CN-121982633-A - Intelligent video system application method and system suitable for smart city

CN121982633ACN 121982633 ACN121982633 ACN 121982633ACN-121982633-A

Abstract

The invention discloses an intelligent video system application method and system applicable to smart cities, and relates to the technical field of visual detection. The intelligent video system application method and system suitable for the smart city comprise the following steps of S1, collecting multi-source video frame data to construct a city video collecting data frame, S2, carrying out edge side preprocessing and picture quality measurement, outputting a video detection input set, S3, constructing an algorithm running data frame and a traceable evaluation index set, and S4, constructing a unified detection result service view and original scene reconstruction interface. The method effectively improves the accuracy of target detection and the adaptability to low-quality and degraded channel pictures in the multi-source urban video scene, and solves the problems that the algorithm result lacks model version and running environment marks and is difficult to realize effect traceability and contrast evaluation.

Inventors

Mai Haozhi
Tang Xiangfei
LIN YUEPENG
ZHENG RUISHENG
DAI YONGHONG
QIU SHAOMING
SONG YONGSHENG

Assignees

中通服中睿科技有限公司

Dates

Publication Date: 20260505
Application Date: 20251231

Claims (10)

1. An intelligent video system application method applicable to a smart city is characterized by comprising the following steps: S1, constructing an urban video acquisition data frame by acquired multi-source video frame data, and constructing an urban video acquisition space-time index view by geometric modeling of a view field and region coding; s2, edge side preprocessing and picture quality measurement are carried out on the urban video acquisition data frame sequence, and channel quality time sequence evaluation and quality driving frame scheduling output video detection input set are carried out; s3, running a data frame set based on a video detection input set construction algorithm, and constructing a model version view and a traceable evaluation index set; And S4, constructing a unified detection result service view based on the algorithm operation data frame set, and constructing a traceable backtracking and original scene reconstruction interface.
2. The method for intelligent video system application of intelligent city according to claim 1, wherein the specific process of constructing city video acquisition data frame from the acquired multi-source video frame data is: the method comprises the steps of inputting multi-source video frame data collected by a fixed camera, a dome camera, a cradle head camera, a vehicle-mounted terminal, a ship-mounted terminal and an unmanned aerial vehicle-mounted camera, carrying out unified time reference construction and channel sampling scheduling on the multi-source video frame data to form a city video collection data frame sequence, distributing unified sampling time stamps for each frame of video sample by taking unified time sources calibrated by unified time service as references in the collection process, sorting sampling records from different video terminals according to the sampling time stamps and classifying the sampling records into corresponding sampling periods, normalizing the sampling time stamps by adopting unified time units and initial reference points, mapping local time of different devices into relative time indexes under the unified time references, carrying out time alignment on each channel video sampling record on the same time axis by taking the sampling time stamps endowed by the unified time sources as a main time axis when the unified collection frame sequence is generated, and outputting the city video collection data frame sequence arranged according to the unified sampling time sequence.
3. The method for intelligent video system application applicable to intelligent cities according to claim 1, wherein the specific process of constructing the urban video acquisition space-time index view by performing field geometric modeling and region coding is as follows: inputting a city basic geographic information data set and a city video acquisition data frame sequence, wherein the city basic geographic information data set is composed of basic topography base maps, road and building vector data, administrative division and functional division data, carrying out geometric modeling and region coding on the installation position and the view field range of each path of video terminal, mapping the city video acquisition data frame sequence to a unified space-time coordinate system, and constructing a city video acquisition space-time index view, namely mapping the installation point of each path of fixed camera to the city unified coordinate system through coordinate conversion, inquiring administrative division boundaries, road center lines and park boundaries, which are intersected with the installation point and peripheral regions in the city basic geographic information data set, taking the administrative division codes, the road or park marks, which are belonged to the installation point, as region codes, and establishing a one-to-one or one-to-many mapping relation to form a time-varying equipment track point set; Binding the region code, the time sequence and the frame-level global unique identifier, the equipment unique identifier, the channel identifier and the unified analysis time in the city video acquisition data frame sequence, generating frame-level time-space association records for each frame of city video acquisition data frame sequence, writing the frame-level time-space association records into a video time-space index table, and outputting a city video acquisition time-space index view.
4. The method for intelligent video system application for intelligent city according to claim 1, wherein the specific process of edge side preprocessing and picture quality measurement for city video acquisition data frame sequence is as follows: The method comprises the steps of inputting a city video acquisition data frame sequence and a city video acquisition space-time index view, carrying out edge side preprocessing and picture quality measurement on the city video acquisition data frame sequence to form a frame-level quality feature vector sequence, configuring an annular frame buffer queue for each path in an edge computing node, writing the city video acquisition data frame sequence into a buffer according to a unified analysis time sequence to enable the buffer to always keep a continuous frame sequence in the last period of time; The method comprises the steps of obtaining quality score and quality grade label of each frame based on frame grade quality feature vector, uniformly writing the frame grade quality feature vector, the quality score and the quality grade label into a frame grade quality feature table to form city video quality feature vector sequence arranged in time sequence, and outputting the frame grade quality feature vector sequence and the quality grade label set.
5. The method for intelligent video system application in intelligent city according to claim 1, wherein the specific process of channel quality time sequence evaluation and quality driving frame scheduling output video detection input set is: The method comprises the steps of inputting a frame-level quality feature vector sequence, a city video acquisition data frame sequence and a city video acquisition space-time index view, carrying out channel quality time sequence assessment on the picture quality change of a multi-source video channel in a time dimension, carrying out quality-driven frame filtering and scheduling according to the channel quality time sequence assessment, and reconstructing the multi-source video channel into a quality-perceived video detection input set: The channel quality health index is built, the channel quality state and the change process are written into a channel quality state table and a quality exception log table while the channel quality health index is generated, and a quality-driven video detection frame scheduling strategy is built based on the channel quality state and the frame level quality grade, namely a quality-aware video detection input set is output.
