] This invention provides a method for deduplicating target images, comprising the following steps.

[0007] Iterate through the data corresponding to each target in the initial deduplication set of the previous capture time of the target captured by the camera device, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target. Based on each of the time intervals, the initial deduplication set of the previous capture time is split to obtain the subset corresponding to each of the time intervals; Sort all the subsets in ascending order of the time intervals to obtain the sorted subsets; Traverse each subset in the sorted subset, and match the small target images in the full set of targets at the target capture time with the small target images in the currently traversed subset to obtain the matching result. The full set of targets includes the small target images corresponding to all targets captured at the target capture time. Based on the matching results, duplicate target small images are removed.

[0008] According to a method for deduplication of a target image provided by the present invention, the method further includes: Acquire at least one piece of raw data sent by the camera device, the raw data including a scene image captured by the camera device, a small image of a first target, and first capture information corresponding to the small image of the first target; Target detection is performed based on each of the scene images to obtain at least one detection data, the detection data including a second target small image and second capture information corresponding to the second target small image; Based on each of the first capture information and each of the second capture information, duplicates are removed from all the first target small images and all the second target small images to obtain the full set of targets at the capture time.

[0009] According to a method for deduplication of target images provided by the present invention, the step of deduplicating small target images based on the matching results includes: If the matching result indicates a successful match, the successfully matched target small image will be predicted as a target small image of the same target; Based on the small images of the target that are predicted to be the same target, deduplication is performed to obtain the initial deduplication set of the target capture time; The initial set of duplicate capture times of the target is deduplicated.

[0010] According to the present invention, a method for deduplicating target images is provided, wherein the full set of targets includes key target identifiers and newly added target identifiers, the key target identifiers are used to characterize the corresponding image as an image in the original data, and the newly added target identifiers are used to characterize the corresponding image as an image in the detection data; The process of deduplicating small images of the target predicted to be the same target to obtain the initial deduplicated set at the time of target capture includes: If all targets predicted to be the same target are traversed, and the first image is identified as a key target while the second image is identified as a newly added target, then the second image is deleted. When both the identifier of the first image and the identifier of the second image are new target identifiers, the first image is an image in the full target set, and the second image is an image in a target subset, the image to be deleted is determined based on the attribute information of the first image and the attribute information of the second image, and the image to be deleted is deleted, and the target subset is the first subset in the sorted subset; If both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset, then delete the image with the smallest capture time between the first image and the second image. The non-target subset is the subset other than the first subset in the sorted subset. The full set of targets obtained from the final traversal and the initial deduplicated set of the previous capture time obtained from the final traversal are merged to obtain the initial deduplicated set of the target capture time.

[0011] According to a target image deduplication method provided by the present invention, the deduplication of the initial deduplication set of the target capture time includes: In the initial deduplication set at the target capture time, at least one third target small image that meets the target conditions is determined, and an image set is determined based on each of the third target small images; Feature extraction is performed on each target small image in the image set to obtain the features of each image; The image set is deduplicated based on the image features described above; The target condition includes one of the following: There are no images matching the third target small image in a preset number of historical capture times prior to the target capture time; There is no image matching the third target small image at the target capture time, and the difference between the current time and the access time of the third target small image is greater than the first preset time. The difference between the current time and the access time of the third target small image is greater than the second preset duration, and the second preset duration is greater than the first preset duration.

[0012] According to a method for deduplicating target images provided by the present invention, determining an image set based on each of the third target small images includes: For each of the aforementioned third target small images, if there is an associated identifier for the third target small image, the suspected target set associated with the third target small image and the associated identifier of the third target small image is added to the image set. The suspected target set includes the image with the largest capture time among the first image and the second image, provided that both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset.

[0013] According to a target image deduplication method provided by the present invention, determining the time interval based on the target capture time and each capture time included in the data corresponding to the target includes: Determine the difference between the target capture time and the capture times included in the data corresponding to the target, and determine the minimum difference as the time interval.

[0014] According to a target image deduplication method provided by the present invention, the step of matching the target mini-images in the full target set at the time of target capture with the target mini-images in the currently traversed subset to obtain a matching result includes: Based on the image information of the target small images in the full target set and the image information of the target small images in the currently traversed subset, the similarity distance is determined. The image information includes at least one of the contour information of the target in the corresponding target small image, the color information of the target, and the texture information of the target. The matching result is determined based on the similarity distance.

[0015] According to a target image deduplication method provided by the present invention, the first capture information includes a first target type, and the second capture information includes a second target type; The step of deduplicating all first target small images and all second target small images based on each of the first capture information and each of the second capture information to obtain the full set of targets at the time of target capture includes: For each of the target capture times, determine the overlap between each first target small image and each second target small image at the target capture time in the scene image; Based on the overlap, the first target type, and the second target type, all the first target small images and all the second target small images at the time of target capture are deduplicated to obtain the full set of targets.

[0016] The present invention also provides a target image deduplication device, comprising: The determining unit is used to traverse the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture time of the camera device, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target. The splitting unit is used to split the initial deduplication set of the previous capture time based on each of the time intervals to obtain the subsets corresponding to each of the time intervals. A sorting unit is used to sort all the subsets in ascending order of the time intervals to obtain sorted subsets; The matching unit is used to traverse each subset in the sorted subset, match the small target images in the full set of targets at the target capture time with the small target images in the currently traversed subset, and obtain the matching result. The full set of targets includes the small target images corresponding to all targets captured at the target capture time. The first deduplication unit is used to deduplicat the target small image based on the matching result.

[0017] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the target image deduplication method as described above.

[0018] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the target image deduplication method as described above.

[0019] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the target image deduplication method as described above.

