AI-BASED IMAGE PROCESSING SYSTEM AND METHOD PROVIDING REAL-TIME RISK DETECTION AND EARLY WARNING OVER EXISTING CAMERA INFRASTRUCTURE.

Abstract

The invention relates to an artificial intelligence-based image processing system and method that detects events posing physical, behavioral, and environmental risks in real time using image data from the existing camera infrastructure and provides early warning to the authorized user according to the determined risk level. The system includes an image input unit that receives the image stream from at least one imaging unit, an edge processing unit that processes the image data locally, an artificial intelligence analysis unit that detects and tracks people, objects, and movements within the image, a risk assessment unit that evaluates the detected events with predetermined risk parameters, and a notification unit that transmits an alert to the authorized user terminal according to the established risk level.A risk score is generated by evaluating person movement, object movement, interpersonal positional relationships, movement intensity, movement continuity, and the temporal characteristics of the event, all obtained by the artificial intelligence analysis unit. If the risk score exceeds a predetermined threshold, the system generates an early warning containing the time and location information of the relevant event. Thanks to the processing of image data in the edge processing unit, there is no need to continuously transfer raw images to an external server, and the system can be integrated into the existing camera infrastructure. This invention is particularly useful for the early detection of risky events such as fights, physical assaults, falls, the use of dangerous objects, unusual movements, and similar incidents in controlled areas such as educational institutions and campuses.

