TW201640419A - Image recognition and monitoring system and its implementing method - Google Patents

Abstract

The present invention discloses image recognition and monitoring system and its implementing method and the system comprises a camera device, a cloud server, and a local server. The method is that the video and image information collected by camera device will be analyzed and selected through cloud server. Moreover, the cloud server gives a mark for each subject in the selected video and image information and the local server conducts a construction for an identification model and generates an identification mechanism. This mechanism will be deployed to the camera devices. Therefore, the camera device can recognize video and image information according to identification mechanism in real time and the conventional device which has recognition problem will be solved.

Description

Image recognition and monitoring system and implementation method thereof

An image recognition and monitoring system and an implementation method thereof, which are matched with a local server by a cloud and through deep learning technology, so that the camera device can accurately perform image recognition.

Through the use of surveillance cameras, the security control of specific areas has been seen daily. People rely on surveillance cameras to achieve many monitoring purposes. They can be seen at all intersections, lanes, downstairs, buildings, etc. Monitor the traces of the camera. With the evolution of many technologies (such as microprocessors, image recognition and analysis, and Internet technologies), smart monitoring has become one of the future trends, and smart monitoring makes surveillance cameras no longer only capable of image information. Taking and storing, it is better to identify the objects and features in the image information. For example, license plate recognition is a common implementation case. However, since the mechanism for identifying images stored in the surveillance camera has been fixed, when it appears If the identification mechanism is unrecognizable, the surveillance camera will lose its ability to recognize, resulting in insufficient effective recognition. The identification mechanism of the surveillance camera can be updated by human operation. However, when the number of surveillance cameras reaches a certain number (for example, the entire factory site may have hundreds or even thousands of surveillance cameras), updating one by one is a rather time consuming procedure. In the case of the "Identification Module" of the Republic of China New Patent Bulletin No. M491214, the inventor disclosed an identification module comprising a server host and a local host, as a local host of the local host. After the device captures an image, a local identification device of the local host extracts features from the image, and a local data device classifies and distinguishes according to the extracted features to identify which object the image is, and the local data device cannot recognize the image. Then, the image is converted into image information and classified and identified by a server model device of the server host, and the result of the identification is transmitted back to the local host by the server host, so that the local data device can add the model information. It will not be necessary to intervene in a large number of artificial processes, that is, by the effect of Machine Learning. Increase local capacity to capture image recognition device of the local host.

However, it can be seen from the above-mentioned prior art that the identification module of the prior art performs the feature recognition of the image information by the servo identification device, and performs two actions of classifying and distinguishing the image information from the servo model device, and simultaneously passes through a single server. The host executes, that is, when the number of monitoring cameras (that is, the local host's local capturing device) reaches a certain number, and most of the monitoring cameras simultaneously fail to recognize the image, the server host will be burdened, thereby reducing Operating efficiency, and due to current network speed limitations, there is a delay in the transmission of data between the local camera and the remote server. The method does not enable real-time identification of local cameras.

In view of the above problems, the present inventors have conducted research and analysis on existing system architectures based on years of experience in related industries and product design, and are capable of designing a better image recognition and monitoring system; thus, the present invention The main purpose of the invention is to provide an image recognition and monitoring system and an implementation system for effectively performing image recognition by enabling the image recognition device to accurately perform identification model construction and category identification mechanism by using the cloud and the local server. method.

In order to achieve the above object, the image recognition and monitoring system of the present invention includes at least one camera device, a cloud server and a local server, and the image recognition and monitoring system is implemented. After the image information, it will be identified according to a category identification mechanism. In addition, the image information will be transmitted to the cloud server, and the image information will be parsed to obtain a plurality of image information, and multiple image information will be selected. And, at least one object in the selected image information is marked, and then, the local server performs an identification model construction according to the marked image information, and generates an update category identification mechanism. The update category identification mechanism is transmitted to the cloud server and deployed to the camera device for the camera device to instantly recognize the image information according to the update category identification mechanism.

In order for your review board to have a clear understanding of the purpose, technical features and effects of the present invention, the following description will be used in conjunction with the illustrations, please refer to it.