[0020] The target image deduplication method and apparatus provided by this invention traverses the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture device. Based on the target capture time and the capture times included in the target-corresponding data, a time interval is determined. Based on each time interval, the initial deduplication set of the previous capture time is split to obtain subsets corresponding to each time interval. All subsets are sorted in ascending order of time intervals to obtain sorted subsets. Each subset in the sorted subsets is traversed, and the target small images in the target full set at the target capture time are matched with the target small images in the currently traversed subsets to obtain matching results. Based on the matching results, target small images are deduplicated. It can be seen that this invention sorts the split subsets in ascending order of time intervals. The size of each time interval reflects the proximity of the capture times. The subsets in the sorted subsets are matched with the target full set from near to far to remove duplicates, thereby reducing the consumption of computing resources in the entire deduplication process. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0022] Figure 1 This is one of the flowcharts illustrating the target image deduplication method provided in this embodiment of the invention.

[0023] Figure 2 This is a schematic diagram of the initial deduplication set provided in an embodiment of the present invention.

[0024] Figure 3 This is the second flowchart illustrating the target image deduplication method provided in this embodiment of the invention.

[0025] Figure 4 This is a schematic diagram of grouping based on the identifier of the camera device provided in an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of scene image deduplication provided in an embodiment of the present invention.

[0027] Figure 6 This is a schematic diagram of the target full set provided in an embodiment of the present invention.

[0028] Figure 7 This is a schematic diagram of the target image deduplication device provided in an embodiment of the present invention.

[0029] Figure 8 This is a schematic diagram of the physical structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0031] The following is combined with Figures 1-6 The present invention describes a method for deduplicating target images. The subject executing this method can be an electronic device such as a terminal, tablet computer, computer, or server, or a target image deduplication device installed in the electronic device. This target image deduplication device can be implemented through software, hardware, or a combination of both.

[0032] Image-based target analysis technology has two main applications in the security industry. The first involves backend devices (servers and other electronic equipment) re-analyzing specified targets based on reports from front-end devices (cameras) using their own algorithms. The second involves performing a full re-analysis of the target data based on the scene image reported by the front-end devices. This avoids data loss during network transmission and storage and compensates for limitations in front-end device support, such as limited target type detection and attribute analysis. However, the second scenario suffers from significant target duplication, which wastes computing and storage resources. Therefore, resolving the target duplication issue in full image analysis is crucial.

[0033] In related technologies, the backend device needs to match all target images obtained from full analysis with all target images in all raw data sent by the camera device to remove duplicates. However, during the deduplication process, the backend device needs to match all target images obtained from full analysis with all target images sent by the camera device, resulting in a large amount of computing resources being consumed in the entire deduplication process.

[0034] Based on this, the present invention provides a method for deduplicating target images. It can first split the initial deduplication set of the previous capture time based on each time interval, and then sort the split subsets according to the order of the time intervals from smallest to largest. The size of each time interval reflects the proximity of the capture time. The subsets in the sorted subsets are matched with the full target set step by step from near to far to deduplicate, thereby reducing the consumption of computing resources in the entire deduplication process.

[0035] Figure 1 This is one of the flowcharts illustrating the target image deduplication method provided in this embodiment of the invention, such as... Figure 1 As shown, the method for deduplicating the target image includes the following steps: Step 101: Iterate through the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture time of the camera device, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target.

[0036] Optionally, the difference between the target capture time and the capture times included in the data corresponding to the target is determined, and the minimum difference is determined as the time interval.

[0037] For example, Figure 2 This is a schematic diagram of the initial deduplication set provided in an embodiment of the present invention, as shown below. Figure 2As shown, the initial deduplication set includes data corresponding to each target captured by the same camera device. Each data entry includes all existing capture times of the target captured by the same camera device. The initial deduplication set is then iterated through from the previous capture time. The data corresponding to each target is used to calculate the full set of targets. The minimum difference between the corresponding target capture time and the capture times included in the data for that target is used to determine the time interval. Using the same method, we can obtain the time intervals corresponding to each target in the initial deduplication set of the previous capture moment.

[0038] It should be noted that after calculating the full target set... When considering the difference between the target capture time and the capture times included in the data corresponding to the target, the second smallest difference among all the differences can be determined as the time interval. This invention does not limit this.

[0039] Step 102: Based on each of the time intervals, split the initial deduplication set of the previous capture time to obtain the subset corresponding to each of the time intervals.

[0040] For example, when obtaining the time intervals corresponding to each target in the initial deduplication set at the previous capture time, the initial deduplication set at the previous capture time is split according to the same time interval to obtain the subsets corresponding to each time interval. That is, identical subsets The time intervals corresponding to each target are the same, but different subsets The time intervals corresponding to the targets are different.

[0041] Step 103: Sort all the subsets in ascending order of the time intervals to obtain the sorted subsets.

[0042] For example, sort the subsets according to their corresponding time intervals in ascending order to obtain the sorted subsets. Assuming the number of subsets is n, then use... arrive Representing each subset, express Minimum, express maximum.

[0043] Step 104: Traverse each subset in the sorted subset, and match the target small images in the full set of targets at the target capture time with the target small images in the currently traversed subset to obtain the matching result. The full set of targets includes the target small images corresponding to all targets captured at the target capture time.

[0044] For example, the entire target set Match it sequentially with each subset in the sorted subset, first matching the entire target set. The target small image and the first subset The target small image in the image is matched to obtain the matching result. The specific matching strategy is as follows: if and If the target types in the small target images are the same, the Hungarian algorithm is used for matching. and If the number of small target images varies, virtual nodes are introduced to act as proxies; if and The target images in the two images are of the same type, and similarity distance is used. As the value coefficient matrix, if the target types are different, the value coefficient matrix is ​​set to infinity, indicating that the two target images cannot be matched. The specific process of the Hungarian algorithm can be found in related technologies, and will not be elaborated upon here.

[0045] It should be noted that, for and In the matching process, if the target represented by two small target images is predicted to be the same target, then these two small target images are called similar targets. In other words, these two small target images are the target images that are predicted to belong to the same target with the smallest capture time difference.

[0046] It should be noted that the Hungarian algorithm is a fuzzy matching algorithm, not a structured feature matching algorithm. Therefore, the matching accuracy is related to the environment, lighting, target size, etc. As the time sequence is closer, the matching accuracy will be higher. However, it is possible that when the time sequence is closest, no target small image predicted to belong to the same target will be matched, while when the time sequence is the second closest, target small images predicted to belong to the same target will be matched. Moreover, the two target small images matched when the time sequence is the second closest are still the small images with the highest matching degree.