Description

REAL-TIME RISK ASSESSMENT OVER EXISTING CAMERA INFRASTRUCTURE. AI-BASED SYSTEM FOR DETECTION AND EARLY WARNING IMAGE PROCESSING SYSTEM AND METHODS Technical Field to Which the Invention Relates The invention involves image processing, artificial intelligence, computer vision, real-time events and anomalies. detection, edge computing and early warning systems technical field It is related. More specifically, the invention utilizes live imagery obtained from existing imaging infrastructures. Analysis of workflows using artificial intelligence at a local edge processing unit; Identifying and determining the positional relationships between people, objects, movements, and individuals within the image. monitoring; multiple risk parameters of the obtained image characteristics Assessment in terms of; a risk associated with the identified event or behavior. the creation of a score and the risk score exceeding a predetermined threshold value in this case, a real-time early warning is sent to an authorized user. It relates to the system and method that provides it. The invention is particularly suitable for schools, educational institutions, university campuses, student dormitories, Controlled areas where children or other vulnerable user groups are present, social The existing camera infrastructure in facilities and similar areas only records images. It is transformed from a passive system that performs tasks into one based on real-time risk analysis. It is intended to be used as an active early warning mechanism. However, The invention's application area is not limited to educational institutions; it also includes real-time imaging. It can also be applied to different controlled areas requiring analysis and pre-event risk assessment. Prior knowledge of the art related to the invention. For security purposes, closed-circuit television (CCTV), IP cameras, and similar surveillance systems are used. The use of these systems has been known for a long time. These systems have different characteristics. Images are captured through cameras placed at various locations, and the captured images... It can be stored on a recording device or server. Known camera systems... 1 A significant part of it involves recording the image and allowing authorized users to access it when needed. It is based on the principle of monitoring. Detecting an incident that occurs in such systems is often the responsibility of the human operator. continuously monitoring the footage or restoring the recordings after the event occurred It depends on being examined in a focused manner. Especially in education with a large physical space. simultaneous image generation by numerous cameras in institutions and campuses Therefore, all camera streams are continuously and efficiently managed by human operators. It may not be possible to track it. In addition, some known image analysis systems show certain objects or simple things While it is possible to automatically detect movement patterns, Determining the risk level of an event based on only a single image feature This assessment can lead to false positive or false negative results. For example, the rapid movement of multiple people alone does not constitute a physical attack or It may not necessarily mean a fight. Similarly, a person falling to the ground is a sporting event. This could occur as a result of an activity, game, or accident, or some other risky event. This can also be the result. Therefore, instead of evaluating the movements of the person in the image in isolation; personal movements, person-to-person positional relationships, object movements, movement intensity, Multiple image features such as the continuity of the event, the location where the event occurred, and so on. A combined assessment will help determine whether the event poses a real risk. This is important for reliable determination. Purposes and Brief Description of the Invention The primary purpose of the invention is to capture live video streams from existing camera infrastructures. Analysis in real time using artificial intelligence-based image processing techniques. and predefined risky events or behaviors that occur in the images a system that automatically detects and provides early warning to authorized users and The aim is to develop a method. Another objective of the invention is to redesign the existing IP camera infrastructure. by enabling its use, it reduces the need for additional camera installations and the camera transforming their systems from passive systems that only record video 2 an active early warning mechanism that performs real-time risk analysis It is about transforming. Another purpose of the invention is edge processing of camera images. This approach enables the analysis of image data within a local processing unit. to reduce the need to constantly transfer it to an external server and thus a more controlled transaction infrastructure in terms of data transfer, latency and data security Another purpose of the invention is to enable the identification of people and objects within an image. movement characteristics and positional relationships between individuals, along with their monitoring and tracking. by enabling the evaluation of relationships, not just a single movement or image the false alarm rate that may occur in feature-based detection systems is the reduction. Especially during normal sporting activities, such as playing, running, or being in a crowd. more reliably separating risk-posing events from each other This is the aim. The current project document also contains errors regarding person / object tracking. Alarm management is envisioned within this scope. Another purpose of the invention is image physical and social behavioral indicators obtained from the data together by evaluating the situation, a risk score is created to determine the risk level of the event, and... The subject is different by comparing the score with predetermined threshold values. It is about generating early warning signals at various levels. Within the scope of the project, these warnings are authorized under different codes depending on the risk level. It is intended to be communicated to the user. Another purpose of the invention is to detect the event. Instead of being considered merely as an automated security output, the generated by allowing the warning to be reviewed by an authorized user. The goal is to maintain control within the system. In this context, the system... the risk warning created, the assessment of the relevant event and the necessary intervention It is presented to the authorized user for implementation. Thus, instead of the system automatically making disciplinary or punishment decisions, It functions as a decision support mechanism that supports human evaluation. This is provided. Another purpose of the invention is to provide educational institutions, universities in particular. Fights, physical assaults, falls, and dangerous objects in campuses and similar controlled areas. 3 its use at the moment when unusual mobility and similar risk-posing events occur or detection of the incident during its occurrence and reporting to the relevant authority to provide. In short, the invention utilizes live images taken from the existing camera infrastructure. transferring the stream to a local edge processing unit, artificially processing the image stream in question. It processes information through intelligence-based analysis units, identifying people, objects, and within the image. extracting characteristics related to movements, and the obtained characteristics are predefined. It evaluates in accordance with risk criteria and the determined risk level threshold. It sends an early warning to the authorized user if the limit is exceeded. Thus, the existing camera infrastructure will transform from a system that merely records events into something more. by being removed from a technical system that provides real-time risk detection and early warning. It is being transformed. Detailed Description of the Invention The invention utilizes image data obtained from existing camera infrastructure to utilize artificial intelligence and processing the image in real time using computer vision techniques Identifying information about people, objects, and movements within it, and different types of this information. Assessment at the analysis levels, identifying a risk related to the determined event or behavior. calculation of the score and the calculated risk score being assigned to one or more predetermined criteria. Early detection of risky situations by comparing them with higher threshold values. a system and method that enables this and notifies the authorized user It is related. Within the scope of the invention, the basic operating principle of the system is based on the existing imaging infrastructure. The live video stream received needs to be continuously monitored by a human. without being heard, the video stream in question is analyzed by an artificial intelligence mechanism. It is based on real-time evaluation. Thus, the camera system It is being transformed from a passive security infrastructure that only records images into one that records images. an active early warning system that can automatically detect specific events and behaviors within it It is being converted into a warning system. In one application, the system includes one or more IP cameras, closed-circuit television. 