Data compatible processing method based on smart campus and internet of things cloud platform

Abstract

This invention provides a data compatibility processing method and IoT cloud platform based on smart campuses, relating to the fields of smart campus and data processing technology. In this invention, multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices are subjected to data compatibility conversion operations to form multiple corresponding target campus IoT surveillance videos; based on the corresponding video frame timestamp information, video frame stitching processing is performed on the target campus IoT surveillance video frames included in the multiple target campus IoT surveillance videos to form a corresponding stitched campus IoT surveillance video; using a video recognition neural network formed through network optimization, video recognition processing is performed on the stitched campus IoT surveillance video to output the target video recognition result corresponding to the stitched campus IoT surveillance video. Based on the above, the problem of low efficiency in data compatibility processing in existing technologies can be improved.

Description

Technical Field

[0001] This invention relates to the field of smart campus and data processing technology, and more specifically, to a data compatibility processing method and an Internet of Things cloud platform based on a smart campus. Background Technology

[0002] A smart campus refers to a smart learning environment aimed at promoting the integration of information technology and education, improving teaching and learning effectiveness, and utilizing new technologies such as the Internet of Things, cloud computing, and big data analytics as core technologies. It provides a comprehensive, intelligent, data-driven, networked, and collaborative integrated system for teaching, research, management, and daily life services, capable of providing insights and predictions regarding education and teaching management. A smart campus consists of: one data center + smart campus infrastructure + eight types of smart campus application systems + smart resources. The eight types of smart campus application systems are: student development smart application systems, teacher professional development smart application systems, scientific research smart application systems, education management smart application systems, security monitoring smart application systems, logistics service smart application systems, social service smart application systems, and comprehensive evaluation smart application systems.

[0003] The Internet of Things (IoT) refers to a network that connects any object to the internet through information sensing devices and according to agreed-upon protocols, enabling information exchange and communication to achieve intelligent identification, location, tracking, monitoring, and management. In simple terms, the IoT is the "internet of interconnected things," encompassing two meanings: First, the IoT is an extension and expansion of the internet, with the internet remaining its core and foundation; second, the user end of the IoT includes not only people but also objects, enabling the exchange and communication of information between people and objects, as well as between objects themselves.

[0004] In existing technologies, data collected from different IoT terminal devices is typically processed separately, which results in low efficiency in data compatibility processing. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a data compatibility processing method and an Internet of Things cloud platform based on smart campuses, so as to improve the problem of low efficiency in data compatibility processing in the prior art.

[0006] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0007] A data compatibility processing method based on smart campuses, applied to an IoT cloud platform, includes:

[0008] Data compatibility conversion is performed on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple target campus IoT surveillance videos, which have the same video format.

[0009] Based on the video frame timestamp information corresponding to each frame of the target campus IoT monitoring video included in each of the target campus IoT monitoring videos, video frame splicing processing is performed on the target campus IoT monitoring video frames included in the multiple target campus IoT monitoring videos to form a corresponding spliced ​​campus IoT monitoring video. The spliced ​​campus IoT monitoring video includes multiple spliced ​​campus IoT monitoring video frames, and each spliced ​​campus IoT monitoring video frame is formed by splicing multiple target campus IoT monitoring video frames with the same video frame timestamp information.

[0010] A video recognition neural network, formed through network optimization, is used to perform video recognition processing on the spliced ​​campus IoT surveillance video to output the target video recognition result corresponding to the spliced ​​campus IoT surveillance video. The target video recognition result is used to reflect the campus security status of the smart campus area corresponding to the spliced ​​campus IoT surveillance video.

[0011] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of performing data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple corresponding target campus IoT surveillance videos includes:

[0012] The synchronous data acquisition instruction is sent to each of the multiple campus IoT terminal devices connected by communication, so that each campus IoT terminal device synchronously acquires data from the corresponding smart campus sub-area according to the synchronous data acquisition instruction, so as to form multiple synchronized campus IoT monitoring videos. Any two campus IoT monitoring video frames with the same video frame time sequence have the same video frame timestamp information.

[0013] Data compatibility conversion is performed on the multiple campus IoT surveillance videos to form multiple corresponding target campus IoT surveillance videos.

[0014] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of performing video frame splicing processing on the target campus IoT monitoring video frames included in the plurality of target campus IoT monitoring videos, based on the video frame timestamp information corresponding to each frame of the target campus IoT monitoring video, to form a corresponding spliced ​​campus IoT monitoring video, includes:

[0015] For each video frame timestamp, extract each target campus IoT monitoring video frame with the video frame timestamp from the multiple target campus IoT monitoring videos to form a set of video frames to be spliced ​​corresponding to the video frame timestamp.

[0016] For each video frame timestamp, based on the positional relationship between the corresponding smart campus sub-regions, the target campus IoT monitoring video frames included in the set of video frames to be spliced ​​corresponding to the video frame timestamp are spliced ​​to form a corresponding spliced ​​campus IoT monitoring video frame. Then, the spliced ​​campus IoT monitoring video frames are combined according to the corresponding video frame timestamp information to form a corresponding spliced ​​campus IoT monitoring video.

[0017] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of using a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, and outputting the target video recognition result corresponding to the stitched campus IoT surveillance video, includes:

[0018] The video feature mining subnetwork, which is part of the video recognition neural network formed through network optimization, is used to perform video feature mining on the spliced ​​campus IoT surveillance video, and outputs the video feature data mining results corresponding to the spliced ​​campus IoT surveillance video.

[0019] The video object feature mining subnetwork included in the video recognition neural network is used to perform video object feature mining operations on the video object information included in the spliced ​​campus IoT monitoring video and the pre-configured video object information set, and output the video object feature data mining results corresponding to the spliced ​​campus IoT monitoring video;

[0020] The video segment feature mining subnetwork included in the video recognition neural network is used to perform video segment feature mining operations on the spliced ​​campus IoT monitoring video and the matching video segments included in the pre-configured set of matching video segments, and output the video segment feature data mining results corresponding to the spliced ​​campus IoT monitoring video.

[0021] The video feature data mining results, the video object feature data mining results, and the video segment feature data mining results are aggregated to form the aggregated data mining results corresponding to the spliced ​​campus IoT surveillance video;

[0022] The aggregated data mining results are loaded into the video recognition subnetwork included in the video recognition neural network, so as to use the video recognition subnetwork to identify the target video recognition result corresponding to the spliced ​​campus IoT surveillance video.