10‧‧‧Image Identification and Monitoring System

101‧‧‧ camera

102‧‧‧Cloud Server

1011‧‧‧Micro Processing Module

1021‧‧‧Image Analysis and Selection Module

1012‧‧‧ Storage Module

1022‧‧‧ Object Marking Module

1013‧‧‧Image capture module

1023‧‧‧ Identification Mechanism Deployment Modeling Group

1014‧‧‧Image recognition module

1024‧‧‧Database

1015‧‧‧ Camera transmission module

103‧‧‧Local server

1031‧‧‧Local transmission module

1032‧‧‧ Identification Model Construction Module

A1‧‧‧Image Information

A2‧‧‧ Image Information

A3‧‧‧Image Information

A4‧‧‧Image Information

B1‧‧‧ Image Information

C1‧‧‧ Image Information

D1‧‧‧ Image Information

S10‧‧‧ Implementation method of image recognition and monitoring system

S101‧‧‧Image capture steps

S102‧‧‧Image analysis and selection steps

S103‧‧‧ Object marking steps

S104‧‧‧ Identification model construction steps

S105‧‧‧ Identification mechanism deployment steps

Figure 1 is a schematic diagram of the system architecture of the present invention.

Fig. 2 is a schematic view showing the structure of an image pickup apparatus of the present invention.

Figure 3A is a schematic diagram (I) of the implementation of image analysis and selection of the present invention.

Figure 3B is a schematic diagram (2) of the implementation of image analysis and selection of the present invention.

Fig. 4 is a schematic view showing the implementation of the object mark of the present invention.

Figure 5 is a diagram showing the steps of the implementation of the present invention.

Please refer to FIG. 1 , which is a schematic diagram of the system architecture of the present invention. As shown in the figure, the image recognition and monitoring system 10 of the present invention includes at least one camera device 101 , a cloud server 102 and a local device. The end server 103 and the cloud server 102 respectively form an information connection with the camera 101 and the local server 103, and then refer to "FIG. 2", which is a schematic diagram of the architecture of the camera device of the present invention. The camera device 101 has a microprocessor module 1011, a storage module 1012, an image capturing module 1013, an image recognition module 1014, and a camera transmission module 1015 and a microprocessor module 1011. In the information link, the storage module 1012 is pre-stored with a classifier (not shown) when the image capture mode is used. After the image information (not shown) is captured by the group 1013, the image recognition module 1014 can identify the image information according to the category identification mechanism in the storage module 1012. In the preferred embodiment, the camera 101 is captured. The image transmission module 1015 transmits the image information captured by the image capturing module 1013 to the cloud server 102. Specifically, the image recognition module 1014 continuously identifies the image information according to the category identification mechanism. At the same time, the camera transmission module 1015 transmits the image information captured by the image capturing module 1013 to the cloud server 102, and the cloud server 102 is shown in FIG. 1 . The invention comprises an image analysis and selection module 1021, an object marking module 1022, an identification mechanism cloth modeling group 1012 and a database 1024, and an image analysis and selection module 1021, an object marking module 1022 and identification. The mechanism cloth modeling group 1012 respectively forms an information link with the data library 1024, and the image analysis and selection module 1021 can be used for analyzing and selecting a plurality of image information for the image information, and the object marking module 1022 is available. At least one object in the image information selected by the image analysis and selection module 1021 is labeled, and the marked image information (ie, image + mark) is stored in the database 1024, and then The local end server 103 has a local end transmission module 1031 and an identification model construction module 1032 connected to the local end transmission module 1031. The local end transmission module 1031 can be used to mark a plurality of tags. The image information is transmitted to the local server 103 for the identification model construction module 1032 to construct an identification model, wherein the identification model is constructed, and the preferred embodiment may be a deep convolutional neural network. (Deep convolutional neural network), of course, other models available for deep learning, such as deep beliefs The deep belief networks or the recurrent neural network may also be the embodiment of the present invention, and are not limited thereto, and the identification model of the deep convolutional neural network is constructed. A conventional technique will not be described in this case. In addition, after the identification model construction module 1032 completes the identification model construction, an updated classifier (New classifier) (not shown) is generated and localized. The end transmission module 1031 is transmitted to the cloud server 102, and the identification mechanism layout group 1012 of the cloud server 102 is configured to deploy the update category identification mechanism to the at least one camera device 101, so that the camera device 101 Instant image recognition and monitoring will be performed based on the updated category identification mechanism.