[0047] Step 105: Based on the matching results, perform deduplication of the target small image.

[0048] For example, if similar distance The value is greater than This indicates that the two small target images represent the same target, where, This indicates the maximum decay time, which defaults to 10 minutes. This represents the initial threshold for similarity distance, with a default value of 0.5. For small images of targets predicted to be the same object, from... The image is removed from the set and will not participate in the matching of subsequent subsets until all subsets have been traversed. This process is used to predict all small images of the same target in the full set of targets at the capture time and the initial deduplication set at the previous capture time. The image with the highest clarity among two small images of the same target can be retained, while the image with the lowest clarity among two small images of the same target can be removed, thus achieving deduplication.

[0049] It should be noted that this invention can be applied to scenarios where deduplication is performed on raw data sent in real time by camera devices, or on scenarios where deduplication is performed on stored raw data; this invention does not limit the application to either.

[0050] The target image deduplication method provided by this invention involves: traversing the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture device; determining time intervals based on the target capture time and the capture times included in the target-corresponding data; splitting the initial deduplication set of the previous capture time into subsets corresponding to each time interval based on each time interval; sorting all subsets in ascending order of time intervals to obtain sorted subsets; traversing each subset in the sorted subsets; matching the target small images in the full target set of the target capture time with the target small images in the currently traversed subsets to obtain matching results; and performing target small image deduplication based on the matching results. As can be seen, this invention sorts the subsets obtained from the splitting process according to the order of the time intervals from smallest to largest. The size of each time interval reflects the similarity of the capture time. The subsets in the sorted subsets are matched and deduplicated with the full target set step by step from near to far, thereby reducing the consumption of computing resources in the entire deduplication process. In addition, by determining the matching relationship between the full target set and the target small images in the traversed subsets from near to far, and then predicting the target small images belonging to the same target based on the matching results, the accuracy of predicting the same target can also be improved.

[0051] In one embodiment, Figure 3 This is a second schematic diagram of the target image deduplication method provided in the embodiments of the present invention, as shown below. Figure 3 As shown, prior to step 101 above, the target image deduplication method further includes the following steps: Step 301: Obtain at least one piece of raw data sent by the camera device, the raw data including a scene image captured by the camera device, a small image of a first target, and first capture information corresponding to the small image of the first target.

[0052] The first target small image can be understood as an image cropped from the scene image that includes a target. The first capture information can include the target type and attribute information of the target in the first target small image. The target type can be, for example, a face, a human body, a motor vehicle or a non-motor vehicle, etc., and the attribute information can be, for example, the target's clarity information, color information, texture information and contour information, etc.

[0053] For example, when the camera captures a scene image, it performs target detection on the scene image to obtain a first target small image and the first capture information corresponding to the first target small image for each target in the scene image. For each target, it generates raw data based on the scene image, the capture time of the scene image, including the first target small image of the target and the first capture information corresponding to the first target small image, and sends each piece of raw data to the backend electronic device, so that the backend electronic device can obtain the raw data sent by the camera.

[0054] It should be noted that when an electronic device receives raw data sent by at least two camera devices, it groups the raw data belonging to the same camera device according to the identifier of each camera device, thereby obtaining the raw data corresponding to each camera device. The camera device can also be called a point, that is, the raw data corresponding to each point is obtained. Figure 4 This is a schematic diagram illustrating grouping based on the identifier of the camera device provided in an embodiment of the present invention, such as... Figure 4 As shown, the received raw data may include face data, human body data, non-motorized vehicle data, and motorized vehicle data. Raw data belonging to the same location are grouped together, that is, raw data of different target types belonging to the same location are grouped together, and a target type is added to each raw data. The following description uses raw data from the same camera device as an example.

[0055] Step 302: Perform target detection based on each of the scene images to obtain at least one detection data, the detection data including a second target small image and the second capture information corresponding to the second target small image.

[0056] For example, when acquiring the raw data sent by the camera device, target detection can be performed on the scene images included in each raw data separately, or target detection can be performed on only one scene image at the same capture moment. The specific method of target detection can be referred to in related technologies, and will not be elaborated here.

[0057] Step 303: Based on each of the first capture information and each of the second capture information, deduplication is performed on all the first target small images and all the second target small images to obtain the full set of targets at the capture time.

[0058] For example, upon obtaining each first capture information and each second capture information, the first capture information and each second capture information are matched to obtain a matching result. Based on the matching result, all first target small images and all second target small images are deduplicated to obtain the full set of targets at the time of capture. .

[0059] In this embodiment, upon acquiring the raw data sent by the camera device, target detection is performed on the scene images included in each raw data, yielding at least one detection data set. Based on the first capture information of the first target small image in each raw data set and the second capture information of the second target small image in each detection data set, deduplication is performed on all first target small images and all second target small images. The complete target set at the target capture time facilitates subsequent deduplication based on the complete target set. It can be seen that the present invention can first perform deduplication on the first target small image in the raw data and the second target small image in the detection data based on the capture information, and then perform a second deduplication on the obtained complete target set. This multi-level deduplication method reduces the consumption of computing resources and the utilization of storage resources throughout the deduplication process.

[0060] In one embodiment, the raw data further includes the capture time of the scene image; step 302 above performs target detection based on each of the scene images to obtain at least one detection data, which can be implemented in the following way: Among all the scene images, identify each scene image belonging to the same capture time; deduplicate each scene image belonging to the same capture time to obtain the target scene image of the same capture time; for each capture time, perform target detection on the target scene image of the capture time to obtain at least one detection data.