[0023] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of using a video feature mining subnetwork included in a video recognition neural network formed through network optimization to perform video feature mining operations on the stitched campus IoT surveillance video and outputting the video feature data mining results corresponding to the stitched campus IoT surveillance video includes:

[0024] The spliced ​​campus IoT surveillance video is segmented to form spliced ​​campus IoT surveillance video segments. Then, the spliced ​​campus IoT surveillance video segments are mapped to output the video segment mapping results corresponding to the spliced ​​campus IoT surveillance video segments.

[0025] The video segment distribution information of the spliced ​​campus IoT surveillance video segments is analyzed, and then the video segment distribution information is mapped to output the video segment distribution information mapping result corresponding to the video segment distribution information.

[0026] The analysis outputs the video segment segment identification information mapping result corresponding to the spliced ​​campus IoT monitoring video segment. Then, the video segment mapping result, the video segment distribution information mapping result, and the video segment segment identification information mapping result are aggregated to form the aggregated information mapping result corresponding to the spliced ​​campus IoT monitoring video segment.

[0027] The aggregated information mapping result is loaded into the video feature mining subnetwork of the video recognition neural network formed by network optimization, so as to use the video feature mining subnetwork to perform video feature mining operation on the aggregated information mapping result, output the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video segment, and then analyze and output the video feature data mining result corresponding to the spliced ​​campus IoT monitoring video based on the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video.

[0028] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of utilizing the video object feature mining subnetwork included in the video recognition neural network to perform video object feature mining operations on the video object information included in the spliced ​​campus IoT surveillance video and the pre-configured video object information set, and outputting the video object feature data mining results corresponding to the spliced ​​campus IoT surveillance video, includes:

[0029] An object comparison operation is performed on the spliced ​​campus IoT surveillance video and the video object information included in the pre-configured video object information set to output the object comparison result corresponding to the spliced ​​campus IoT surveillance video;

[0030] If the object comparison result reflects that the spliced ​​campus IoT surveillance video contains video object information included in the video object information set, the video object information contained in the spliced ​​campus IoT surveillance video is marked to form corresponding video object information to be processed.

[0031] The video object information to be processed is loaded into the video object feature mining sub-network included in the video recognition neural network, so as to use the video object feature mining sub-network to perform video object feature mining operation on the video object information to be processed, and output the video object information feature distribution corresponding to the video object information to be processed.

[0032] Based on the distribution of video object information features, the video object feature data mining results corresponding to the spliced ​​campus IoT surveillance video are analyzed and output.

[0033] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of utilizing the video segment feature mining subnetwork included in the video recognition neural network to perform video segment feature mining operations on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured matching video segment set, and outputting the video segment feature data mining results corresponding to the spliced ​​campus IoT surveillance video, includes:

[0034] A video segment comparison operation is performed on the spliced ​​campus IoT surveillance video and the pre-configured set of matching video segments to output the video segment comparison result corresponding to the spliced ​​campus IoT surveillance video;

[0035] If the video segment comparison results show that there is a matching video segment in the matching video segment set that matches the spliced ​​campus IoT monitoring video, then the matching video segment in the matching video segment set that matches the spliced ​​campus IoT monitoring video is marked to form a corresponding matching video segment to be processed.

[0036] The video segment to be matched is loaded into the video segment feature mining sub-network included in the video recognition neural network, so as to use the video segment feature mining sub-network to perform video segment feature mining operation on the video segment to be matched and output the matching video segment feature distribution corresponding to the video segment to be matched.

[0037] Based on the feature distribution of the matched video segments, the feature data mining results of the video segments corresponding to the spliced ​​campus IoT surveillance video are analyzed and output.

[0038] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of using a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, and outputting the target video recognition result corresponding to the stitched campus IoT surveillance video, further includes:

[0039] The extracted example spliced ​​campus IoT surveillance video and the example video annotation results corresponding to the example spliced ​​campus IoT surveillance video are used to reflect the real campus security status corresponding to the example spliced ​​campus IoT surveillance video;

[0040] The video feature mining subnetwork, which is included in the video recognition neural network to be optimized, is used to perform video feature mining on the example spliced ​​campus IoT surveillance video, and outputs the example video feature data mining results corresponding to the example spliced ​​campus IoT surveillance video.

[0041] Using the video object feature mining subnetwork included in the video recognition neural network to be optimized, video object feature mining is performed on the video object information included in the example spliced ​​campus IoT surveillance video and the pre-configured video object information set, and the example video object feature data mining results corresponding to the example spliced ​​campus IoT surveillance video are output.

[0042] The video segment feature mining subnetwork, which is included in the video recognition neural network to be optimized, is used to perform video segment feature mining operation on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured set of matching video segments, and outputs the example video segment feature data mining result corresponding to the example spliced ​​campus IoT surveillance video.

[0043] Based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, a network optimization operation is performed on the video recognition neural network to be optimized to form the video recognition neural network corresponding to the video recognition neural network to be optimized.

[0044] In some preferred embodiments, in the above-described data compatibility processing method based on smart campuses, the step of performing network optimization on the video recognition neural network to be optimized, based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, to form the video recognition neural network corresponding to the video recognition neural network to be optimized, includes:

[0045] The example video feature data mining results, the example video object feature data mining results, and the example video segment feature data mining results are aggregated to form the example aggregated data mining results of the example spliced ​​campus IoT surveillance video. The example aggregated data mining results are then loaded into the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, so as to use the video recognition sub-network to be optimized to identify the example video recognition results corresponding to the example spliced ​​campus IoT surveillance video.

[0046] Based on the example video recognition results and the example video annotation results, the network optimization cost corresponding to the neural network for the video recognition to be optimized is analyzed and output;

[0047] If the network optimization cost value corresponding to the video recognition neural network to be optimized is greater than or equal to the pre-configured network optimization cost reference value, the network optimization operation of the video recognition neural network to be optimized is performed according to the network optimization cost value.

[0048] The video recognition neural network to be optimized after network optimization is marked as a candidate video recognition neural network. Then, network optimization is performed on the candidate video recognition neural network. If the network optimization cost corresponding to the candidate video recognition neural network after network optimization is less than the network optimization cost reference value, the current candidate video recognition neural network is marked as a video recognition neural network.

[0049] This invention also provides an Internet of Things (IoT) cloud platform, including a processor and a memory. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned data compatibility processing method based on a smart campus.

[0050] This invention provides a data compatibility processing method and IoT cloud platform based on a smart campus. The method performs data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple target campus IoT surveillance videos. Based on the corresponding video frame timestamp information, it performs video frame stitching processing on the target campus IoT surveillance video frames included in the multiple target campus IoT surveillance videos to form a corresponding stitched campus IoT surveillance video. Using a video recognition neural network formed through network optimization, it performs video recognition processing on the stitched campus IoT surveillance video to output the target video recognition result corresponding to the stitched campus IoT surveillance video. Based on the foregoing, by first stitching video frames and then performing recognition processing on the stitched video, compared to conventional technical solutions that perform recognition processing separately, it can improve the low efficiency problem of data compatibility processing in existing technologies to a certain extent.

