Intelligent video monitoring and behavior analysis early warning system based on AI

Abstract

The invention relates to an AI-based intelligent video monitoring and behavior analysis early warning system in the field of video monitoring, when an AI processing platform terminal detects that a video has a non-standard behavior, a central server gives an alarm to a feedback layer, and when a crane is in an operation period, the central server identifies the non-standard behavior and then gives an alarm to the feedback layer. During operation, when a central server identifies non-standard behaviors and the non-standard behaviors cannot be processed in time through the feedback layer, the central server accesses an emergency center and contacts maintenance personnel to process the non-standard behaviors, and meanwhile, a light supplementing unit is designed to perform a light supplementing function according to the illumination intensity in the operation environment detected by an illumination detection unit. According to the AI intelligent video monitoring and behavior analysis early warning system, video data acquired by the data acquisition end are clearer, the illumination detection unit is in a double-detection mode, the illumination detection unit is in a normal working state for a long time, the AI intelligent video monitoring and behavior analysis early warning system has better environmental adaptability, and the practical effect is improved.

Description

Technical Field

[0001] The video surveillance and behavior analysis and early warning system involved in the present invention, in particular, relates to an AI-based intelligent video surveillance and behavior analysis and early warning system applied to the field of video monitoring. Background Art

[0002] An AI-based intelligent video surveillance and behavior analysis and early warning system is a system that uses artificial intelligence technology to analyze video content. Through deep learning algorithms, it can monitor video images in real time, identify and analyze behaviors in the images. The system can automatically detect abnormal behaviors or potential threats and immediately issue early warnings, thereby improving the efficiency and accuracy of security monitoring.

[0003] The specification of invention patent CN202410573491.2 discloses a monitoring method for an intelligent doorbell, including the following steps: obtaining a video captured by a camera of the doorbell, and identifying an object in the captured video to obtain an identified object video; removing pets from the identified object video to obtain a human identified object video. If the object in the human identified object video does not match a preset object in the atlas knowledge base, spatial parameters of the object in the human identified object video are extracted to obtain human behavior characteristics; determining whether there are abnormal behavior characteristics in the human behavior characteristics; if there are potential intrusion behavior characteristics, performing in-depth analysis on the potential intrusion behavior characteristics through a preset human pose estimation algorithm to assist in judging the intention and behavior of a human, and obtaining an in-depth analysis result; if the in-depth analysis result is that there is a dangerous behavior, pushing the video corresponding to the dangerous behavior to the user's terminal; ensuring that the user can quickly obtain potential risk information and improving the home security level.

[0004] The specification of invention patent CN202310910118.7 discloses an underground image enhancement method, device, equipment and medium based on LED control, including the following steps: performing initial illuminance setting on an LED lamp, and obtaining an underground low-illuminance image in an illumination area; obtaining an above-ground natural light image, and inputting all low-illuminance images and the above-ground natural light image into an image enhancement model for training; inputting the above-ground natural light image into an image quality evaluation model for training; inputting the low-illuminance image into the trained image enhancement model to enhance the low-illuminance image, and inputting it into the trained image quality evaluation model to obtain a current evaluation result; adjusting the illuminance of the LED lamp according to the current evaluation result, collecting images in real time, inputting the enhanced images into the image quality evaluation model for evaluation, comparing the current evaluation result with the previous evaluation result, and stopping adjusting the illuminance of the LED lamp until the current evaluation result is greater than or equal to the previous evaluation result. The video surveillance system of the present invention has good imaging quality and improves the generalization of intelligent recognition and detection of the AI model.

[0005] In the prior art, it is the prior art in this field to use recognition to identify and warn against specific behaviors, and at the same time use light to assist in processing low-illumination images to improve the versatility of recognition and detection. However, in an environment where the light conditions change greatly in a short period of time, such as the early morning and evening stages of outdoor work, using the real-time image quality to adjust the intensity of supplementary light will greatly reduce the data collection efficiency and affect the recognition and warning of images. Summary of the Invention

[0006] In view of the above prior art, the technical problem to be solved by the present invention is that using the implementation image quality to adjust the intensity of supplementary light will greatly reduce the data collection efficiency and cannot meet the work requirements.