4 (CCTV) camera, network camera or similar imaging unit existing It works in conjunction with the camera infrastructure. Live footage is generated by the camera infrastructure. image streaming, image processing over a wired or wireless communication network. It is being transferred to the infrastructure. In the preferred application mode of the invention, the image stream is primarily where the image is received. a local processing unit located close to the physical environment, in other words The images are transferred to an edge processing unit. Thus, the images are processed... For analysis purposes, it is not necessary to transfer the data to a remote cloud server. In the current project structure, the image is processed on the local server and then transferred to the cloud environment. It is not expected to be sent. End processing units; central processing unit, graphics processor, artificial intelligence accelerator, memory the unit, the storage unit, the network communication unit, and the software that enables them to function. It may include one or more of the following components: the end processing unit. capacity depends on the number of cameras to be processed simultaneously, image resolution, and image frame rate. This can be determined based on the speed and the processing load of the artificial intelligence models to be implemented. After the camera image is received by the end processing unit, the image data is processed in a preliminary manner. It can be processed through a pre-processing stage. In this pre-processing stage, the image converting the resolution of the image to the level required by the artificial intelligence model. Adjusting the frame rate, reducing image noise, normalizing light levels. processing such as contrast adjustment, image field determination, and similar operations. It can be accomplished. Following the preprocessing stage, the image stream, including the people and objects within the image, is processed. It is transferred to an identification mechanism aimed at determining this. This mechanism, analyze a single frame of an image or multiple consecutive image frames by identifying the people, objects, and other predefined classes within the image. It determines the relevant location information. In person and object detection, it identifies the positions of the relevant objects within the image. bounding boxes, center points, area information, or similar positional indicators It can be created. For each person or object identified, the relevant identification information must be provided. The class it belongs to and its confidence level can also be determined. The detection process performed by the system is based on only a single image frame. It is not limited to object identification. Multiple consecutive image frames are analyzed. This allows the same person or object to be tracked over time. For this purpose, a tracking ID can be created for each identified person or object, and position changes of the person or object in successive image frames It can be monitored. Thanks to person and object tracking, it is possible to track not only the movement that occurs at a specific moment, It is also possible to assess the temporal continuity and direction of the movement. This is the case, for example, with a short, quick movement over a certain period of time. ongoing patterns of pursuit, approach, withdrawal, containment, or reciprocal movement They can be separated from each other. One of the key features of the invention is that risk assessment is based on a single image feature. Instead, it allows for the simultaneous evaluation of multiple image features. In this context, the system monitors person movement, object movement, and the distance between people, the relative positions of individuals, their directions of movement, their speeds of movement, and the intensity of their movement. The continuity of an event depends on the region where the event occurred and the temporal characteristics of the event. or more can be removed. These features are used to analyze events within the image at different layers of analysis. can be evaluated. In the preferred method of application, the analysis is at least a physical analysis. It includes multiple layers of analysis, including social / interactional analysis. The project includes three layers of analysis defined as physical, social, and psychological. It is located. At the physical analysis level, the physical characteristics of people and objects within the image are analyzed. Characteristics related to their movements are being evaluated. This includes sudden acceleration, running, falling, sudden change of direction, rapid approach, retreat, sudden movements towards each other movements, remaining motionless on the ground, and similar predefined movement patterns. It can be detected. 6 The physical analysis layer also examines whether specific object classes are present in the image. It can be assessed as not being present. For example, something defined as dangerous. The detection of an object within the image means that the object in question has been identified by a person. the movement of the object or the positional relationship of the object with a specific person, risk It can provide data to the evaluation mechanism. For an event to be assessed as a physical risk, it must involve the presence of a specific object or... It is not necessary to detect the movement. Multiple physical characteristics together. By evaluating the situation, the risk level of the event can be determined. Thus, for example, in a sporting event... Rapid movements occurring during the activity may constitute a physical assault. The aim is to differentiate between the movements. At the social / interactional analysis level, the relationships between multiple people within the image... Spatial and temporal relationships are evaluated within this scope. the ways in which they approach and distance themselves from each other, a person by more than one person being surrounded, a person's freedom of movement being restricted by other people limitations, the concentration of people around a particular person, and similar interactions. The models can be analyzed. Within the scope of social analysis, it is the physical isolation of one person from other people over a specific period of time. such as being separated or consistently existing alone in certain areas Behavioral indicators can also be used as an input for risk assessment. However, such indicators alone do not make any person risky, criminal, or It is not used for the purpose of classifying items as hazardous; only when necessary. as an indicator of support that can be evaluated by a qualified expert It is used. One of the fundamental technical elements used in social analysis is individual tracking. The system uses sequential methods. By correlating the findings that were determined to belong to the same person in the image frames, the relevant It is possible to create a person's movement trajectory over a specific time interval. This movement the trajectory of a person is compared with the trajectories of other people, thus creating a distinction between individuals. The level of interaction can be determined. 