[0051] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0052] Figure 1 This is a structural block diagram of an IoT cloud platform provided in an embodiment of the present invention.

[0053] Figure 2 This is a flowchart illustrating the steps of the data compatibility processing method based on a smart campus provided in this embodiment of the invention.

[0054] Figure 3 This is a schematic diagram of the modules included in the data compatibility processing device based on a smart campus provided in an embodiment of the present invention. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0056] like Figure 1As shown, this embodiment of the invention provides an Internet of Things (IoT) cloud platform. The IoT cloud platform may include a memory and a processor.

[0057] In detail, the memory and the processor are electrically connected directly or indirectly to enable data transmission or interaction. For example, they can be electrically connected via one or more communication buses or signal lines. The memory may store at least one software functional module (computer program) that exists in the form of software or firmware. The processor can be used to execute the executable computer program stored in the memory, thereby implementing the data compatibility processing method based on a smart campus provided in this embodiment of the invention.

[0058] For example, in some embodiments, the memory may be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The processor may be a general-purpose processor, including a Central Processing Unit (CPU), Network Processor (NP), System on Chip (SoC), etc.; it may also be a Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0059] For example, in some implementations, the IoT cloud platform may be a server (or server cluster) with data processing capabilities.

[0060] Combination Figure 2 This invention also provides a data compatibility processing method based on a smart campus, which can be applied to the aforementioned IoT cloud platform. The method steps defined in the relevant process of the data compatibility processing method based on a smart campus can be implemented by the IoT cloud platform.

[0061] The following will be about Figure 2 The specific process shown will be explained in detail.

[0062] Step S110: Perform data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple corresponding target campus IoT surveillance videos.

[0063] In this embodiment of the invention, the IoT cloud platform can perform data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple corresponding target campus IoT surveillance videos. This ensures that the multiple target campus IoT surveillance videos have the same video format (the specific video format is not limited).

[0064] Step S120: Based on the video frame timestamp information corresponding to each frame of the target campus IoT monitoring video included in each target campus IoT monitoring video, video frame splicing processing is performed on the target campus IoT monitoring video frames included in the multiple target campus IoT monitoring videos to form a corresponding spliced ​​campus IoT monitoring video.

[0065] In this embodiment of the invention, the IoT cloud platform can perform video frame splicing processing on the target campus IoT monitoring video frames included in the multiple target campus IoT monitoring videos, based on the video frame timestamp information corresponding to each frame of the target campus IoT monitoring video, to form a corresponding spliced ​​campus IoT monitoring video. The spliced ​​campus IoT monitoring video includes multiple spliced ​​campus IoT monitoring video frames, and each spliced ​​campus IoT monitoring video frame is formed by splicing multiple target campus IoT monitoring video frames with the same video frame timestamp information.

[0066] Step S130: Using a video recognition neural network formed through network optimization, the spliced ​​campus IoT surveillance video is processed for video recognition to output the target video recognition result corresponding to the spliced ​​campus IoT surveillance video.

[0067] In this embodiment of the invention, the IoT cloud platform can utilize a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, so as to output the target video recognition result corresponding to the stitched campus IoT surveillance video. The target video recognition result is used to reflect the campus security status (such as a security level value, or both secure and unsecured results) of the smart campus area corresponding to the stitched campus IoT surveillance video.

[0068] Based on the foregoing, by first stitching video frames together and then performing recognition processing on the stitched video, compared to conventional technical solutions that perform recognition processing separately, the problem of low efficiency in data compatibility processing in existing technologies can be improved to some extent.

[0069] For example, in some implementations, step S110 above may specifically include the following sub-steps:

[0070] A synchronous data acquisition command is sent to each of the multiple campus IoT terminal devices connected in a communication link. This enables each campus IoT terminal device to synchronously acquire data from the corresponding smart campus sub-area according to the synchronous data acquisition command, thereby forming multiple synchronized campus IoT monitoring videos. Any two campus IoT monitoring video frames with the same video frame timing sequence have the same video frame timestamp information (i.e., the video frame timestamp information corresponding to the first campus IoT monitoring video frame in each campus IoT monitoring video is the same, the video frame timestamp information corresponding to the second campus IoT monitoring video frame in each campus IoT monitoring video is the same, the video frame timestamp information corresponding to the third campus IoT monitoring video frame in each campus IoT monitoring video is the same, the video frame timestamp information corresponding to the fourth campus IoT monitoring video frame in each campus IoT monitoring video is the same, etc.).

[0071] Data compatibility conversion is performed on the multiple campus IoT surveillance videos (the specific data compatibility conversion method is not limited, as long as the format of the multiple target campus IoT surveillance videos is unified) to form multiple corresponding target campus IoT surveillance videos.

[0072] For example, in some implementations, step S120 above may specifically include the following sub-steps:

[0073] For each video frame timestamp, extract each target campus IoT monitoring video frame with the video frame timestamp from the multiple target campus IoT monitoring videos to form a set of video frames to be spliced ​​corresponding to the video frame timestamp.

[0074] For each video frame timestamp, based on the positional relationship between the corresponding smart campus sub-regions, the target campus IoT monitoring video frames included in the set of video frames to be stitched according to the timestamp information are processed to form a corresponding stitched campus IoT monitoring video frame (that is, if smart campus sub-region A is to the left of smart campus sub-region B, then the target campus IoT monitoring video frame corresponding to smart campus sub-region A is also to the left of the target campus IoT monitoring video frame corresponding to smart campus sub-region B). Then, the stitched campus IoT monitoring video frames are combined according to the corresponding timestamp information (e.g., sorted from earliest to latest timestamp) to form a corresponding stitched campus IoT monitoring video.

[0075] For example, in some implementations, step S130 above may specifically include the following sub-steps:

[0076] The video feature mining subnetwork, which is part of the video recognition neural network formed through network optimization, is used to perform video feature mining on the spliced ​​campus IoT surveillance video, and outputs the video feature data mining results corresponding to the spliced ​​campus IoT surveillance video.

[0077] The video object feature mining subnetwork included in the video recognition neural network is used to perform video object feature mining operations on the video object information included in the spliced ​​campus IoT monitoring video and the pre-configured video object information set, and output the video object feature data mining results corresponding to the spliced ​​campus IoT monitoring video;

[0078] The video segment feature mining subnetwork included in the video recognition neural network is used to perform video segment feature mining operations on the spliced ​​campus IoT monitoring video and the matching video segments included in the pre-configured set of matching video segments, and output the video segment feature data mining results corresponding to the spliced ​​campus IoT monitoring video.

