CN-122024016-A - Method, apparatus, program product and system for image enhancement for intelligent analysis tasks

CN122024016ACN 122024016 ACN122024016 ACN 122024016ACN-122024016-A

Abstract

The present disclosure relates to a method, apparatus, computer program product and security monitoring system for image enhancement for intelligent analysis tasks. The method includes, in response to receiving an image from a unified service interface and an indication of a type of intelligent analysis task to be performed based on the image, determining whether to perform image enhancement on the image, in response to determining to perform image enhancement, determining a set of image enhancement algorithm models to perform, in response to determining that a first subset of the image enhancement algorithm models in the set are disposed on a cloud platform and a second subset of the image enhancement algorithm models in the set are disposed in a security monitoring system, invoking one of the first subset or the second subset to process the image to obtain a first processed image, invoking the other of the first subset or the second subset to process the first processed image to obtain a second processed image, and transmitting the second processed image via the unified service interface for the intelligent analysis task.

Inventors

WANG YONGJIAN

Assignees

深圳市联洲国际技术有限公司

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (12)

1. A method for enhancing images aiming at intelligent analysis tasks of a security monitoring system comprises the following steps: responsive to receiving an image acquired by the security monitoring system from a unified service interface and an indication of a type of intelligent analysis task to be performed based on the image, determining whether to perform image enhancement on the image; in response to determining that image enhancement is to be performed, determining a set of image enhancement algorithm models to be performed; In response to determining that a first subset of the image enhancement algorithm models in the set are deployed on a cloud platform and a second subset of the image enhancement algorithm models in the set are deployed in the security monitoring system, invoking one of the first subset or the second subset to process the image to obtain a first processed image; invoking the other of the first subset or the second subset to process the first processed image to obtain a second processed image, and The second processed image is sent via the unified service interface for the intelligent analysis task.
2. The method of claim 1, wherein the image enhancement algorithm models in the second subset are stored in a local algorithm repository of the security monitoring system and model instantiation is implemented by a warm-boot approach or a cold-boot approach, wherein the warm-boot approach includes an approach in which the image enhancement algorithm models have been partially or fully loaded into system memory before being invoked, and the cold-boot approach includes an approach in which the image enhancement algorithm models have not been loaded into system memory before being invoked.
3. The method of claim 2, wherein the image enhancement algorithm models in the second subset implement model instantiation in either a hot-start mode or a cold-start mode, comprising the image enhancement algorithm models in the second subset determining to implement model instantiation in either a hot-start mode or a cold-start mode based on at least one of a model call frequency and a system resource occupancy.
4. The method of claim 1, wherein a cloud algorithm repository of the cloud platform stores a plurality of historical versions of each image enhancement algorithm model and performance metrics corresponding to each historical version in the first subset, and a local algorithm repository of the security monitoring system stores a plurality of historical versions of each image enhancement algorithm model and performance metrics corresponding to each historical version in the second subset, the method further comprising: responsive to determining that a performance index of a current version of a first image enhancement algorithm model in the first subset does not meet a performance benchmark for the first image enhancement algorithm model, rollback the first image enhancement algorithm model to a version with an optimal performance index among a plurality of historical versions of the first image enhancement algorithm model, and In response to determining that the performance index of the current version of the second image enhancement algorithm model in the second subset does not meet the performance benchmark for the second image enhancement algorithm model, rollback the second image enhancement algorithm model to a version of the plurality of historical versions of the second image enhancement algorithm model that has the optimal performance index, The performance indexes of the image enhancement algorithm model comprise the running speed, the resource occupancy rate and the firmware compatibility of the image enhancement algorithm model.
5. The method of claim 1, wherein determining whether to perform image enhancement on the image is based on at least one of: comparing an image quality of the image to a predefined quality threshold; A predefined profile or a user profile indicating whether the image enhancement is enabled based on a type of the intelligent analysis task, and And comparing the local hardware resource use parameter of the security monitoring system with a predefined resource use threshold.
6. The method of claim 1, wherein responsive to determining that image enhancement is to be performed, determining a set of image enhancement algorithm models to be performed comprises: Determining a set of image enhancement algorithm models to be executed based on a predefined mapping relationship of the type of the intelligent analysis task and one or more image enhancement algorithm models, or And determining a set of image enhancement algorithm models to be executed by using a reinforcement learning model based on the type of the intelligent analysis task, wherein the reinforcement learning model updates the mapping relation between the type of the intelligent analysis task and one or more image enhancement algorithm models based on the feedback result of the historical intelligent analysis task.
7. The method of claim 1, wherein the received image is stored in a message queue based on a type of the intelligent analysis task.
8. The method of claim 1, wherein the image enhancement algorithm models in the first subset are stored in a cloud algorithm repository of the cloud platform and run in a server-less architecture environment of the cloud platform.
9. The method of claim 1, wherein the invocation of the second subset is implemented based on gRPC protocol or WebSocket protocol.
10. An apparatus for image enhancement for intelligent analysis tasks of security monitoring systems, comprising: one or more processors; A memory coupled to a processor of the processors, and A set of computer program instructions stored in the memory, which when executed by a processor of the processor performs the method according to any one of claims 1-9.
11. A computer program product comprising a computer program which, when executed by a processor, implements the method according to any of claims 1-9.
12. A security monitoring system comprising an acquisition unit, the device of claim 10 and an analysis unit, The acquisition unit is configured to acquire an image; The apparatus is configured to: Receiving the image from the acquisition unit and receiving an indication of the type of intelligent analysis task from the analysis unit, and Sending the second processed image to the analysis unit, The analysis unit is configured to perform the intelligent analysis task for the second processed image.