7 For example, during a sporting activity, multiple people moving in the same direction and at similar speeds as a person moves, they are surrounded by other people and the direction of their movement The situations in which the individuals in question are followed are different types of behavior. This allows the system to evaluate only the number of people in the image. It also includes the relationship of movement between individuals in the risk assessment. Another application of the invention involves the behavior of the person within the image. Features and visual characteristics obtained from the facial region can also be analyzed. In this context, specific expressions or situations are defined by visual characteristics of the facial region. An artificial intelligence model can be used to predict class classifications. Project This layer in the document analyzes facial expressions related to situations such as extreme stress or anger. It includes. This analysis alone does not constitute a definitive psychological or clinical diagnosis of an individual. It is not used for the purpose of creating. The result obtained is other physical and social Risk assessment is carried out by evaluating the indicators obtained from the analyses. It can be used as a visual feature to aid in the mechanism. This allows the system to, for example, observe physical motion occurring within the same time interval. change, interpersonal interaction change, and behavioral changes derived from the image. By evaluating the indicators together, we can get a more comprehensive picture than focusing on a single image feature. It is able to conduct an incident assessment. The data obtained from the analysis layers described above can be used for a risk assessment. It is transferred to the unit. The risk assessment unit obtains data from different layers of analysis. by evaluating the indicators obtained within the scope of one or more risk criteria, the relevant event It creates a risk score for them. The risk score can be calculated by weighting the different analysis outcomes. For example, physical movement indicator, person-to-person interaction indicator, object detection. The indicators related to the continuity of the event and the region where the event occurred are different. They can have varying weights. 8 In one application, the risk score is calculated using a function as follows: It can be calculated as follows: R = f(F, S, B, O, T, Z) Here; • R represents the generated risk score. • F, physical movement characteristics, • S, social / interactional characteristics, • B, behavioral or visual characteristics, • It, the characteristics of the object or event, • T represents the temporal continuity of the event. • Z, characteristics relating to the region where the event occurred. It expresses. The function in question is based on a fixed-weight scoring system, a learned machine learning approach. to the model, the classification model, rule-based evaluation, or these It can be achieved based on a combination of these. The risk score is calculated by detecting the event in a single image frame. The difference lies in the fact that it continues for a certain period of time in successive image frames. An assessment can be made. Thus, a short-term and low-reliability diagnosis is possible. an event that is continuous and supported by multiple layers of analysis A distinction can be made between them. The risk score generated by the risk assessment unit is at least a predetermined minimum. It is compared with a threshold value. When the risk score falls below the relevant threshold value... The system can continue image analysis without generating any critical warnings. A low or medium level warning is issued if the risk score exceeds the first threshold value. can be generated. If the risk score exceeds a higher threshold value. In this case, a high-priority alert can be generated. Within the scope of the project, this alert The levels are communicated to the user panel with different codes such as "Yellow" and "Red". It is anticipated. 9 One of the key technical aims of the invention is to reduce the intensity of motion or sudden movement. false alarms that may occur as a result of relying on individual features such as these is the reduction. For this purpose, the system uses temporal and spatial data obtained from tracking people and objects. This information is incorporated into the risk assessment. Thus, for example, a sporting activity... the rapid movement of multiple people, a ball or sports equipment the object moves and the people have similar movement trajectories When the characteristics obtained in this situation are evaluated together, it is seen that the event is a normal activity. The probability of this happening can be increased. In contrast, it involves a particular person constantly interacting with other people. The people approaching each other quickly, their directions of movement being opposite each other. change, surrounding a particular person or a particular object at risk Multiple indicators detected simultaneously within the same time interval, such as their movements. If this occurs, the risk score can be increased. Thanks to this mechanism, the system only detects "movement" or "many people present". Instead of making a simple decision in that way, it takes into account the temporal and spatial context of the event. This is ensured. The project document also mentions a fight between students playing basketball. The use of person / object tracking to differentiate students and the final The decision is expected to be made by a human. The risk score determined by the risk assessment unit exceeds the relevant threshold value. In this case, the system generates an event log and a corresponding alert. Alert log; information about the camera or area where the event occurred, and the time the event was detected. information, risk level, event type, relevant risk indicators, and system-generated data. It may include other technical information. In one application, the system captures the image segment or related image where the event was detected. also a link that provides access to the image to the authorized user. It can provide. The image itself is presented to the user by the system. depending on the defined access rights and relevant data processing policies It can be accomplished. The generated alert is sent to a local user terminal, computer, mobile device, or web-based platform. It can be transmitted to the user panel or a combination thereof. User The panel lists events from multiple cameras simultaneously, and the importance of those events is indicated. It is possible to rank them according to their level and examine the relevant event. Thus, instead of constantly monitoring all camera streams, the user can use the system It can prioritize events that are deemed risky and handle the relevant event. By examining the situation, they can carry out the necessary human intervention. The invention, particularly in its application form used in educational institutions, is implemented by the system. the warning issued is not a final disciplinary, punitive, or guilty judgment. It is not intended to be used. The system should be used by an authorized PDR specialist or institution. It functions as a decision support and early warning tool for designated authorized personnel. Authorized users can review the incident alert they received, including footage of the event. can evaluate the context and, if deemed necessary, can speak with students, teachers, and parents. or can initiate human-centered assessment processes with other relevant individuals. Thanks to this structure, automatic detection is performed by the artificial intelligence system. The assessment carried out by human experts is different from the project. The document also states that the system does not automatically impose penalties and the final decision rests with the psychological counseling specialist. Its presence is clearly foreseen. When used in educational institutions, the system outputs may label individuals as "potential criminals" or similar. Instead of a classification, "support" that can be submitted for expert evaluation when needed. It can be defined as "need" or "situation requiring investigation". Thus The output of