[0079] The video feature data mining results, the video object feature data mining results, and the video segment feature data mining results are aggregated to form the aggregated data mining results corresponding to the spliced ​​campus IoT surveillance video. (Example: aggregating the video feature data mining results, the video object feature data mining results, and the video segment feature data mining results can mean superimposing or splicing the video feature data mining results, the video object feature data mining results, and the video segment feature data mining results; superimposing the video feature data mining results, the video object feature data mining results, and the video segment feature data mining results can be done by directly superimposing them when the dimensions and sizes of each data mining result are the same, or it can be done by first scaling the dimensions and sizes of each data mining result to make the dimensions and sizes the same before superimposing them; in addition, splicing the video feature data mining results, the video object feature data mining results, and the video segment feature data mining results can be done in a certain order or arbitrarily.)

[0080] The aggregated data mining results are loaded into the video recognition subnetwork of the video recognition neural network to identify the target video recognition result corresponding to the spliced ​​campus IoT surveillance video. (For example, when the aggregated data mining results are loaded into the video recognition subnetwork of the video recognition neural network, the video recognition subnetwork can determine the probability that the spliced ​​campus IoT surveillance video is in a safe state, i.e., the probability corresponding to the spliced ​​campus IoT surveillance video. Based on this probability, the target safety state corresponding to the spliced ​​campus IoT surveillance video can be determined, i.e., the target video recognition result. For example, when the probability is greater than the probability threshold, the target video recognition result is marked as safe; when the probability is less than or equal to the probability threshold, the target video recognition result can be marked as unsafe. In addition, the video recognition subnetwork may include a softmax function to calculate and output the corresponding probability.)

[0081] For example, in some implementations, the step of using the video feature mining sub-network of the video recognition neural network formed through network optimization to perform video feature mining operations on the stitched campus IoT surveillance video and outputting the video feature data mining results corresponding to the stitched campus IoT surveillance video may specifically include the following implementable sub-steps:

[0082] The spliced ​​campus IoT surveillance video is segmented to form spliced ​​campus IoT surveillance video segments (for example, the video segmentation operation can be the process of dividing and combining multiple spliced ​​campus IoT surveillance video frames included in the spliced ​​campus IoT surveillance video into segments; for example, two spliced ​​campus IoT surveillance video frames with low similarity can be assigned to two spliced ​​campus IoT surveillance video segments). Then, a video segment mapping operation is performed on the spliced ​​campus IoT surveillance video segments to output the video segment mapping result corresponding to the spliced ​​campus IoT surveillance video segments (for example, the video segment mapping operation can refer to encoding the spliced ​​campus IoT surveillance video segments through an encoding network to form the corresponding video segment mapping result).

[0083] The distribution information of the video segments in the spliced ​​campus IoT surveillance video is analyzed, and then a distribution information mapping operation is performed on the video segment distribution information to output the video segment distribution information mapping result corresponding to the video segment distribution information (for example, the distribution information mapping operation may refer to encoding the video segment distribution information through an encoding network to form the corresponding video segment distribution information mapping result, wherein the video segment distribution information may refer to the video frame timing of the spliced ​​campus IoT surveillance video segment in the spliced ​​campus IoT surveillance video).

[0084] The system analyzes and outputs the video segment segment identification information mapping results corresponding to the spliced ​​campus IoT monitoring video segments (for example, the video segment segment identification information mapping results can be obtained by encoding the video segment segment identification information corresponding to the spliced ​​campus IoT monitoring video segments through an encoding network, and the video segment segment identification information is used to identify the corresponding spliced ​​campus IoT monitoring video segments in the spliced ​​campus IoT monitoring video). Then, the video segment mapping results, the video segment distribution information mapping results, and the video segment segment identification information mapping results are aggregated to form the aggregated information mapping results corresponding to the spliced ​​campus IoT monitoring video segments (for example, the video segment mapping...). The result aggregation operation of the video segment distribution information mapping result and the video segment segment identification information mapping result can refer to adding the video segment mapping result, the video segment distribution information mapping result, and the video segment segment identification information mapping result together, or performing a weighted summation, etc.; Alternatively, after forming the video segment mapping result and the video segment distribution information mapping result corresponding to the spliced ​​campus IoT monitoring video segment, it is also possible to aggregate only the video segment mapping result and the video segment distribution information mapping result, that is, without needing the video segment segment identification information mapping result, to output the aggregated information mapping result corresponding to the spliced ​​campus IoT monitoring video segment.

[0085] The aggregated information mapping result is loaded into the video feature mining subnetwork of the video recognition neural network formed through network optimization. The video feature mining subnetwork is used to perform video feature mining on the aggregated information mapping result, and outputs the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video segment. Then, based on the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video segment, the video feature data mining result corresponding to the spliced ​​campus IoT monitoring video is analyzed and output. (For example, if the number of spliced ​​campus IoT monitoring video segments is greater than a pre-configured segment number threshold, the spliced ​​campus IoT monitoring video segments can be processed in batches. The aggregated information mapping result corresponding to the first batch of spliced ​​campus IoT monitoring video segments is loaded into the video feature mining subnetwork, and then the aggregated information mapping result corresponding to the second batch of spliced ​​campus IoT monitoring video segments is loaded into the video feature mining subnetwork, and so on. The number of each batch of spliced ​​campus IoT monitoring video segments is equal to the segment number threshold.)

[0086] For example, in some implementations, the step of loading the aggregated information mapping result into the video feature mining sub-network of the video recognition neural network formed through network optimization, so as to use the video feature mining sub-network to perform video feature mining operations on the aggregated information mapping result, output the initial video feature data mining result corresponding to the stitched campus IoT monitoring video segment, and then analyzing and outputting the video feature data mining result corresponding to the stitched campus IoT monitoring video based on the initial video feature data mining result corresponding to the stitched campus IoT monitoring video segment, may specifically include the following implementable sub-steps:

[0087] The aggregated information mapping result is loaded into the video feature mining sub-network of the video recognition neural network formed by network optimization, so as to use the internal focusing network model included in the video feature mining sub-network to perform key data mining operations on the aggregated information mapping result and output the initial background data mining result corresponding to the aggregated information mapping result (for example, in this embodiment, the background refers to the information recorded by the memory cells at the current time step).