[0061] For example, for all raw data sent by the same camera device, the raw data are sorted according to the capture time, and all scene images corresponding to the raw data belonging to the same capture time are deduplicated so that only one scene image is retained at the same capture time. This scene image is used as the target scene image at that capture time, and the target scene image is re-detected to obtain at least one detection data corresponding to the target scene image. The detection data includes the second target small image and the second capture information corresponding to the second target small image. Figure 5 This is a schematic diagram of scene image deduplication provided in an embodiment of the present invention, such as... Figure 5As shown, the left side displays the raw data sent by the same camera device. Each piece of raw data includes a scene image, the capture time of the scene image, the access time of the electronic device receiving the raw data, a first target small image, the position coordinates of the first target small image in the scene image, the target type of the first target small image, and the target recognition information of the target represented by the first target small image. This target recognition information can include attribute information such as the target's color, texture, contour, and sharpness. The right side displays the merged data corresponding to the same capture time from the same camera device. The merged data includes the capture time, access time, target scene image, and detection data corresponding to each target. The detection data includes a second target small image, the target type of the second target small image, the position coordinates of the second target small image in the target scene image, and the target recognition information of the target represented by the second target small image. Figure 5 As can be seen, by deduplicating the scene images belonging to the same capture time in each original data on the left, only one target scene image is retained for the same capture time on the right, thus achieving scene image deduplication.

[0062] It should be noted that for each piece of raw data acquired, the electronic device assigns a corresponding cache space to each camera device, and the maximum duration for each piece of raw data to remain in the corresponding cache space is [duration to be specified]. , It can be configured based on specific scenarios, for example, For a duration of 1 minute, the maximum amount of raw data that can be stored in the buffer space of each camera device is [amount missing]. The amount of original data stored in the cache space is greater than At that time, release the original data with the smallest capture time in the cache space. It can be configured based on specific scenarios, for example, There are 100 entries.

[0063] In this embodiment, duplicate scene images belonging to the same capture time are deduplicated, and target detection is only performed on one target scene image at the same capture time, which further reduces the consumption of computing resources.

[0064] In one embodiment, step 105 above performs deduplication of the target small image based on the matching result, which can be implemented in the following way: If the matching result indicates a successful match, the successfully matched target small image is predicted as a target small image of the same target; deduplication is performed on each target small image predicted as the same target to obtain the initial deduplication set of the target capture time; deduplication is performed on the initial deduplication set of the target capture time.

[0065] For example, if the matching result indicates a successful match, the successfully matched target image is predicted as a target image of the same target, and the full set of targets at the time of capture is: The initial deduplication set of the previous snapshot moment is Then predict and For target images belonging to the same target, either of the two predicted target images belonging to the same target can be removed from the corresponding set. Alternatively, based on a preset strategy, an image to be deleted can be selected from the two predicted target images belonging to the same target and deleted from the corresponding set. Following the same method, all target images belonging to the same target are deduplicated. The full target set obtained after the final deletion at the current capture time is merged with the initial deduplicated set obtained after the final deletion at the previous capture time to obtain the initial deduplicated set for the target capture time. This leads to the initial deduplication set of the captured moment. Further deduplication is performed. For example... Figure 2 As shown, the initial deduplication set includes data corresponding to each target captured by the same camera device. Each data entry includes the optimal capture time, the latest capture time, the latest access time, all existing capture times for that target and a list of attributes corresponding to each capture time, the target small image corresponding to the optimal capture time (which is the final retained target small image), the target type of the target small image corresponding to the optimal capture time, and the scene image to which the target small image corresponding to the optimal capture time belongs. The optimal capture time is... and The selected target's final small image capture time corresponds to the capture time of the target. The latest capture time is the maximum capture time among all existing capture times of the target. The latest access time is the time when the electronic device acquires the scene image corresponding to the latest capture time.

[0066] It should be noted that the initial deduplication set at the first capture moment is the full target set at the first capture moment. The initial deduplication set at the second capture moment can be determined based on the full target set at the second capture moment and the initial deduplication set at the first capture moment, and so on. This invention will not elaborate further here.

[0067] In this embodiment, the full set of targets at the target capture time is compared with the initial deduplicated set at the previous capture time. The target small images that belong to the same target at the two capture times are predicted. Then, the target small images predicted to be the same target are deduplicated, which further reduces the repetition of target small images and thus reduces the consumption of computing resources.

[0068] In one embodiment, the full set of targets includes key target identifiers and newly added target identifiers. The key target identifiers are used to indicate that the corresponding image is an image in the original data, and the newly added target identifiers are used to indicate that the corresponding image is an image in the detection data. The above-mentioned deduplication based on each small image of a target predicted to be the same target to obtain the initial deduplication set at the time of target capture can be implemented in the following way: If all targets predicted to be the same target are traversed, and the first image is identified as a key target while the second image is identified as a newly added target, then the second image is deleted. When both the identifier of the first image and the identifier of the second image are new target identifiers, the first image is an image in the full target set, and the second image is an image in a target subset, the image to be deleted is determined based on the attribute information of the first image and the attribute information of the second image, and the image to be deleted is deleted, and the target subset is the first subset in the sorted subsets; If both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset, then delete the image with the smallest capture time between the first image and the second image. The non-target subset is the subset other than the first subset in the sorted subset. The full set of targets obtained from the final traversal and the initial deduplicated set of the previous capture time obtained from the final traversal are merged to obtain the initial deduplicated set of the target capture time.

[0069] For example, when traversing all targets predicted to be the same objective, deduplication is performed using the following preset strategy: if the identifier of the first image is a key objective identifier and the identifier of the second image is a newly added objective identifier, the first image is retained because the image quality detected by the camera is better, and the second image is deleted from the corresponding set. When both the identifier of the first image and the identifier of the second image are newly added objective identifiers, the first image is an image in the full set of objectives, and the second image is an image in a subset of objectives, it indicates that the first image and the second image are two small images of the same objective predicted to have the smallest capture time difference. In this case, the attribute information of the first image and the attribute information of the second image are obtained. Taking clarity as an example, the image with the lowest clarity is identified as the image to be deleted and deleted from the corresponding set. Of course, the attribute information can also be brightness, etc., and this invention does not limit this. If both the first and second images are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset, it indicates that the first and second images are not two small target images predicted to belong to the same target with the smallest capture time difference. In this case, the first and second images are identified as suspected identical images, suspected identifiers are added to the first and second images respectively, and the target is generated. Suspected target set Used to store the target All suspected images are used to add the image with the smallest capture time between the first and second images to the suspected target set. Then, the image with the largest capture time between the first and second images is added to the initial deduplication set. In the initial deduplication set To remove duplicates, the image with the smallest capture time is not added; and in the initial deduplication set... Add the set of images with the largest capture time and the suspected target. The association identifier can be used to find a set of suspected targets. All images in the dataset. After traversing all targets, the final set of all targets obtained from the traversal is merged with the initial deduplicated set of the previous capture moment obtained from the traversal, to obtain the initial deduplicated set of the target capture moment.