[0007] To solve the above problems, the present invention provides an AI-based intelligent video surveillance and behavior analysis and warning system, including a central server, the central server is signal-connected to an edge node, the edge node is signal-connected to a data acquisition end, the data acquisition end is a camera, and an AI processing platform terminal is installed in the central server;

[0008] The central server is signal-connected to an emergency center, a feedback layer, an auxiliary end and a data storage layer. The feedback layer includes a sounder and an alarm lamp. The auxiliary end includes a supplementary light unit and a light detection unit;

[0009] The light detection unit includes a mounting base. The lower end of the mounting base is fixedly connected with a movable bin, a protective ring sleeve and a light sensor II. The movable bin, the protective ring sleeve and the light sensor II are distributed from outside to inside in sequence. The movable bin includes a bin wall. A sliding groove is formed on the inner wall of the bin wall. A sliding mounting table matching with itself is slidably connected in the bin wall, and the sliding mounting table is in contact with the protective ring sleeve. A plurality of ear blocks matching the shape of the sliding groove are fixedly connected to the outer wall of the sliding mounting table. A light sensor I is fixedly connected to the lower end of the sliding mounting table. A fixing ring matching with itself is threadedly connected to the opening of the bin wall. A plurality of compression springs are connected between the plurality of ear blocks and the fixing ring. A limiting column is fixedly connected to the inner wall of each of the plurality of sliding grooves. The limiting column penetrates through the ear block and the compression spring. An intensity groove is formed on the outer wall of the bin wall. The intensity groove is located above the sliding groove. An electric heating ring is fixedly connected in the intensity groove. One end of the electric heating ring away from the bin wall is fixedly connected with hot melt adhesive.

[0010] In the above-mentioned AI-based intelligent video surveillance and behavior analysis and warning system, it is possible to realize real-time makeup of the environment according to the light conditions, and increase the accuracy and efficiency of video data recognition.

[0011] As a further improvement of the present application, the unstandardized behaviors of the operators recognized by the AI processing platform terminal are manually verified and corrected through the feedback layer, and the corrected results can be fed back to the AI processing platform terminal, thereby improving the recognition accuracy and alarm feedback timeliness of the AI processing platform terminal.

[0012] As a further improvement of the present application, the AI processing platform terminal is autonomously trained by feeding "corpus", so as to improve the accuracy of model recognition.

[0013] As a further improvement of the present application, the edge node has data storage capacity. When the network signal is weak or the network is interrupted, the detected data can be stored in the node device and synchronized to the central server after the network is restored. The storage duration on the edge side is not less than 24 hours.

[0014] As another improvement of the present application, multiple supplementary lighting units and light detection units are arranged in the working environment at the auxiliary end and are evenly distributed in the columnar body of the activity range of the operator. The setting of multiple supplementary lighting units can avoid obvious shadows during the supplementary lighting process and reduce the difficulty of subsequent identification of irregular behaviors.

[0015] As a supplement to another improvement of the present application, a pair of connecting cables are fixedly connected between the mounting base and the sliding mounting table. Both connecting cables are made of elastic materials woven, so that the remaining parts of the broken activity bin are not likely to directly fall to the ground and are not likely to cause secondary injuries to the staff.

[0016] As a supplement to another improvement of the present application, an LED light board is fixedly connected to one end of the sliding mounting table close to the mounting base. The position of the LED light board matches that of the second light sensor, which can regularly detect the working state of the second light sensor and replace the faulty second light sensor in time to avoid the situation where both the second light sensor and the protective ring sleeve fail and the light detection unit fails.

[0017] As another improvement of the present application, a plurality of heat conduction units are fixedly connected to one end of the electric heating ring far from the bin wall. All the plurality of heat conduction units are in a three-dimensional spiral shape. On the one hand, it improves the heat conduction efficiency and speeds up the melting of the hot melt adhesive. After the first light sensor fails, the second light sensor can quickly enter the working position. On the other hand, the heat conduction units can capture most of the molten hot melt adhesive in the strength groove, reducing the phenomenon of the molten hot melt adhesive dripping and scalding the operator.