Please refer to "3A" and "3B" for the implementation of the image analysis and selection of the present invention (1) ~ (2), as shown in the figure, when the image is analyzed After the selection module 1021 analyzes the plurality of image information on the image information, in order to avoid the identification model construction module 1032 in the construction of the identification model, an overfitting phenomenon is generated, that is, a map required for constructing the identification model is constructed. The information is too singular, so that the identification model can only analyze and learn for specific image information, which makes the update category identification mechanism still not the basis of the camera device 101, accurately for image recognition, and the image analysis and selection of the present invention. The module 1021 selects a plurality of image information obtained by analyzing the image information according to a variability criterion (not shown). As shown in FIG. 3A, the image information A1 and the image are displayed. The characteristics in the information A2, the image information A3, and the image information A4 are not much different, in other words, the variability is quite low (Low variance), so The image analysis and selection module 1021 will only select one image information, such as image information A2, and image information B1, image information C1, and image information D1, whether compared to each other or compared with image information A2. In fact, the difference in features in the image information is quite large, in other words, the variability is high (high variance), so in addition to the image information A2, the image information B1, the image information C1 and the image information D1 will also The image analysis and selection module 1021 is selected by the image analysis and selection module 1021. More specifically, the image analysis and selection module 1021 described in the present invention selects a plurality of image information with high variability according to the variability criterion. The model construction module 1032 performs the construction of the identification model.

Please refer to "Figure 4" for matching. The figure shows the implementation of the object mark of the present invention. As shown in the figure, when the image analysis and selection module 1021 completes the selection of multiple image information, The object tagging module 1022 marks at least one object of the image information. For example, the object tagging module 1022 marks "tree" and "person" in the image information A2, and "car" in the image information B1. The object marking module 1022 can be manually marked, but not limited thereto.

Please refer to "Figure 5", which shows the steps of the implementation of the present invention, and please refer to "Figure 1" and "Figure 2" together, as shown in the figure, the image recognition and monitoring system of the present invention The method S10 includes the following steps: an image capturing step S101: an image capturing module 1013 of the camera 101 performs image capturing, and the image information is transmitted through the image capturing device 101. The image transmission module 1015 is transmitted to a cloud server 102. The image recognition module 1014 of the camera device also performs image information according to a category identification mechanism pre-stored in a storage module 1012. Identification; an image analysis and selection step S102: after the image information is transmitted to the cloud server 102, an image analysis and selection module 1021 of the cloud server 102 analyzes the plurality of image information for the image information, and then According to a variability criterion, a plurality of image information of high variability is selected; an object marking step S103: the selected image information, an object marking module 1022 of the cloud server 102 will at least one of the image information The object is marked, such as the method of manual marking, and the image information after the completion of the marking is stored in a database 1024 of the cloud server 102; an identification model construction step S104: a local end of the local server 103 The transmission module 1031 is linked to the cloud server 102, and obtains the marked image information in the database 1024. An identification model construction module 1032 of the local server 103 is provided. Constructing an identification model based on the marked image information, and generating an updated category identification mechanism, wherein the identification model is constructed, and in the preferred embodiment, such as a deep convolutional neural network, but not For this reason, the update category identification mechanism will be transmitted to the cloud server 102 by the local end transmission module 1031; an identification mechanism deployment step S105: after the cloud server 102 receives the update category identification mechanism, the cloud server 102 The identification mechanism layout group 1023 deploys the update category identification mechanism to the camera device 101 for the camera device 101 to perform instant image information recognition according to the update category identification mechanism. To achieve security monitoring.

As can be seen from the above, the image recognition and monitoring system of the present invention includes at least one camera device, a cloud server and a local server. The camera device has a storage module pre-stored with a category identification mechanism. An image capture module, an image recognition module and a camera transmission module, and the cloud server has an image analysis and selection module, an object marking module and an identification mechanism cloth modeling group, and then The local end server has a local end transmission module and an identification model construction module, and the image recognition and monitoring system of the present invention implements an image capture module of the camera device to capture an image information. After that, the image recognition module can identify the image recognition module according to the category identification mechanism, and the image information will be transmitted to the cloud server by the camera transmission module, and the image analysis and selection module will parse the image information to obtain a plurality of images. Image information, and selecting a plurality of image information according to a variability criterion, the object marking module will at least one object in the selected image information Line marking, in addition, after the cloud server transmits the marked image information to the local server, the identification model construction module will construct an identification model according to the marked image information, and generate an update category. The identification mechanism, the update category identification mechanism will be transmitted to the cloud server through the local transmission module, and the identification mechanism identification group is used to construct the update category identification mechanism to at least one camera device, so that the camera device can be based on Updating the category identification mechanism to instantly identify the image information to achieve security monitoring; therefore, the invention can be implemented to provide a kind of server through the cloud and the local server. Collocation, effective identification model construction and category identification mechanism deployment, so that the camera device can accurately complete the image recognition image recognition and monitoring system and its implementation method.