the technical system is prevented from replacing pedagogical evaluation. In the preferred application of the invention, image data should be as close as possible to the maximum extent possible. analysis in a processing unit close to the physical environment in which the image was acquired 11 This ensures that the raw image data is continuously sent to an external data center. It does not need to be transferred. Instead of directly storing image data in the system, it is used for risk assessment. Extraction of necessary features and, where possible, processing of the raw image. Afterwards, it can be ensured that it is not held at all or is held for a limited period of time. In an application format, the integrity of image data and protection against unauthorized access hash or similar data integrity mechanisms to ensure its protection It can be used. The project document includes the hashing and processing of images and personalization. The data should only be viewed by an authorized PDR (Psychological Counseling and Guidance) specialist in case of a risk. an approach is envisioned. The system can also implement user-based access authorization. Accordingly, system administrator, guidance counselor, security personnel and other users' images, events And access levels to report data can be differentiated. If a user is not authorized to view the event footage, that user will be notified. only information such as the type of event, time information, camera or area information, and risk level. Technical information that does not include images can be displayed. The method developed within the scope of the invention initially involves one or more cameras Live video streaming is obtained by this. The resulting image stream is transferred to the edge processing unit, and the image is processed by artificial intelligence. It is converted into a format that analysis models can process. People and objects within the converted image stream are identified, and the identified individuals are... People and objects are tracked through successive image frames. Positional, temporal, and motion characteristics of the tracked individuals and objects. is being removed. 12 The extracted features are analyzed using physical, social / interactional, and, where necessary, behavioral methods. It is evaluated at different levels. By combining the results obtained from the analysis layers, a risk assessment is determined for the event or situation. A score is being generated. The risk score is compared against one or more threshold values. If the risk score does not exceed the relevant threshold value, the system returns to normal image analysis. The process continues; if the threshold value is exceeded, the event log and related early warnings are triggered. A warning is being generated. The early warning generated includes information about the location and time of the incident, along with relevant information, to the authorities. It is transferred to the user terminal. The incident is being investigated by the authorized user, and the relevant person will be contacted if deemed necessary. intervention or the psychological counseling process is initiated. In one application of the invention, artificial intelligence analysis models can be used with different cameras. using sample images obtained from various locations and different environmental conditions They can be trained or retrained. Educational data includes normal school activity, sports activity, play, crowding, running, falls, physical conflict, unusual personal behavior, and other predetermined events It may include tagged image or video samples related to their categories. An event obtained from the actual use of the system and verified by a qualified expert. The results will be further modeled by applying appropriate data security and anonymization processes. It can be used in development studies. This allows the system to be used in different school environments, with different camera angles and different lighting conditions. It is possible to improve the model's performance for operation under these conditions. It is possible. 13 The invention is not limited to specific brands or models of camera systems. The system integrates different cameras that provide images over the network to the existing IP camera infrastructure. to systems or other display systems with a suitable video output It is possible to implement it. Similarly, the end processing unit must be in the form of a single physical device. It is not. A distributed local environment consisting of multiple processors or processing devices. Processing architecture can be used. While in one application a separate end processing unit can be used for each camera, In another application format, the image stream from multiple cameras is the same local network. It can be processed in the processing unit. Also, some image processing operations are done on the camera side, and some operations are done at the edge processing level. a hybrid system where some operations can be performed on the unit and some operations can be performed on a central server. It can also be used in architecture. Risk threshold values ​​do not have to be the same for all camera systems. Camera location, intended use, area characteristics, time of day, user density, and institution Different threshold values ​​depending on the security policies determined by It can be defined. For example, high intensity of movement might be considered normal in a gym. While the weighting of physical movement indicators can be determined differently, corridor, stairs, entrance or the same motion indicator is higher in an area with low motion intensity It may have a risk weighting. The same technical infrastructure will also be used in the application forms of the invention outside of educational institutions. It can be used in university campuses, student dormitories, childcare centers, social facilities, sports facilities and other controlled areas requiring real-time risk assessment These areas can be evaluated within this scope. 14 Thus, thanks to this invention, images obtained from the existing camera infrastructure are only The data recorded for examination after the event is being removed and is being presented as real. It is used for time-based technical analysis. Physical, social, and behavioral aspects of person and object detection and tracking processes. By evaluating its features together, it is not based on a single image feature. A more comprehensive risk assessment mechanism is provided compared to the findings. Thanks to the edge processing approach, image analysis is performed close to the image source. This implementation reduces real-time processing latency and raw image reducing the need for continuous transfer of data to external systems It provides an opportunity. Comparison of the risk score with threshold values ​​and only under specific conditions If provided, an alert will be generated, alerts generated by the system. It contributes to prioritization and reduces the false alarm rate. By providing event-based alerts to the authorized user, the human operator's workload is significantly reduced. without needing to constantly monitor numerous camera streams, by the system This makes it possible to intervene in prioritized events. In particular, in applications carried out in educational institutions, the system automatically Instead of a mechanism that punishes or determines guilt, it could be risky. by identifying problems at an early stage and presenting them for evaluation by authorized human users. It operates as a technical decision support system. This approach is outlined in the project documentation. It is stated that "the system never automatically imposes penalties" and the final decision is made by a human expert. It is consistent with the principle of evaluation. The system and method described above explain the basic application form of the invention. The invention is granted for this purpose, and its scope is limited to the application methods described herein. It is not. Without changing the fundamental principle of the invention, different display units, artificial intelligence models, image processing techniques, risk assessment algorithms, communication Various protocols, processing architectures, and user terminals can be used.