[0018] In summary, when the AI processing platform terminal detects irregular behaviors in the video, the central server will send an alarm to the feedback layer. When the crane is in operation, after the central server identifies irregular behaviors, it will shout at the operator through the feedback layer. During non-operation periods, when the central server identifies irregular behaviors and cannot handle them in time through the feedback layer, it will connect to the emergency center and contact the maintenance personnel for handling.

[0019] The supplementary lighting unit performs a supplementary lighting function according to the light intensity detected by the light detection unit in the working environment, making the video data collected by the data acquisition end clearer.

[0020] The light detection unit is designed in a dual-detection mode of light sensor one and light sensor two. During daily work, light sensor one is in the working state, while light sensor two is in the standby state. The light unit adjusts the light intensity according to the data detected by light sensor one to assist the data acquisition end in working. At the same time, it conducts daily detections on light sensor one and light sensor two to ensure that at least one of light sensor one and light sensor two is in the normal working state, enabling the AI intelligent video surveillance and behavior analysis early warning system to have better environmental adaptability and enhancing its practical effect. Description of the Drawings

[0021] Figure 1 Simplified diagram of the intelligent video surveillance and behavior analysis early warning system of the first implementation mode of this application;

[0022] Figure 2 Schematic diagram of the feedback layer of the first implementation mode of this application;

[0023] Figure 3 Schematic diagram of the auxiliary end of the first implementation mode of this application;

[0024] Figure 4 Simplified table of the recognition requirements of the intelligent video surveillance and behavior analysis early warning system of the first implementation mode of this application;

[0025] Figure 5 Simplified table of the early warning requirements of the intelligent video surveillance and behavior analysis early warning system of the first implementation mode of this application;

[0026] Figure 6 Structural schematic diagram of the light detection unit of the second implementation mode of this application;

[0027] Figure 7 Side sectional view of the light detection unit of the second implementation mode of this application;

[0028] Figure 8 For Figure 7 Schematic diagram of the structure at position A in

[0029] Figure 9 For Figure 7 Schematic diagram of the structure at position B in

[0030] Figure 10 Schematic diagram of the working state change of the light detection unit of the second implementation mode of this application.

[0031] Explanation of the reference numerals in the figures:

[0032] 1 Installation base, 2 movable bin, 201 bin wall, 202 sliding groove, 203 strength groove, 204 electric heating ring, 205 hot melt adhesive, 206 heat conduction unit, 3 fixing ring, 4 sliding installation table, 5 light sensor one, 6 LED light board, 7 light sensor two, 8 protective ring sleeve, 9 connecting cable, 10 limiting post, 11 compression spring. Detailed implementation manners

[0033] The following describes in detail two implementation manners of the present application with reference to the accompanying drawings.

[0034] The first implementation manner:

[0035] Figures 1 - 3 An intelligent video monitoring and behavior analysis and early warning system showing AI includes a central server. The central server is signal-connected to an edge node, and the edge node is signal-connected to a data acquisition end. The data acquisition end is a camera, and an AI processing platform terminal is installed in the central server;

[0036] The central server is signal-connected to an emergency center, a feedback layer, an auxiliary end, and a data storage layer. The feedback layer includes a sounder and an alarm light, and the auxiliary end includes a supplementary light unit and a light detection unit.

[0037] Among them, the AI processing platform terminal can identify people and their actions through the video data collected by the data acquisition terminal, and timely detect non-standard operations according to the currently non-standard behaviors stored in advance in the industry. This is well-known technology to those skilled in the art, so it is not disclosed in detail in the present application. Those skilled in the art can reasonably design the AI processing platform terminal according to the existing technology and meet the usage requirements of the present application.

[0038] Taking a lifting device as an example, the feedback layer, the data acquisition end, and the auxiliary end are all arranged in the operating room of the lifting device. Among them, multiple cameras should be set for the data acquisition end to ensure that there are no monitoring blind spots in the operating room of the lifting device. The data collected by the data acquisition end is transmitted to the central server through the edge node, and the video is processed by the AI processing platform terminal, and the data is correspondingly transmitted to the data storage layer for backup.