[0070] It should be noted that this applies to the set of suspected targets. Establishment: When the target first appears When two images are suspected to be identical, generate the target. Suspected target set In the subsequent deduplication process, if the target is matched... If a new suspected image is found, it will be added to the suspected target set. middle.

[0071] In this embodiment, based on a preset strategy, the target small images in the full set of the target at the target capture time and the initial deduplication set of the previous capture time are deduplicated to obtain the initial deduplication set of the target capture time. This reduces the repetition of target small images in the initial deduplication set and further reduces the consumption of computing resources.

[0072] In one embodiment, the deduplication of the initial set of target capture times can be achieved in the following way: In the initial deduplication set at the target capture time, at least one third target small image that meets the target conditions is determined. Based on each third target small image, an image set is determined. Feature extraction is performed on each target small image in the image set to obtain each image feature. Based on each image feature, the image set is deduplicated.

[0073] The target condition includes one of the following: There are no images matching the third target small image in a preset number of historical capture times prior to the target capture time; There is no image matching the third target small image at the target capture time, and the difference between the current time and the access time of the third target small image is greater than the first preset time. The difference between the current time and the access time of the third target small image is greater than the second preset duration, and the second preset duration is greater than the first preset duration.

[0074] The target quantity can be adopted as follows: This represents the number of times a match was unsuccessful. The value can be configured based on specific scenarios, for example, Equals 5; the first preset duration can be adopted. To indicate, The value can be configured based on specific scenarios, for example, Equals 3 minutes; the second preset duration can be adopted. To indicate, The value can be configured based on specific scenarios, for example, That's 10 minutes.

[0075] For example, the initial deduplicated set for the capture time of this target. For any target small image in the dataset, in the first case, if no matching image exists within a preset number of historical capture times prior to the target's capture time, then the target small image is identified as the third target small image satisfying the target condition. In the second case, if no matching image exists at the target's capture time, the difference between the current time and the access time of the target small image is determined to be greater than a first preset duration; if the difference is greater than the first preset duration, then the target small image is identified as the third target small image satisfying the target condition. In the third case, the difference between the current time and the access time of the target small image is determined to be greater than a second preset duration; if the difference is greater than the second preset duration, then the target small image is identified as the third target small image satisfying the target condition. The target corresponding to the identified third target small image is the disappeared target, so it will be removed from the initial deduplication set of the target's capture time. All the small images of the third target identified in the process are considered as requiring exit. The image allows the entire deduplication process to proceed normally, and the initial deduplication set from the capture time of the target will be used. All the small images of the third target identified were added to the image set.

[0076] In this embodiment, based on three dimensions—capture time, access time, and number of unsuccessful matches—a mechanism is set up to remove the target small image from the initial deduplication set at the target capture time, so as to further achieve deduplication and further reduce the consumption of computing resources.

[0077] In one embodiment, the determination of the image set based on each of the third target small images can be implemented in the following way: For each of the aforementioned third target small images, if there is an associated identifier for the third target small image, the suspected target set associated with the third target small image and the associated identifier of the third target small image is added to the image set. The suspected target set includes the image with the largest capture time among the first image and the second image, provided that both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset.

[0078] For example, the initial deduplicated set for the capture time of this target. Each third target small image in the image is determined. Whether to add an associated identifier to the third target small image, in order to determine When a correlation marker is added to the third target small image, a set of suspected related targets is found based on the correlation marker. Both the set of suspected related targets and the third target small image are added to the image set. Then, features are extracted from each target small image in the image set to obtain semi-structured features and / or structured features of each target small image. Taking image features as semi-structured features as an example, when obtaining the semi-structured features of each target small image, a cosine distance is calculated based on every two semi-structured features. If the cosine distance is greater than a preset distance, the targets in the two target small images corresponding to the two semi-structured features are determined to be different targets. If the cosine distance is less than or equal to the preset distance, the targets in the two target small images corresponding to the two semi-structured features are determined to be the same target. Then, the image with the highest clarity among the two target small images is retained, and the image with the lowest clarity among the two target small images is removed, thus achieving the final deduplication.

[0079] In this embodiment, if the third target small image in the initial deduplication set at the time of target capture has an association identifier, the third target small image and the suspected target set associated with the association identifier of the third target small image are added to the image set together. This facilitates subsequent deduplication of the image set, minimizes the amount of duplicate target small images, and improves the accuracy of deduplication.

[0080] In one embodiment, step 104 above involves matching the small target images in the full set of targets at the time of target capture with the small target images in the currently traversed subset to obtain a matching result. This can be achieved in the following way: Based on the image information of the target small images in the full target set and the image information of the target small images in the currently traversed subset, a similarity distance is determined. The image information includes at least one of the contour information, color information, and texture information of the target in the corresponding target small image. The matching result is determined based on the similarity distance.

[0081] For example, for small target images accessed by the same camera device, the background changes are generally small, while the lighting and external scene of each small target image usually change gradually over time, and the changes are generally small at similar time points. Therefore, this invention uses at least one of the target's contour information, target's color information, and target's texture information to determine the similarity distance between two small target images. Taking the image information as contour information, color information, and texture information as an example, the specific method is as follows: For the contour information, Hu moments are used to calculate the distance, and normalization processing is performed to generate the distance. For color information, a target color histogram is used, and distance is calculated using chi-square distance, followed by normalization to generate the distance. For texture information, Euclidean distance is used for calculation, and the generated distance is normalized. .against , and The values ​​range from 0 to 1, with smaller numbers indicating greater similarity between the two target images. The methods for calculating these distances can be found in related technologies, and will not be elaborated upon here. Using the three types of image information, a weighted fusion is performed to generate the similarity distance. At similar distances The value is greater than When the sum of the weighted values ​​of the three types of image information is 1, it indicates that the two small target images represent the same target. The weighted values ​​of the contour information, color information and texture information are set to 0.3, 0.3 and 0.4 respectively.