6. The method for intelligent video system application applicable to intelligent cities according to claim 1, wherein the specific process of running the data frame set based on the video detection input set construction algorithm is as follows: The method comprises the steps of inputting a video detection input set, uniformly collecting and structuring and packaging detection results, model configuration and operation environment characteristics generated in the execution process of each detection task, constructing an algorithm operation data frame set, selecting model versions and operation configuration matched with task types, scene types and quality conditions according to task items in the video detection input set, positioning and loading corresponding preprocessed image frames in a city video collection data frame table according to a frame-level global unique identifier on an inference node side, inputting images into a designated model version, structuring and packaging detection results output by the models, obtaining detection result basic records, collecting operation characteristic information called by current reasoning in real time in the reasoning execution process, quantifying the reasoning environment and resource consumption into operation characteristic vectors, constructing algorithm operation data frames, writing the algorithm operation data frame records into the algorithm operation data frame table, and outputting the algorithm operation data frame set.
7. The method for intelligent video system application of intelligent city according to claim 1, wherein the specific process of constructing model version view and traceable evaluation index set is: Inputting an algorithm operation data frame set, a city video acquisition data frame sequence, a city video acquisition space-time index view, a frame-level quality feature vector sequence and a channel quality state record, aggregating and comparing detection results and operation characteristics of different model versions on the basis of unified data, and constructing a model version view and a traceable evaluation index set which are oriented to version management, wherein the algorithm operation data frame set is grouped and aggregated by taking a model identifier and a model version number as main dimensions, so as to construct the model version view; And respectively counting samples in different quality grades and different channel quality states, so that the performances of the same model version under the conditions of ideal image quality, general image quality and degraded image quality can be distinguished, and for a scene configured with multi-model version parallel detection, carrying out one-to-one comparison on detection results given by a plurality of model versions on the same frame and the same target area by using the same frame-level global unique identifier as an associated key to construct a cross-version comparison sub-view, and outputting the model version view and an evaluation index set.
8. The method for intelligent video system application applicable to intelligent cities according to claim 1, wherein the specific process of constructing a unified detection result service view based on the algorithm running data frame set is as follows: The method comprises the steps of inputting an algorithm operation data frame set, a model version view and an evaluation index set, a city video acquisition data frame sequence, a city video acquisition space-time index view, a frame-level quality feature vector sequence and a channel quality state record, carrying out field cutting, semantic merging and service encapsulation on a multi-source algorithm operation data frame to construct a unified detection result service view, carrying out layered extraction and semantic merging on fields based on the algorithm operation data frame set, carrying out structural decoupling on an operation process field used internally and a detection result field of external service, only preserving technical fields directly related to an upper layer system in an outer layer service view, and injecting version health degree information and recommended grade in the model version view and the evaluation index set into each detection result record to output the unified detection result service view.
9. The method for intelligent video system application applicable to intelligent cities according to claim 1, wherein the specific process for constructing the traceable backtracking and original scene reconstruction interface is as follows: A unified detection result service view and an algorithm operation data frame set are input, association relations among detection results, model versions, picture quality, operation environments and an original video acquisition process are organized and indexed, a traceable retrospective interface supporting combined query and original scene reconstruction is constructed, wherein a combined condition query mechanism is designed, a user can specify model versions, time intervals, regional ranges, picture quality conditions, channel quality state characteristics and operation anomaly marking conditions, and matched detection result records are rapidly screened under the support of multi-level indexes of the service view to obtain candidate result sets; After candidate detection result records and corresponding operation data frame identifiers are obtained, the original technical context is reconstructed layer by layer through a tracing index, namely the operation data frame identifiers are used as keys, model identifiers and model version numbers used in the detection are recovered, a frame identifier frame-level global unique identifier is used as a key, unified analysis time, space coordinate positions and view field coverage areas of frames are obtained, picture quality changes and channel health index time curves in time windows before and after the frames are recovered, and a unified traceable tracing interface is provided for an upper layer system.
10. An intelligent video system application system for a smart city, applying a method for applying an intelligent video system for a smart city according to any one of claims 1 to 9, comprising: The video acquisition module is used for constructing urban video acquisition data frames from acquired multi-source video frame data, and constructing urban video acquisition space-time index views by performing field geometric modeling and region coding; the quality perception module is used for carrying out edge side preprocessing and picture quality measurement on the urban video acquisition data frame sequence, and carrying out channel quality time sequence evaluation and quality driving frame scheduling output video detection input set; The visual detection module is used for operating the data frame set based on a video detection input set construction algorithm and constructing a model version view and a traceable evaluation index set; And the view and trace back evaluation module is used for constructing a unified detection result service view based on the algorithm operation data frame set and constructing a traceable trace back and original scene reconstruction interface.