[0039] Please refer to Figures 4 - 5 , still taking a lifting device as an example, various behavior identifications in the scenario are divided into three priorities: high, medium, and low according to the urgency and difficulty. Among them, high-priority non-standard behaviors include alarm for non-staff intrusion, alarm for personnel using mobile phones, alarm for no video signal and movement or occlusion of the video detection range, and alarm for smoke and fire in the cab;

[0040] Medium-priority non-standard behaviors include alarm for personnel yawning, dozing off, smoking and eating behaviors, alarm for non-palm operation of buttons and handles, and alarm for abnormal postures such as personnel lying flat, crossing legs, and shaking legs.

[0041] Low-priority non-compliant behaviors include alarms due to personnel's emergency calls for help due to illness or unconscious state, alarms due to personnel's out-of-control language and emotions, and alarms for personnel's use of Bluetooth headsets.

[0042] During system deployment, the accuracy rates of identifying high-priority, medium-priority, and low-priority behaviors are required to reach 90%, 70%, and 50% respectively. After subsequent algorithm optimization and AI learning, the accuracy rates of identifying high-priority, medium-priority, and low-priority behaviors are required to reach 98%, 90%, and 80% respectively.

[0043] When the AI processing platform terminal detects non-compliant behaviors in the video, the central server will send an alarm to the feedback layer. When the crane is in operation, after the central server identifies non-compliant behaviors, it will shout at the operators through the feedback layer. During non-operation periods, when the central server identifies non-compliant behaviors and cannot process them in time through the feedback layer, it will connect to the emergency center and contact the maintenance personnel for handling.

[0044] The central server classifies, statistically analyzes, and analyzes the received video data, generates statistical charts according to actual needs, including bar charts, pie charts, line charts, etc. The central server has a warning storage function. After the AI processing platform identifies non-compliant behaviors of the operators, it intercepts the corresponding video information, forms a warning video, and stores it in the data storage layer. The cause of the accident can be found by retrieving the video. Various warning signals can be video-linked, and marks are added to the warning video data for subsequent query work.

[0045] The non-compliant behaviors of the operators identified by the AI processing platform terminal are manually verified and corrected through the feedback layer. The corrected results can be fed back to the AI processing platform terminal, thereby improving the accuracy rate of identification and the alarm feedback timeliness of the AI processing platform terminal.

[0046] At the same time, the AI processing platform terminal is autonomously trained by means of "corpus" feeding, so as to improve the accuracy rate of model identification.

[0047] The edge node has data storage capabilities. When the network signal is weak or the network is interrupted, the detection data can be stored in the node device and synchronized to the central server after the network is restored. The storage duration on the edge side is not less than 24 hours.

[0048] In the last auxiliary end, the supplementary lighting unit performs supplementary lighting according to the lighting intensity in the working environment detected by the lighting detection unit, making the video data collected by the data acquisition end clearer. In particular, multiple supplementary lighting units and lighting detection units in the auxiliary end should be set in the working environment and evenly distributed in the column within the activity range of the operator. The setting of multiple supplementary lighting units can avoid obvious shadows during the supplementary lighting process and reduce the difficulty of subsequent identification of irregular behaviors.

[0049] The second implementation method:

[0050] Figures 3 - 9 The lighting detection unit is shown, including a mounting base 1. The mounting base 1 is fixedly connected to the top of the working environment, such as the top of a crane. The lower end of the mounting base 1 is fixedly connected with a movable bin 2, a protective ring sleeve 8, and a second lighting sensor 7. The movable bin 2, the protective ring sleeve 8, and the second lighting sensor 7 are distributed from outside to inside in sequence. The movable bin 2 includes a bin wall 201. A sliding groove 202 is drilled on the inner wall of the bin wall 201. A sliding mounting table 4 matching itself is slidably connected in the bin wall 201, and the sliding mounting table 4 is in contact with the protective ring sleeve 8. A plurality of ear blocks matching the shape of the sliding groove 202 are fixedly connected to the outer wall of the sliding mounting table 4. A first lighting sensor 5 is fixedly connected to the lower end of the sliding mounting table 4. A fixing ring 3 matching itself is threadedly connected to the opening of the bin wall 201. A plurality of compression springs 11 are connected between the plurality of ear blocks and the fixing ring 3. A limiting post 10 is fixedly connected to the inner wall of each of the plurality of sliding grooves 202. The limiting post 10 penetrates through the ear block and the compression spring 11. A strength groove 203 is drilled on the outer wall of the bin wall 201. The strength groove 203 is located above the sliding groove 202. An electric heating ring 204 is fixedly connected in the strength groove 203. One end of the electric heating ring 204 away from the bin wall 201 is fixedly connected with a hot melt adhesive 205.