[0082] It should be noted that when matching two target images, features can be extracted from each target image to obtain image features of each target image. Image features include semi-structured features and / or structured features. Semi-structured features are mainly used to characterize the comprehensive information of the target and to determine the uniqueness of the target, that is, to analyze whether the target in the two target images is the same person or the same vehicle, etc. Structured features are mainly attribute information that can be understood by humans, such as color information, vehicle style information, and clothing style information. Taking image features as semi-structured features as an example, when obtaining the semi-structured features of each target small image, the cosine distance, which is the similarity distance, is calculated based on every two semi-structured features. When the cosine distance is greater than a preset distance, the targets in the two target small images corresponding to the two semi-structured features are determined to be different targets; when the cosine distance is less than or equal to the preset distance, the targets in the two target small images corresponding to the two semi-structured features are determined to be the same target. Then, the image with the highest clarity among the two target small images can be retained, and the image with the lowest clarity among the two target small images can be removed. For the specific calculation of the cosine distance, please refer to relevant technologies. This invention will not elaborate further here.

[0083] It should be noted that, for different target types, semi-structured analysis methods and / or structured analysis methods corresponding to different target types can be used. For details, please refer to relevant technologies. This invention will not elaborate further here.

[0084] In this embodiment, the target images in the full set of targets at the time of target capture are matched with the target images in the corresponding subset at the previous capture time. Taking advantage of the fact that the target changes relatively little in the captured images at adjacent capture times, the similarity distance between the two target images is calculated by combining the target's contour information, target's color information, and target's texture information. Then, based on the similarity distance, it is determined whether the two target images belong to the same target, which improves the accuracy of the matching result determination.

[0085] In one embodiment, the first capture information includes a first target type, and the second capture information includes a second target type. The above-mentioned process of deduplicating all first target small images and all second target small images based on each of the first and second capture information to obtain the full set of targets at the time of capture can be implemented in the following way: For each target capture time, the overlap between each first target small image and each second target small image at the target capture time in the scene image is determined; based on the overlap, the first target type and the second target type, all first target small images and all second target small images at the target capture time are deduplicated to obtain the full set of targets.

[0086] For example, for each first target small image, based on the first position coordinates of the first target small image in the scene image and the second position coordinates of each second target small image in the scene image, the Intersection of Union (IOU) ratio of the first position coordinates and the second position coordinates is calculated. The IOU ratio is used to represent the degree of overlap. When the first target type of the first target small image and the second target type of the second target small image are the same, and the IOU ratio is greater than or equal to a preset threshold, it is determined that the targets represented by the first target small image and the second target small image belong to the same target. In this case, the second target small images belonging to the same target are removed, and the first target small images belonging to the same target are retained. When the first target type of the first target small image and the second target type of the second target small image are the same, and the IOU ratio is less than the preset threshold, it is determined that the targets represented by the first target small image and the second target small image do not belong to the same target. In this case, the second target small image is considered to be a newly detected image by the electronic device, that is, an image missed by the camera device during target detection or an image lost during transmission. Therefore, this second target small image needs to be retained. After comparing all the first target small images and all the second target small images, the complete set of targets of the same camera device at the target capture time is obtained. Figure 6 This is a schematic diagram of the target full set provided in an embodiment of the present invention, such as... Figure 6As shown, the full target set includes the capture time, access time, scene image, and data corresponding to each target. Each data entry includes a small target image retained after deduplication, the target type, a list of attributes of the small target image retained after deduplication (including attribute information obtained by feature detection using detection algorithms such as contour, color, and texture), and the target identifier of the small target image retained after deduplication. When the small target image retained after deduplication is an image in the detection data, the target identifier of the small target image retained after deduplication is the newly added target identifier; when the small target image retained after deduplication is an image in the original data, the target identifier of the small target image retained after deduplication is the key target identifier.

[0087] In this embodiment, based on the overlap, the first target type, and the second target type, all the first target small images and all the second target small images at the time of target capture are deduplicated, which further reduces the consumption of computing resources.

[0088] The target image deduplication device provided by the present invention is described below. The target image deduplication device described below can be referred to in correspondence with the target image deduplication method described above.

[0089] Figure 7 This is a schematic diagram of the target image deduplication device provided in an embodiment of the present invention, as shown below. Figure 7 As shown, the target image deduplication device 700 includes a determining unit 701, a splitting unit 702, a sorting unit 703, a matching unit 704, and a first deduplication unit 705; wherein: The determining unit 701 is used to traverse the data corresponding to each target in the initial deduplication set of the previous capture time of the camera device at the target capture time, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target. The splitting unit 702 is used to split the initial deduplication set of the previous capture time based on each of the time intervals to obtain the subsets corresponding to each of the time intervals. The sorting unit 703 is used to sort all the subsets in ascending order of the time intervals to obtain the sorted subsets. The matching unit 704 is used to traverse each subset in the sorted subset, match the small target images in the full set of targets at the target capture time with the small target images in the currently traversed subset, and obtain the matching result. The full set of targets includes the small target images corresponding to all targets captured at the target capture time. The first deduplication unit 705 is used to deduplicat the target small image based on the matching result.

[0090] The target image deduplication device provided by this invention traverses the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture device. Based on the target capture time and the capture times included in the target-corresponding data, a time interval is determined. Based on each time interval, the initial deduplication set of the previous capture time is split into subsets corresponding to each time interval. All subsets are sorted in ascending order of time intervals to obtain sorted subsets. Each subset in the sorted subsets is traversed, and the target small images in the target full set at the target capture time are matched with the target small images in the currently traversed subsets to obtain matching results. Based on the matching results, target small images are deduplicated. It can be seen that this invention sorts the split subsets in ascending order of time intervals. The size of each time interval reflects the proximity of the capture times. The subsets in the sorted subsets are matched with the target full set from near to far to remove duplicates, thereby reducing the consumption of computing resources in the entire deduplication process.