Description

Intelligent video system application method and system suitable for smart city Technical Field The invention relates to the technical field of visual detection, in particular to an intelligent video system application method and system applicable to intelligent cities. Background With the continuous expansion of the video monitoring scale of the smart city and the continuous improvement of the iteration frequency of the visual algorithm, the prior art exposes the problems of non-uniform time base, missing space-time index, non-traceable algorithm version and running environment, difficulty in quantifying the influence of channel image quality degradation on a detection result and the like in the aspects of multi-source video acquisition, algorithm deployment and effect evaluation. The front-end cameras and the edge gateway are provided by different manufacturers, the local clocks of the front-end cameras and the edge gateway are independently maintained, the uploaded data only carries a single equipment time field, and the cloud side also uses the database warehouse-in time as a sequencing basis in a mixed mode, so that the time sequence of the same event between multiple tables and multiple channels is contradicted. For example, the invention patent with bulletin number CN109271554B discloses an intelligent video recognition system and application thereof, wherein the intelligent video recognition system comprises front-end access equipment, a video image intelligent analysis system, a video big data analysis system, a video cloud platform and a comprehensive application system. The invention realizes networking of established cameras, unified application of resource integration, wide coverage, rich functions, multiple characteristic quantity capable of being monitored and comprehensive monitoring elements, simultaneously builds a video monitoring front end based on the Internet, has high speed and low cost, selects high-value point positions, deploys an intelligent front end and realizes people and vehicle snapshot and control alarm. For example, the invention patent with the publication number of CN103475870B discloses a distributed extensible intelligent video monitoring system which comprises a decoding management module, a polling scheduling module, a combination analysis module, a resource management module, a picture display module, an alarm pushing module and an external communication module. Aiming at the new characteristics of digitization, intellectualization and distribution of the modern video monitoring system, the invention can simultaneously meet the application requirements of a small video monitoring system and a large intelligent video monitoring system through flexible configuration, and has the capacity of large-scale smooth expansion. In the prior art, the existing system usually records the detection result only by a single timestamp and a device number, lacks a city video acquisition space-time index view based on a unified time base, lacks joint modeling of channel quality health indexes and detection processes, and lacks a model version view and operation environment view linkage mechanism taking an algorithm operation data frame as a core, so that cross-camera event reconstruction is difficult, the model upgrading effect is difficult to objectively compare, and false alarm missing reasons are difficult to quickly locate. Therefore, in view of the above problems, there is a need for an intelligent video system application method and system suitable for smart cities. Disclosure of Invention Technical problem to be solved Aiming at the defects of the prior art, the invention provides an intelligent video system application method and system applicable to smart cities, and solves the problems that an algorithm result lacks a model version and an operation environment mark, and traceability and comparison evaluation of effects are difficult to realize. Technical proposal The intelligent video system application method and system suitable for the smart city comprise the following steps of S1, constructing a city video acquisition data frame by using acquired multi-source video frame data, constructing a city video acquisition space-time index view by visual field geometric modeling and region coding, S2, carrying out edge side preprocessing and picture quality measurement on a city video acquisition data frame sequence, carrying out channel quality time sequence assessment and quality driving frame dispatching and outputting a video detection input set, S3, constructing an algorithm operation data frame set based on the video detection input set, constructing a model version view and a traceable assessment index set, S4, constructing a unified detection result service view based on the algorithm operation data frame set, and constructing a traceable retrospective and original scene reconstruction interface. The method comprises the specific processes of constructing an urban video acqu