Claims

1. Realizing image data obtained from an existing imaging infrastructure. By analyzing events in a timely manner, risks can be identified and early warning can be provided. It is an artificial intelligence-based image processing system that enables its production, and its feature is; • configured to receive live video streams from at least one display unit a video input unit (10), • the physical environment where the unit displaying the live video stream is located at least one edge processing configured to handle locally in the vicinity unit (20), • Detecting people and objects in the image stream and identifying the detected people and objects to follow by correlating successive image frames a structured person and object detection and tracking unit (30), • Positional, temporal, and motion characteristics of the people and objects being tracked. extracting and analyzing these features in at least two different layers of analysis. an AI-based analysis unit structured to evaluate (40), • joint evaluation of different analysis outputs obtained from the analysis unit a risk score for an event or situation detected in the related image stream a risk assessment unit structured to create (50), • the generated risk score with at least a predetermined threshold value a threshold evaluation unit structured for comparison (60) and • The incident is detected if the risk score exceeds the specified threshold value. Including information about the display unit or area, as well as the time of the incident and the risk level. to generate an early warning and send it to at least one authorized user terminal It should include a notification unit structured to (70) and • The risk score is based on the person and / or object, not on a single image feature. the evaluation of at least two different analysis outcomes regarding their movements created by It is an artificial intelligence-based image processing system characterized by its...