[0091] Based on any of the above embodiments, the target image deduplication device 700 further includes: The acquisition unit is used to acquire at least one piece of raw data sent by the camera device, the raw data including a scene image captured by the camera device, a small image of a first target, and first capture information corresponding to the small image of the first target; A detection unit is used to perform target detection based on each of the scene images to obtain at least one detection data, the detection data including a second target small image and second capture information corresponding to the second target small image; The second deduplication unit is used to deduplicatize all the first target small images and all the second target small images based on each of the first capture information and each of the second capture information, so as to obtain the full set of targets at the time of the target capture.

[0092] Based on any of the above embodiments, the first deduplication unit 705 is specifically used for: If the matching result indicates a successful match, the successfully matched target small image will be predicted as a target small image of the same target; Based on the small images of the target that are predicted to be the same target, deduplication is performed to obtain the initial deduplication set of the target capture time; The initial set of duplicate capture times of the target is deduplicated.

[0093] Based on any of the above embodiments, the target full set includes key target identifiers and newly added target identifiers. The key target identifiers are used to characterize that the corresponding image is an image in the original data, and the newly added target identifiers are used to characterize that the corresponding image is an image in the detection data; the first deduplication unit 705 is further specifically used for: If all targets predicted to be the same target are traversed, and the first image is identified as a key target while the second image is identified as a newly added target, then the second image is deleted. When both the identifier of the first image and the identifier of the second image are new target identifiers, the first image is an image in the full target set, and the second image is an image in a target subset, the image to be deleted is determined based on the attribute information of the first image and the attribute information of the second image, and the image to be deleted is deleted, and the target subset is the first subset in the sorted subset; If both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset, then delete the image with the smallest capture time between the first image and the second image. The non-target subset is the subset other than the first subset in the sorted subset. The full set of targets obtained from the final traversal and the initial deduplicated set of the previous capture time obtained from the final traversal are merged to obtain the initial deduplicated set of the target capture time.

[0094] Based on any of the above embodiments, the first deduplication unit 705 is further specifically used for: In the initial deduplication set at the target capture time, at least one third target small image that meets the target conditions is determined. Based on each of the third target small images, an image set is determined. Feature extraction is performed on each target small image in the image set to obtain the features of each image. The image set is deduplicated based on the image features described above.

[0095] The target condition includes one of the following: There are no images matching the third target small image in a preset number of historical capture times prior to the target capture time; There is no image matching the third target small image at the target capture time, and the difference between the current time and the access time of the third target small image is greater than the first preset time. The difference between the current time and the access time of the third target small image is greater than the second preset duration, and the second preset duration is greater than the first preset duration.

[0096] Based on any of the above embodiments, the first deduplication unit 705 is further specifically used for: For each of the aforementioned third target small images, if there is an associated identifier for the third target small image, the suspected target set associated with the third target small image and the associated identifier of the third target small image is added to the image set. The suspected target set includes the image with the largest capture time among the first image and the second image, provided that both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset.

[0097] Based on any of the above embodiments, the determining unit 701 is specifically used for: Determine the difference between the target capture time and the capture times included in the data corresponding to the target, and determine the minimum difference as the time interval.

[0098] Based on any of the above embodiments, the matching unit 704 is specifically used for: Based on the image information of the target small images in the full target set and the image information of the target small images in the currently traversed subset, the similarity distance is determined. The image information includes at least one of the contour information of the target in the corresponding target small image, the color information of the target, and the texture information of the target. The matching result is determined based on the similarity distance.

[0099] Based on any of the above embodiments, the first capture information includes a first target type, and the second capture information includes a second target type; the second deduplication unit is specifically used for: For each of the target capture times, determine the overlap between each first target small image and each second target small image at the target capture time in the scene image; Based on the overlap, the first target type, and the second target type, all the first target small images and all the second target small images at the time of target capture are deduplicated to obtain the full set of targets.

[0100] Figure 8 This is a schematic diagram of the physical structure of the electronic device provided in the embodiments of the present invention, such as... Figure 8As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communications bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communications bus 840. The processor 810 can call logic instructions in the memory 830 to execute a target image deduplication method. This method includes: traversing the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture device; determining time intervals based on the target capture time and the capture times included in the target-corresponding data; splitting the initial deduplication set of the previous capture time based on each time interval to obtain subsets corresponding to each time interval; sorting all subsets in ascending order of the time intervals to obtain sorted subsets; traversing each subset in the sorted subsets; matching the target small images in the full target set of the target capture time with the target small images in the currently traversed subsets to obtain a matching result, where the full target set includes the target small images corresponding to all targets captured at the target capture time; and performing target small image deduplication based on the matching result.

[0101] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0102] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the target image deduplication method provided by the above methods. The method includes: traversing the data corresponding to each target in the initial deduplication set of the previous capture time of the camera device at the target capture time; determining a time interval based on the target capture time and the capture times included in the target-corresponding data; splitting the initial deduplication set of the previous capture time based on each time interval to obtain a subset corresponding to each time interval; sorting all the subsets in ascending order of each time interval to obtain a sorted subset; traversing each subset in the sorted subset, matching the target small images in the target full set at the target capture time with the target small images in the currently traversed subset to obtain a matching result, wherein the target full set includes the target small images corresponding to all targets captured at the target capture time; and performing target small image deduplication based on the matching result.

[0103] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a target image deduplication method provided by the above methods. The method includes: traversing the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture time; determining a time interval based on the target capture time and the capture times included in the target-corresponding data; splitting the initial deduplication set of the previous capture time based on each time interval to obtain a subset corresponding to each time interval; sorting all the subsets in ascending order of the time intervals to obtain a sorted subset; traversing each subset in the sorted subset, matching the target small images in the target full set at the target capture time with the target small images in the currently traversed subset to obtain a matching result, wherein the target full set includes the target small images corresponding to all targets captured at the target capture time; and performing target small image deduplication based on the matching result.