[0088] The aggregated information mapping result and the initial background data mining result are loaded into the primary background stabilization network model included in the video feature mining sub-network. The primary background stabilization network model is then used to perform regression prediction error analysis on the aggregated information mapping result and the initial background data mining result, outputting the corresponding primary regression prediction error analysis result. Then, the primary regression prediction error analysis result is subjected to data centering operation to output the primary data centering feature distribution corresponding to the aggregated information mapping result (for example, the primary background stabilization network model may include a regression prediction error analysis sub-model and a data centering sub-model. The regression prediction error analysis sub-model can reduce the attenuation of information during propagation, and the data centering sub-model can prevent the value from being too large or too small due to the position being too large or too small; in addition, the regression prediction error analysis operation can be used to analyze the difference between the predicted value and the observed value, and the data centering operation can make the data distribution have a mean of 0 and a variance of 1).

[0089] The primary data-driven feature distribution is loaded into the unidirectional data-driven feature mining network model included in the video feature mining sub-network, so as to use the unidirectional data-driven feature mining network model to perform data feature mining operations on the primary data-driven feature distribution and output the intermediate background data mining results corresponding to the primary data-driven feature distribution (for example, the unidirectional data-driven feature mining network model can be an artificial neural network, in which each neuron starts from the input layer, receives the input of the previous level, and inputs it to the next level, until the output layer).

[0090] The primary data-driven feature distribution and the intermediate background data mining results are loaded into the intermediate background stable network model included in the video feature mining sub-network. The intermediate background stable network model is then used to perform regression prediction error analysis on the primary data-driven feature distribution and the intermediate background data mining results, and the corresponding intermediate regression prediction error analysis results are output. The intermediate regression prediction error analysis results are then subjected to data-driven operation to output the intermediate data-driven feature distribution corresponding to the aggregated information mapping results.

[0091] Based on the intermediate data center's digitized feature distribution analysis, the initial video feature data mining results corresponding to the spliced ​​campus IoT surveillance video segments are output. Then, based on the initial video feature data mining results corresponding to the spliced ​​campus IoT surveillance video segments, the video feature data mining results corresponding to the spliced ​​campus IoT surveillance video are analyzed and output. (For example, the intermediate data center's digitized feature distribution can be directly used as the initial video feature data mining results; alternatively, the initial video feature data mining results corresponding to each spliced ​​campus IoT surveillance video segment can be merged, such as splicing or overlaying, to form the video feature data mining results corresponding to the spliced ​​campus IoT surveillance video.)

[0092] For example, in one instance, aggregated information mapping results A1, A2, ..., An can be loaded into a video feature mining sub-network. Using this sub-network, initial video feature data mining results B1, B2, ..., Bn can be obtained. Specifically, the initial video feature data mining result corresponding to aggregated information mapping result A1 can be initial video feature data mining result B1, the initial video feature data mining result corresponding to aggregated information mapping result A2 can be initial video feature data mining result B2, ..., and the initial video feature data mining result corresponding to aggregated information mapping result An can be initial video feature data mining result Bn. Based on this, the aggregated information mapping results can reflect the inherent information of the stitched campus IoT surveillance video clip, and the initial video feature data mining results can reflect both the inherent information of the stitched campus IoT surveillance video clip and the enhanced information of the stitched campus IoT surveillance video. The aggregated information mapping results A1, A2, ..., An are loaded into the target encoding unit 5b in the video feature mining sub-network. The target encoding unit 5b can output the initial video feature data mining results P1, P2, ..., Pn. Specifically, the initial video feature data mining result corresponding to aggregated information mapping result A1 can be P1, the initial video feature data mining result corresponding to aggregated information mapping result A2 can be P2, ..., and the initial video feature data mining result corresponding to aggregated information mapping result An can be Pn. Further, the initial video feature data mining results P1, P2, ..., Pn are loaded into the encoding network included in the video feature mining sub-network for encoding processing to output the initial video feature data mining results B1, B2, ..., Bn. The initial video feature data mining results B1, B2, ..., Bn can be collectively referred to as the initial video feature data mining results. Based on these initial video feature data mining results, the video feature data mining results corresponding to the spliced ​​campus IoT surveillance videos can be determined. Furthermore, the output of each encoding network in the video feature mining sub-network can also be collectively referred to as the initial video feature data mining results.Thus, the initial video feature data mining results B1 and P1 can be collectively referred to as the initial video feature data mining results corresponding to the aggregated information mapping result A1, the initial video feature data mining results B2 and P2 can be collectively referred to as the initial video feature data mining results corresponding to the aggregated information mapping result A2, and so on, the initial video feature data mining results Bn and Pn can be collectively referred to as the initial video feature data mining results corresponding to the aggregated information mapping result An.

[0093] For example, in some implementations, the step of loading the aggregated information mapping result into the video feature mining sub-network of the video recognition neural network formed through network optimization, and using the internal focusing network model included in the video feature mining sub-network to perform key data mining operations on the aggregated information mapping result and output the initial background data mining result corresponding to the aggregated information mapping result, may specifically include the following implementable sub-steps:

[0094] The aggregated information mapping result is loaded into the video feature mining sub-network of the video recognition neural network formed by network optimization;

[0095] The first internal focusing network sub-model is determined by analyzing the multiple internal focusing network sub-models of the internal focusing network model included in the video feature mining sub-network (for example, each internal focusing network sub-model can be used as the first internal focusing network sub-model in turn).

[0096] The primary feature integration sub-model included in the internal focusing network model is used to perform feature integration on the aggregated information mapping result, so as to output the first, second, and third feature integration representative information corresponding to the aggregated information mapping result. (For example, aggregated information mapping result 1, aggregated information mapping result 2, and aggregated information mapping result 3 can be loaded into the primary feature integration sub-model corresponding to the first internal focusing network sub-model, and the primary feature integration sub-model is used to perform linear mapping on the aggregated information mapping result to output the first, second, and third feature integration representative information corresponding to the aggregated information mapping result. Wherein, aggregated information mapping result 1, aggregated information mapping result 2, and aggregated information mapping result 3 can be the same, and the primary feature integration sub-model can output the first feature integration representative information corresponding to aggregated information mapping result 1, the second feature integration representative information corresponding to aggregated information mapping result 2, and the third feature integration representative information corresponding to aggregated information mapping result 3; additionally...) Aggregated information mapping results 1, 2, and 3 can be input into a primary feature integration sub-model corresponding to the first internal focusing network sub-model to perform the same linear mapping on them. A primary feature integration sub-model corresponding to the first internal focusing network sub-model can include primary feature integration sub-models F1, F2, and F3. Primary feature integration sub-model F1 performs a linear mapping on aggregated information mapping result 1, primary feature integration sub-model F2 performs a linear mapping on aggregated information mapping result 2, and primary feature integration sub-model F3 performs a linear mapping on aggregated information mapping result 3. In other words, primary feature integration sub-models F1, F2, and F3 can perform a linear mapping on the same aggregated information mapping result to output corresponding first feature integration representative information, second feature integration representative information, and third feature integration representative information.