[0051] Among them, the movable bin 2 and the sliding mounting table 4 are made of non-light-transmitting materials, and the protective ring sleeve 8 is made of light-transmitting materials.

[0052] In this implementation method, the lighting detection unit is designed in a dual-detection mode of the first lighting sensor 5 and the second lighting sensor 7. During the daily work process, the first lighting sensor 5 is in the working state, while the second lighting sensor 7 is in the standby state. The lighting unit adjusts the lighting intensity according to the data detected by the first lighting sensor 5 to achieve the purpose of assisting the data acquisition end to work.

[0053] At the same time, the light sensor 5 is regularly detected for its working state during non-working hours. The lighting unit changes its light intensity regularly according to the preset detection instruction. When the light intensity data detected by the light sensor 5 matches the preset detection instruction, the light sensor 5 is in a normal working state and the working state detection passes. When the light intensity data detected by the light sensor 5 does not match the preset detection instruction, the light sensor 5 is in an abnormal working state. At this time, please refer to Figure 10 , the heating ring 204 works to generate a large amount of heat, melting the hot melt adhesive 205. The strength groove 203 can no longer obtain the strength support of the hot melt adhesive 205. Under the action of its own structural weight, the bin wall 201 will break along the position where the strength groove 203 is dug. Among them, the load-bearing limit of the movable bin 2 can be controlled by controlling the depth of the strength groove 203. The movable bin 2 breaks into two parts as a whole. One part falls with the light sensor 5 and exposes the light sensor 2 7, enabling the light sensor 2 7 to be converted into a working state. At the same time, in the part of the movable bin 2 that falls, under the action of the compression spring 11 in the compressed state, the sliding mounting table 4 and the light sensor 5 will move inward towards the bin wall 201 to protect the light sensor 5 and prevent further damage to the light sensor 5.

[0054] A pair of connecting cables 9 are fixedly connected between the mounting base 1 and the sliding mounting table 4. Both of the two connecting cables 9 are woven from elastic materials, making it difficult for the remaining parts of the broken movable bin 2 to directly fall to the ground and not easily cause secondary injuries to the staff.

[0055] One end of the sliding mounting table 4 close to the mounting base 1 is fixedly connected with an LED light board 6. The position of the LED light board 6 matches that of the light sensor 2 7, which can regularly detect the working state of the light sensor 2 7 and replace the faulty light sensor 2 7 in a timely manner to avoid the situation where both the light sensor 2 7 and the protective ring sleeve 8 fail and the light detection unit fails.

[0056] A plurality of heat conduction units 206 are fixedly connected to the end of the heating ring 204 far from the bin wall 201. The plurality of heat conduction units 206 are all in a three-dimensional spiral shape. On the one hand, it improves the heat conduction efficiency and speeds up the melting of the hot melt adhesive 205. After the light sensor 5 fails, the light sensor 2 7 can quickly enter the working position. On the other hand, the heat conduction units 206 can capture most of the molten hot melt adhesive 205 in the strength groove 203, reducing the phenomenon of molten hot melt adhesive 205 dripping and scalding the operators.