[0104] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for deduplicating target images, characterized in that, include: Iterate through the data corresponding to each target in the initial deduplication set of the previous capture time of the target captured by the camera device, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target. Based on each of the time intervals, the initial deduplication set of the previous capture time is split to obtain the subset corresponding to each of the time intervals; Sort all the subsets in ascending order of the time intervals to obtain the sorted subsets; Traverse each subset in the sorted subset, and match the small target images in the full set of targets at the target capture time with the small target images in the currently traversed subset to obtain the matching result. The full set of targets includes the small target images corresponding to all targets captured at the target capture time. Based on the matching results, duplicate target small images are removed.

2. The method for deduplicating target images according to claim 1, characterized in that, The method further includes: Acquire at least one piece of raw data sent by the camera device, the raw data including a scene image captured by the camera device, a small image of a first target, and first capture information corresponding to the small image of the first target; Target detection is performed based on each of the scene images to obtain at least one detection data, the detection data including a second target small image and second capture information corresponding to the second target small image; Based on each of the first capture information and each of the second capture information, duplicates are removed from all the first target small images and all the second target small images to obtain the full set of targets at the capture time.

3. The method for deduplicating target images according to claim 2, characterized in that, The process of deduplicating the target small image based on the matching result includes: If the matching result indicates a successful match, the successfully matched target small image will be predicted as a target small image of the same target; Based on the small images of the target that are predicted to be the same target, deduplication is performed to obtain the initial deduplication set of the target capture time; The initial set of duplicate capture times of the target is deduplicated.

4. The target image deduplication method according to claim 3, characterized in that, The full set of targets includes key target identifiers and newly added target identifiers. The key target identifiers are used to indicate that the corresponding image is an image in the original data, and the newly added target identifiers are used to indicate that the corresponding image is an image in the detection data. The process of deduplicating small images of the target predicted to be the same target to obtain the initial deduplicated set at the time of target capture includes: If all targets predicted to be the same target are traversed, and the first image is identified as a key target while the second image is identified as a newly added target, then the second image is deleted. When both the identifier of the first image and the identifier of the second image are new target identifiers, the first image is an image in the full target set, and the second image is an image in a target subset, the image to be deleted is determined based on the attribute information of the first image and the attribute information of the second image, and the image to be deleted is deleted, and the target subset is the first subset in the sorted subset; If both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full target set, and the second image is an image in a non-target subset, then delete the image with the smallest capture time between the first image and the second image. The non-target subset is the subset other than the first subset in the sorted subset. The full set of targets obtained from the final traversal and the initial deduplicated set of the previous capture time obtained from the final traversal are merged to obtain the initial deduplicated set of the target capture time.

5. The target image deduplication method according to claim 4, characterized in that, The initial deduplication of the target capture time set includes: In the initial deduplication set at the target capture time, at least one third target small image that meets the target conditions is determined, and an image set is determined based on each of the third target small images; Feature extraction is performed on each target small image in the image set to obtain the features of each image; The image set is deduplicated based on the image features described above; The target condition includes one of the following: There are no images matching the third target small image in a preset number of historical capture times prior to the target capture time; There is no image matching the third target small image at the target capture time, and the difference between the current time and the access time of the third target small image is greater than the first preset time. The difference between the current time and the access time of the third target small image is greater than the second preset duration, and the second preset duration is greater than the first preset duration.

6. The method for deduplicating target images according to claim 5, characterized in that, The process of determining the image set based on each of the third target small images includes: For each of the aforementioned third target small images, if there is an associated identifier for the third target small image, the suspected target set associated with the third target small image and the associated identifier of the third target small image is added to the image set. The suspected target set includes the image with the largest capture time among the first image and the second image, provided that both the identifier of the first image and the identifier of the second image are newly added target identifiers, the first image is an image in the full set of targets, and the second image is an image in a non-target subset.

7. The method for deduplicating target images according to any one of claims 1-6, characterized in that, The step of determining the time interval based on the target capture time and the capture times included in the data corresponding to the target includes: Determine the difference between the target capture time and the capture times included in the data corresponding to the target, and determine the minimum difference as the time interval.

8. The method for deduplicating target images according to any one of claims 1-6, characterized in that, The step of matching the small target images in the full set of targets at the time of target capture with the small target images in the currently traversed subset to obtain the matching result includes: Based on the image information of the target small images in the full target set and the image information of the target small images in the currently traversed subset, the similarity distance is determined. The image information includes at least one of the contour information of the target in the corresponding target small image, the color information of the target, and the texture information of the target. The matching result is determined based on the similarity distance.

9. The method for deduplicating target images according to any one of claims 2-6, characterized in that, The first capture information includes a first target type, and the second capture information includes a second target type; The step of deduplicating all first target small images and all second target small images based on each of the first capture information and each of the second capture information to obtain the full set of targets at the time of target capture includes: For each of the target capture times, determine the overlap between each first target small image and each second target small image at the target capture time in the scene image; Based on the overlap, the first target type, and the second target type, all the first target small images and all the second target small images at the time of target capture are deduplicated to obtain the full set of targets.

10. A target image deduplication device, characterized in that, include: The determining unit is used to traverse the data corresponding to each target in the initial deduplication set of the previous capture time of the target capture time of the camera device, and determine the time interval based on the target capture time and the capture times included in the data corresponding to the target. The splitting unit is used to split the initial deduplication set of the previous capture time based on each of the time intervals to obtain the subsets corresponding to each of the time intervals. A sorting unit is used to sort all the subsets in ascending order of the time intervals to obtain sorted subsets; The matching unit is used to traverse each subset in the sorted subset, match the small target images in the full set of targets at the target capture time with the small target images in the currently traversed subset, and obtain the matching result. The full set of targets includes the small target images corresponding to all targets captured at the target capture time. The first deduplication unit is used to deduplicat the target small image based on the matching result.

Images