[0097] The first feature integration representative information, the second feature integration representative information, and the third feature integration representative information are loaded into the first internal focusing network sub-model to perform information fusion operation on the first feature integration representative information, the second feature integration representative information, and the third feature integration representative information using the first internal focusing network sub-model, and output the information fusion feature distribution corresponding to the first internal focusing network sub-model (for example, the first feature integration representative information, the second feature integration representative information, and the third feature integration representative information can be matrix multiplied to output the corresponding information fusion feature distribution).

[0098] When each of the internal focusing network sub-models included in the internal focusing network model is analyzed and determined as the first internal focusing network sub-model so that it has the information fusion feature distribution corresponding to each internal focusing network sub-model, the feature distribution fusion sub-model included in the internal focusing network model is used to perform a feature distribution fusion operation on the information fusion feature distribution corresponding to each internal focusing network sub-model to form the information focusing fusion feature distribution corresponding to the aggregated information mapping result (for example, the feature distribution fusion operation on the information fusion feature distribution corresponding to each internal focusing network sub-model may refer to splicing the information fusion feature distribution corresponding to each internal focusing network sub-model).

[0099] The information-focused fusion feature distribution is loaded into the intermediate feature integration sub-model included in the internal focusing network model, so as to use the intermediate feature integration sub-model to perform feature integration operation on the information-focused fusion feature distribution (that is, to transform the information-focused fusion feature distribution into a one-dimensional feature distribution through the intermediate feature integration sub-model), thereby forming the initial background data mining result corresponding to the aggregated information mapping result.

[0100] For example, in some implementations, the step of using the video object feature mining sub-network included in the video recognition neural network to perform video object feature mining operations on the video object information included in the stitched campus IoT surveillance video and the pre-configured video object information set, and outputting the video object feature data mining results corresponding to the stitched campus IoT surveillance video, may specifically include the following implementable sub-steps:

[0101] An object comparison operation is performed on the spliced ​​campus IoT surveillance video and the video object information included in the pre-configured video object information set to output the object comparison result corresponding to the spliced ​​campus IoT surveillance video. If the object comparison result reflects that the spliced ​​campus IoT surveillance video contains video object information included in the video object information set (such as the same person, vehicle, etc.), the video object information in the spliced ​​campus IoT surveillance video is marked to form corresponding video object information to be processed. The video object information to be processed is loaded into the video object feature mining sub-network included in the video recognition neural network to utilize the video object features. The feature mining subnetwork performs video object feature mining operations on the video object information to be processed (for example, the video object feature mining subnetwork can be a coding network) to output the video object information feature distribution corresponding to the video object information to be processed; based on the video object information feature distribution, the video object feature data mining result corresponding to the spliced ​​campus IoT monitoring video is analyzed and output (for example, one of the video object information feature distributions can be directly used as the video object feature data mining result; or, the mean of multiple video object information feature distributions can be calculated to output the video object feature data mining result corresponding to the spliced ​​campus IoT monitoring video).

[0102] For example, in some implementations, the step of using the video segment feature mining sub-network included in the video recognition neural network to perform video segment feature mining operations on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured matching video segment set, and outputting the video segment feature data mining results corresponding to the spliced ​​campus IoT surveillance video, may specifically include the following implementable sub-steps:

[0103] A video segment comparison operation is performed on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured matching video segment set to output the video segment comparison result corresponding to the spliced ​​campus IoT surveillance video; if the video segment comparison result reflects that there is a matching video segment in the matching video segment set that matches the spliced ​​campus IoT surveillance video, the matching video segment in the matching video segment set that matches the spliced ​​campus IoT surveillance video is marked to form the corresponding matching video segment to be processed;

[0104] The video segment to be matched is loaded into the video segment feature mining sub-network included in the video recognition neural network, so as to use the video segment feature mining sub-network to perform video segment feature mining operation on the video segment to be matched (for example, the video segment feature mining sub-network can be an encoding network to perform encoding processing), and output the matching video segment feature distribution corresponding to the video segment to be matched;

[0105] Based on the feature distribution of the matched video segments, the feature data mining results of the video segments corresponding to the spliced ​​campus IoT monitoring video are analyzed and output (for example, one of the feature distributions of the matched video segments can be directly used as the feature data mining result of the video segments corresponding to the spliced ​​campus IoT monitoring video; or, the average of multiple feature distributions of the matched video segments can be calculated to obtain the corresponding video segment feature data mining result).

[0106] For example, in some implementations, the step of using a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video to output the target video recognition result corresponding to the stitched campus IoT surveillance video may further include the following implementable sub-steps:

[0107] The extracted example spliced ​​campus IoT surveillance video and the example video annotation results corresponding to the example spliced ​​campus IoT surveillance video are used to reflect the real campus security status corresponding to the example spliced ​​campus IoT surveillance video;

[0108] By utilizing the sub-network for video feature mining included in the video recognition neural network to be optimized, video feature mining is performed on the example spliced ​​campus IoT surveillance video, and the example video feature data mining results corresponding to the example spliced ​​campus IoT surveillance video are output (see above).

[0109] Using the sub-network for mining video object features, which is included in the video recognition neural network to be optimized, video object feature mining is performed on the video object information included in the example spliced ​​campus IoT surveillance video and the pre-configured video object information set, and the example video object feature data mining results corresponding to the example spliced ​​campus IoT surveillance video are output (refer to the above).

[0110] Using the video segment feature mining subnetwork included in the video recognition neural network to be optimized, video segment feature mining operation is performed on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured matching video segment set, and the example video segment feature data mining result corresponding to the example spliced ​​campus IoT surveillance video is output (refer to the above).

[0111] Based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, a network optimization operation is performed on the video recognition neural network to be optimized to form the video recognition neural network corresponding to the video recognition neural network to be optimized.

[0112] For example, in some implementations, the step of performing network optimization on the video recognition neural network to form the video recognition neural network corresponding to the video recognition neural network to be optimized, based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, may specifically include the following implementable sub-steps:

[0113] The example video feature data mining results, the example video object feature data mining results, and the example video segment feature data mining results are aggregated to form the example aggregated data mining results of the example spliced ​​campus IoT surveillance video. The example aggregated data mining results are then loaded into the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, so as to use the video recognition sub-network to be optimized to identify the example video recognition results corresponding to the example spliced ​​campus IoT surveillance video (see above).

[0114] Based on the differences between the example video recognition results and the example video annotation results, the network optimization cost corresponding to the neural network for the video recognition to be optimized is analyzed and output.