The above description is only for the preferred embodiment of the present invention, and is not intended to limit the scope of the present invention; any changes and modifications made by those skilled in the art without departing from the spirit and scope of the invention All should be covered by the patent of the present invention.

In summary, the effects of the present invention are patents such as "industry availability," "novelty," and "progressiveness" of the invention; the applicant filed an invention patent with the bureau in accordance with the provisions of the Patent Law. Application.

10‧‧‧Image Identification and Monitoring System

101‧‧‧ camera

102‧‧‧Cloud Server

1021‧‧‧Image Analysis and Selection Module

1022‧‧‧ Object Marking Module

1023‧‧‧ Identification Mechanism Deployment Modeling Group

1024‧‧‧Database

103‧‧‧Local server

1031‧‧‧Local transmission module

1032‧‧‧ Identification Model Construction Module

Claims (10)

An image recognition and monitoring system for identifying an image information includes: a cloud server having an image analysis and selection module and an object tagging module, the image analysis and selection module of the image information Performing analysis and selection of a plurality of image information, the object marking module marking at least one object of the image information; a local server, which is connected to the cloud server, the local server has a Identifying a model construction module, the identification model construction module constructing an identification model based on the plurality of marked image information, and generating an update category identification mechanism; and at least one camera device, presenting information to the cloud server The image capture device has a micro processing module, and an image capture module, an image recognition module and a camera transmission module are respectively connected to the micro processing module, and the image capture module performs The image recognition module discriminates the image information according to the update category identification mechanism, and the camera transmission module images the image Drive information transmitted to the server. The image recognition and monitoring system of claim 1, wherein the cloud server has a database, the database stores a plurality of the image information, the image information, and the marked image information. . The image recognition and monitoring system of claim 1, wherein the local server has a local transmission module, and the local transmission module and the identification model construction module are linked by information, and the plurality of The marked image information is transmitted to the local server through the local transmission module, and the category identification mechanism is transmitted to the cloud server via the local transmission module. The image recognition and monitoring system of claim 3, wherein the cloud server has an identification mechanism cloth modeling group, and the identification mechanism cloth modeling group updates the category identification mechanism. The cloth is built on at least one of the camera devices. The image recognition and monitoring system of claim 4, wherein the camera device has a storage module and is connected to the micro-processing module, the storage module pre-storing a category identification mechanism for the The image recognition module of the camera device identifies the image information according to the category identification mechanism. The image recognition and monitoring system of claim 5, wherein when the identification mechanism modeling group deploys the update category identification mechanism to at least one of the camera devices, the category identification mechanism of the storage module is Update the category identification mechanism to replace it. An implementation method of an image recognition and monitoring system includes the following steps: an image analysis and selection step: an image analysis and selection module of a cloud server performs a plurality of image information analysis on an image information, and Selecting a plurality of the image information with high variability according to a variability criterion; an object marking step: after completing the selection of the plurality of image information, an object tagging module of the cloud server At least one object is marked, and the marked image information is stored in a database of the cloud server; an identification model construction step: a local end transmission module information of the local server is linked to the cloud The server obtains the marked image information in the database, and an identification model construction module of the local server, constructs an identification model according to the marked image information, and generates an update category identification mechanism. The update category identification mechanism is transmitted to the cloud server via the local end transmission module; an identification mechanism deployment step: the cloud server receives the After the new category identification mechanism, the cloud server identification mechanism of a set of cloth modeling category the update deployment mechanism to identify at least An imaging device for the camera device to perform identification of the image information according to the update category identification mechanism. The method for implementing the image recognition and monitoring system of claim 7, wherein the image analysis and selection step further comprises the following steps: an image capture step: an image capture of the camera device The module performs the capturing of the image information, and the image information is transmitted to the cloud server via a camera transmission module of the camera device. The method for implementing the image recognition and monitoring system of claim 8, wherein the image capturing module of the camera device captures the image information, and an image recognition module of the camera device is based on A category identification mechanism stored in a storage module performs identification of the image information. The method for implementing the image recognition and monitoring system according to claim 9, wherein the identification mechanism of the identification mechanism completes the deployment of the update category identification mechanism, and the category identification mechanism of the storage module is The image recognition module of the camera device performs the identification of the image information according to the update category identification mechanism.