[0057] Intersecting with the first embodiment, the specific structure of the light detection unit is disclosed in this embodiment. The light detection unit is designed in a dual-detection mode of light sensor 1 5 and light sensor 2 7. During daily work, light sensor 1 5 is in the working state, while light sensor 2 7 is in the standby state. The light unit adjusts the light intensity according to the data detected by light sensor 1 5 to achieve the purpose of assisting the data acquisition end in working. At the same time, the daily detection of light sensor 1 5 and light sensor 2 7 is carried out to ensure that at least one of light sensor 1 5 and light sensor 2 7 is in the normal working state, so that the AI intelligent video monitoring and behavior analysis warning system has better environmental adaptability and increases its practical effect.

[0058] Combined with the current actual needs, the above-mentioned embodiment adopted in this application, the protection scope is not limited to this. Within the scope of knowledge possessed by those skilled in the art, various changes made without departing from the concept of this application still fall within the protection scope of the present invention.

Claims

1. An AI-based intelligent video surveillance and behavior analysis and early warning system, including a central server, characterized in that: The central server is signal-connected to edge nodes, and the edge nodes are signal-connected to data acquisition terminals. The data acquisition terminal is a camera, and an AI processing platform terminal is installed in the central server; The central server is signal-connected to an emergency center, a feedback layer, an auxiliary terminal, and a data storage layer. The feedback layer includes a sounder and an alarm light, and the auxiliary terminal includes a supplementary lighting unit and a light detection unit; The light detection unit includes a mounting base (1). The lower end of the mounting base (1) is fixedly connected to a movable bin (2), a protective ring sleeve (8), and a second light sensor (7). The movable bin (2), the protective ring sleeve (8), and the second light sensor (7) are distributed from outside to inside in sequence. The movable bin (2) includes a bin wall (201). A sliding groove (202) is formed in the inner wall of the bin wall (201). A sliding mounting table (4) matching the bin wall (201) is slidably connected in the bin wall (201), and the sliding mounting table (4) is in contact with the protective ring sleeve (8). A plurality of ear blocks matching the shape of the sliding groove (202) are fixedly connected to the outer wall of the sliding mounting table (4). A first light sensor (5) is fixedly connected to the lower end of the sliding mounting table (4). A fixing ring (3) matching the bin wall (201) is threadedly connected to the opening of the bin wall (201). A plurality of compression springs (11) are connected between the plurality of ear blocks and the fixing ring (3). A limiting post (10) is fixedly connected to the inner wall of each of the plurality of sliding grooves (202). The limiting post (10) penetrates through the ear block and the compression spring (11). An intensity groove (203) is formed in the outer wall of the bin wall (201). The intensity groove (203) is located above the sliding groove (202). An electric heating ring (204) is fixedly connected in the intensity groove (203). A hot melt adhesive (205) is fixedly connected to one end of the electric heating ring (204) away from the bin wall (201).

2. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, characterized in that: Irregular behaviors of operators identified by the AI processing platform terminal are manually verified and corrected through the feedback layer, and the corrected results can be fed back to the AI processing platform terminal.

3. The AI-based intelligent video surveillance and behavior analysis warning system according to claim 1, characterized in that: The AI processing platform terminal is autonomously trained by means of "corpus" feeding.

4. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, characterized in that: The edge nodes have data storage capabilities. When the network signal is weak or the network is interrupted, the detection data can be stored in the node devices and synchronized to the central server after the network is restored.

5. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, wherein: The supplementary lighting unit and the light detection unit of the auxiliary terminal are set in multiple numbers in the working environment and are evenly distributed in the columnar body of the operator's activity range.

6. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, characterized in that: A pair of connecting cables (9) are fixedly connected between the mounting base (1) and the sliding mounting table (4). Both of the two connecting cables (9) are woven with elastic materials, so that the remaining parts of the broken movable bin (2) are not likely to directly fall to the ground and are not likely to cause secondary injuries to the staff.

7. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, wherein: An LED light board (6) is fixedly connected to one end of the sliding mounting table (4) close to the mounting base (1). The position of the LED light board (6) matches that of the second light sensor (7).

8. The AI-based intelligent video surveillance and behavior analysis and early warning system according to claim 1, characterized in that: One end of the electrothermal ring (204) far from the bin wall (201) is fixedly connected with a plurality of heat conduction units (206), and the plurality of heat conduction units (206) are all in a three-dimensional spiral shape.

Images