[0115] If the network optimization cost value corresponding to the video recognition neural network to be optimized is greater than or equal to the pre-configured network optimization cost reference value, the network optimization operation of the video recognition neural network to be optimized is performed according to the network optimization cost value; then, the video recognition neural network to be optimized after the network optimization operation can be marked as a candidate video recognition neural network, and the candidate video recognition neural network is then subjected to network optimization operation, and if the network optimization cost value corresponding to the candidate video recognition neural network after the network optimization operation is less than the network optimization cost reference value, the current candidate video recognition neural network is marked as a video recognition neural network.

[0116] Combination Figure 3 This invention also provides a data compatibility processing device based on a smart campus, which can be applied to the aforementioned IoT cloud platform. The data compatibility processing device based on a smart campus may include the following software functional modules:

[0117] The data compatibility conversion module (refer to the explanation of step S110 above) is used to perform data compatibility conversion on multiple campus IoT monitoring videos collected by multiple campus IoT terminal devices to form multiple target campus IoT monitoring videos, wherein the multiple target campus IoT monitoring videos have the same video format.

[0118] The video frame splicing processing module (refer to the explanation of step S120 above) is used to perform video frame splicing processing on the target campus IoT monitoring video frames included in the multiple target campus IoT monitoring videos based on the video frame timestamp information corresponding to each target campus IoT monitoring video frame included in each target campus IoT monitoring video to form a corresponding spliced ​​campus IoT monitoring video. The spliced ​​campus IoT monitoring video includes multiple spliced ​​campus IoT monitoring video frames, and each spliced ​​campus IoT monitoring video frame is formed by splicing multiple target campus IoT monitoring video frames with the same video frame timestamp information.

[0119] The video recognition processing module (refer to the explanation of step S130 above) is used to perform video recognition processing on the spliced ​​campus IoT monitoring video using a video recognition neural network formed by network optimization, so as to output the target video recognition result corresponding to the spliced ​​campus IoT monitoring video. The target video recognition result is used to reflect the campus security status of the smart campus area corresponding to the spliced ​​campus IoT monitoring video.

[0120] In summary, the present invention provides a data compatibility processing method and IoT cloud platform based on smart campuses. This method performs data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple target campus IoT surveillance videos. Based on the corresponding video frame timestamp information, it performs video frame stitching processing on the target campus IoT surveillance video frames included in the multiple target campus IoT surveillance videos to form a corresponding stitched campus IoT surveillance video. Finally, it utilizes a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, outputting the target video recognition result corresponding to the stitched campus IoT surveillance video. Based on the foregoing, by first stitching video frames and then performing recognition processing on the stitched video, compared to conventional technical solutions that perform recognition processing separately, it achieves higher processing efficiency and can, to a certain extent, improve the problem of low efficiency in data compatibility processing in existing technologies.

[0121] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A data compatibility processing method based on a smart campus, characterized in that, The data compatibility processing method based on smart campuses, applied to IoT cloud platforms, includes: Data compatibility conversion is performed on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple target campus IoT surveillance videos, which have the same video format. Based on the video frame timestamp information corresponding to each frame of the target campus IoT monitoring video included in each target campus IoT monitoring video, for each video frame timestamp information, each frame of the target campus IoT monitoring video with the video frame timestamp information is extracted from the plurality of target campus IoT monitoring videos to form a set of video frames to be spliced ​​corresponding to the video frame timestamp information. For each video frame timestamp, based on the positional relationship between the corresponding smart campus sub-regions, the target campus IoT monitoring video frames included in the set of video frames to be stitched according to the video frame timestamp information are subjected to video frame stitching processing to form a corresponding stitched campus IoT monitoring video frame. Then, based on the corresponding video frame timestamp information, the stitched campus IoT monitoring video frames are combined to form a corresponding stitched campus IoT monitoring video. The stitched campus IoT monitoring video includes multiple stitched campus IoT monitoring video frames, and each stitched campus IoT monitoring video frame is formed by stitching together multiple target campus IoT monitoring video frames with the same video frame timestamp information. A video recognition neural network, formed through network optimization, is used to perform video recognition processing on the spliced ​​campus IoT surveillance video to output the target video recognition result corresponding to the spliced ​​campus IoT surveillance video. The target video recognition result is used to reflect the campus security status of the smart campus area corresponding to the spliced ​​campus IoT surveillance video.

2. The data compatibility processing method based on smart campus as described in claim 1, characterized in that, The step of performing data compatibility conversion on multiple campus IoT surveillance videos collected by multiple campus IoT terminal devices to form multiple corresponding target campus IoT surveillance videos includes: The synchronous data acquisition instruction is sent to each of the multiple campus IoT terminal devices connected by communication, so that each campus IoT terminal device synchronously acquires data from the corresponding smart campus sub-area according to the synchronous data acquisition instruction, so as to form multiple synchronized campus IoT monitoring videos. Any two campus IoT monitoring video frames with the same video frame time sequence have the same video frame timestamp information. Data compatibility conversion is performed on the multiple campus IoT surveillance videos to form multiple corresponding target campus IoT surveillance videos.

3. The data compatibility processing method based on smart campus as described in claim 1 or 2, characterized in that, The step of using a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, and outputting the target video recognition result corresponding to the stitched campus IoT surveillance video, includes: The video feature mining subnetwork, which is part of the video recognition neural network formed through network optimization, is used to perform video feature mining on the spliced ​​campus IoT surveillance video, and outputs the video feature data mining results corresponding to the spliced ​​campus IoT surveillance video. The video object feature mining subnetwork included in the video recognition neural network is used to perform video object feature mining operations on the video object information included in the spliced ​​campus IoT monitoring video and the pre-configured video object information set, and output the video object feature data mining results corresponding to the spliced ​​campus IoT monitoring video; The video segment feature mining subnetwork included in the video recognition neural network is used to perform video segment feature mining operations on the spliced ​​campus IoT monitoring video and the matching video segments included in the pre-configured set of matching video segments, and output the video segment feature data mining results corresponding to the spliced ​​campus IoT monitoring video. The video feature data mining results, the video object feature data mining results, and the video segment feature data mining results are aggregated to form the aggregated data mining results corresponding to the spliced ​​campus IoT surveillance video; The aggregated data mining results are loaded into the video recognition subnetwork included in the video recognition neural network, so as to use the video recognition subnetwork to identify the target video recognition result corresponding to the spliced ​​campus IoT surveillance video.