Description

Method, apparatus, program product and system for image enhancement for intelligent analysis tasks Technical Field The disclosure relates to the field of intelligent security monitoring, in particular to a method, a device, a computer program product and a security monitoring system for image enhancement aiming at an intelligent analysis task of the security monitoring system. Background With the rapid development of artificial intelligence and computer vision technologies, security monitoring systems have evolved gradually from traditional passive video modes to intelligent and automated active analysis modes. The security monitoring system not only can collect video streams or images, but also can execute intelligent analysis tasks such as face recognition, target detection, behavior recognition and the like based on the collected images, thereby playing an important role in various fields such as public security, intelligent buildings, traffic control and the like. The monitoring camera is used as a front end core acquisition component of the security monitoring system, the quality of an output image of the monitoring camera directly influences the performance of a follow-up intelligent analysis task, and the accuracy and reliability of key tasks such as face recognition, target detection and behavior recognition are basic preconditions for guaranteeing the stable operation of the whole security monitoring system. In the actual application scene of security monitoring, the monitoring camera often faces complex and changeable natural environment and working condition challenges, and these adverse factors become bottlenecks for restricting the image acquisition quality. For example, low illumination conditions (such as scenes of night, cloudy days, tunnels and the like) can cause insufficient image brightness and enhanced noise signals, haze and dust weather can cause image contrast reduction and detail shielding, rain and snow weather can introduce random noise and cause image blurring distortion due to refraction and reflection of light, strong light and backlight scenes can easily cause partial overexposure or underexposure of images to lose key area information, and camera shake, rapid movement of targets and the like can generate dynamic blurring to further damage the detail integrity of the images. The quality of the acquired image is obviously reduced, so that effective features are difficult to extract by a follow-up intelligent analysis algorithm, and further the problems of reduced recognition accuracy, increased misjudgment rate and missed detection rate and the like are caused, and the efficiency of the security monitoring system is seriously influenced. In order to improve image acquisition quality in complex environments, various image enhancement algorithms have been proposed in recent years, aiming at improving visual quality and information readability of images, including, for example, contrast enhancement algorithms, noise reduction algorithms, defogging algorithms, color enhancement algorithms, low light enhancement algorithms, deblurring algorithms, and the like. Disclosure of Invention The present disclosure provides a method, an apparatus, a computer program product, and a security monitoring system for image enhancement for an intelligent analysis task of the security monitoring system, which can support a dual-mode deployment mechanism of multiple image enhancement algorithm models, that is, a part of the multiple image enhancement algorithm models can be deployed on a cloud platform, another part of the multiple image enhancement algorithm models can be deployed in the security monitoring system, and call each image enhancement algorithm model according to actual requirements, so as to enhance images for various subsequent intelligent analysis tasks. Therefore, the image enhancement effect can be achieved through the image enhancement algorithm model, the image enhancement algorithm model can be adapted to hardware conditions of the security monitoring system through the dual-mode deployment mechanism, when hardware resources of the security monitoring system are insufficient to deploy all or part of the image enhancement algorithm model, image enhancement processing is completed through elastic computing force of the cloud platform, local hardware resources of the security monitoring system are not required to be occupied, the problems of processing delay, task blocking, even system breakdown and the like caused by insufficient local computing force of the security monitoring system are effectively avoided, and therefore stable, reliable and efficient operation of the image enhancement algorithm under diversified hardware deployment scenes is achieved. In addition, the method, the device, the computer program product and the security monitoring system can realize standardized interaction through the unified service interface, and the difference between the local and cloud impleme