4. The data compatibility processing method based on smart campus as described in claim 3, characterized in that, The step of using a video feature mining subnetwork included in a video recognition neural network formed through network optimization to perform video feature mining operations on the stitched campus IoT surveillance video and outputting the video feature data mining results corresponding to the stitched campus IoT surveillance video includes: The spliced ​​campus IoT surveillance video is segmented to form spliced ​​campus IoT surveillance video segments. Then, the spliced ​​campus IoT surveillance video segments are mapped to output the video segment mapping results corresponding to the spliced ​​campus IoT surveillance video segments. The video segment distribution information of the spliced ​​campus IoT surveillance video segments is analyzed, and then the video segment distribution information is mapped to output the video segment distribution information mapping result corresponding to the video segment distribution information. The analysis outputs the video segment segment identification information mapping result corresponding to the spliced ​​campus IoT monitoring video segment. Then, the video segment mapping result, the video segment distribution information mapping result, and the video segment segment identification information mapping result are aggregated to form the aggregated information mapping result corresponding to the spliced ​​campus IoT monitoring video segment. The aggregated information mapping result is loaded into the video feature mining subnetwork of the video recognition neural network formed by network optimization, so as to use the video feature mining subnetwork to perform video feature mining operation on the aggregated information mapping result, output the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video segment, and then analyze and output the video feature data mining result corresponding to the spliced ​​campus IoT monitoring video based on the initial video feature data mining result corresponding to the spliced ​​campus IoT monitoring video.

5. The data compatibility processing method based on smart campus as described in claim 3, characterized in that, The step of utilizing the video object feature mining subnetwork included in the video recognition neural network to perform video object feature mining operations on the video object information included in the spliced ​​campus IoT surveillance video and the pre-configured video object information set, and outputting the video object feature data mining results corresponding to the spliced ​​campus IoT surveillance video, includes: An object comparison operation is performed on the spliced ​​campus IoT surveillance video and the video object information included in the pre-configured video object information set to output the object comparison result corresponding to the spliced ​​campus IoT surveillance video; If the object comparison result reflects that the spliced ​​campus IoT surveillance video contains video object information included in the video object information set, the video object information contained in the spliced ​​campus IoT surveillance video is marked to form corresponding video object information to be processed. The video object information to be processed is loaded into the video object feature mining sub-network included in the video recognition neural network, so as to use the video object feature mining sub-network to perform video object feature mining operation on the video object information to be processed, and output the video object information feature distribution corresponding to the video object information to be processed. Based on the distribution of video object information features, the video object feature data mining results corresponding to the spliced ​​campus IoT surveillance video are analyzed and output.

6. The data compatibility processing method based on smart campus as described in claim 3, characterized in that, The step of utilizing the video segment feature mining subnetwork included in the video recognition neural network to perform video segment feature mining operations on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured matching video segment set, and outputting the video segment feature data mining results corresponding to the spliced ​​campus IoT surveillance video, includes: A video segment comparison operation is performed on the spliced ​​campus IoT surveillance video and the pre-configured set of matching video segments to output the video segment comparison result corresponding to the spliced ​​campus IoT surveillance video; If the video segment comparison results show that there is a matching video segment in the matching video segment set that matches the spliced ​​campus IoT monitoring video, then the matching video segment in the matching video segment set that matches the spliced ​​campus IoT monitoring video is marked to form a corresponding matching video segment to be processed. The video segment to be matched is loaded into the video segment feature mining sub-network included in the video recognition neural network, so as to use the video segment feature mining sub-network to perform video segment feature mining operation on the video segment to be matched and output the matching video segment feature distribution corresponding to the video segment to be matched. Based on the feature distribution of the matched video segments, the feature data mining results of the video segments corresponding to the spliced ​​campus IoT surveillance video are analyzed and output.

7. The data compatibility processing method based on smart campus as described in claim 3, characterized in that, The step of using a video recognition neural network formed through network optimization to perform video recognition processing on the stitched campus IoT surveillance video, and outputting the target video recognition result corresponding to the stitched campus IoT surveillance video, further includes: The extracted example spliced ​​campus IoT surveillance video and the example video annotation results corresponding to the example spliced ​​campus IoT surveillance video are used to reflect the real campus security status corresponding to the example spliced ​​campus IoT surveillance video; The video feature mining subnetwork, which is included in the video recognition neural network to be optimized, is used to perform video feature mining on the example spliced ​​campus IoT surveillance video, and outputs the example video feature data mining results corresponding to the example spliced ​​campus IoT surveillance video. Using the video object feature mining subnetwork included in the video recognition neural network to be optimized, video object feature mining is performed on the video object information included in the example spliced ​​campus IoT surveillance video and the pre-configured video object information set, and the example video object feature data mining results corresponding to the example spliced ​​campus IoT surveillance video are output. The video segment feature mining subnetwork, which is included in the video recognition neural network to be optimized, is used to perform video segment feature mining operation on the spliced ​​campus IoT surveillance video and the matching video segments included in the pre-configured set of matching video segments, and outputs the example video segment feature data mining result corresponding to the example spliced ​​campus IoT surveillance video. Based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, a network optimization operation is performed on the video recognition neural network to be optimized to form the video recognition neural network corresponding to the video recognition neural network to be optimized.

8. The data compatibility processing method based on smart campus as described in claim 7, characterized in that, The step of performing network optimization on the video recognition neural network to be optimized based on the example video feature data mining results, the example video object feature data mining results, the example video segment feature data mining results, the example video annotation results, and the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, to form the video recognition neural network corresponding to the video recognition neural network to be optimized, includes: The example video feature data mining results, the example video object feature data mining results, and the example video segment feature data mining results are aggregated to form the example aggregated data mining results of the example spliced ​​campus IoT surveillance video. The example aggregated data mining results are then loaded into the video recognition sub-network to be optimized included in the video recognition neural network to be optimized, so as to use the video recognition sub-network to be optimized to identify the example video recognition results corresponding to the example spliced ​​campus IoT surveillance video. Based on the example video recognition results and the example video annotation results, the network optimization cost corresponding to the neural network for the video recognition to be optimized is analyzed and output; If the network optimization cost value corresponding to the video recognition neural network to be optimized is greater than or equal to the pre-configured network optimization cost reference value, the network optimization operation of the video recognition neural network to be optimized is performed according to the network optimization cost value. The video recognition neural network to be optimized after network optimization is marked as a candidate video recognition neural network. Then, network optimization is performed on the candidate video recognition neural network. If the network optimization cost corresponding to the candidate video recognition neural network after network optimization is less than the network optimization cost reference value, the current candidate video recognition neural network is marked as a video recognition neural network.

9. An Internet of Things (IoT) cloud platform, characterized in that, It includes a processor and a memory, the memory being used to store a computer program, and the processor being used to execute the computer program to implement the data compatibility processing method based on a smart campus as described in any one of claims 1-8.

Images