2. According to Claim 1, it is an artificial intelligence-based image processing system, the feature of which is artificial The intelligence-based analysis unit determines the speed and movement of people within the image. will determine the direction, duration of motion, intensity of motion and trajectory of motion It is structured in this way. 16 3. According to Claim 1, it is an artificial intelligence-based image processing system whose feature is the identification of a person and The object detection and tracking unit detects the same person or object in consecutive image frames. By associating the findings related to the object, a temporal movement is determined for each person or object. It is structured in a way that will create its orbit.

4. According to Claim 1, it is an artificial intelligence-based image processing system, the feature of which is artificial The first analysis layer of the intelligence-based analysis unit analyzes the physical movement of people and objects. It is structured in a way that will evaluate its characteristics and the physical properties in question. movement characteristics include at least one sudden change in speed, sudden change in direction, falling, running, fast It includes the characteristic of approaching, moving away, or reciprocating.

5. According to Claim 1, it is an artificial intelligence-based image processing system, the feature of which is artificial In the second analysis layer of the intelligence-based analysis unit, there are at least two people in the image. structured to evaluate the spatial and temporal relationship between them It is the fact that.

6. According to claim 5, it is an artificial intelligence-based image processing system whose feature is personal information. The spatial and temporal relationship between them is at least the point where people approach each other or distancing situation, a person being surrounded by more than one person, people in specific concentration around a person, reciprocal movement, or the trajectory of a person's movement being followed by another person or group of people, one or more of these characteristics It contains more than that.

7. According to Claim 1, it is an artificial intelligence-based image processing system whose feature is the identification of a person and object detection and tracking unit detects predefined objects within the image. It will identify their classes and the positional and temporal movements of the objects in question. It is structured in a way that allows it to follow certain instructions.

8. According to Claim 7, it is an artificial intelligence-based image processing system, the feature of which is risk the evaluation unit determines the type of object detected, the object's position within the image. its position, the movement of the object, and the positional relationship between the object and at least one person. structured to be used as input in the creation of a risk score It is the fact that. 17 9. According to Claim 1, it is an artificial intelligence-based image processing system, the feature of which is artificial intelligence-based analysis unit's behavioral and / or behavioral characteristics of a person within the image It will analyze visual features obtained from the facial region and these features to be used as an analytical input in the creation of a risk score It is structured.

10. An artificial intelligence-based image processing system according to claim 9, whose feature is facial recognition. visual features obtained from the region predefined expression or situation It is the process of generating a classification output relating to at least one of the classes.

11. According to Claim 1, it is an artificial intelligence-based image processing system whose feature is risk The risk score of the assessment unit is determined by; physical movement characteristics, interpersonal positional and temporal relationship characteristics, object detection characteristics, temporal continuity of the event, and by considering at least two of the characteristics of the region where the incident occurred It is structured in a way that will create 12. According to Claim 11, it is an artificial intelligence-based image processing system whose feature is risk by applying different weighting coefficients to different analysis inputs of the evaluation unit It is structured in a way that will generate a risk score.

13. According to Claim 11, it is an artificial intelligence-based image processing system whose feature is risk The evaluation unit is sequential with the detection of an event in a single image frame. Different risk scores depending on whether the image frames continue for a certain period of time. It is structured in a way that will create 14. According to Claim 1, it is an artificial intelligence-based image processing system, and its feature is thresholding. at least two different threshold values ​​corresponding to different risk levels of the assessment unit It is configured in a way that allows its use.

15. According to Claim 14, it is an artificial intelligence-based image processing system, the feature of which is risk If the score exceeds the first threshold value, it is at the first level and if it exceeds the second threshold value If it exceeds this level, it will receive a second-level early warning with a higher priority than the first level. is the creation of. 18 According to Claim 16, it is an artificial intelligence-based image processing system whose feature is risk the movement of people and objects to identify a risk event by the assessment unit considering both temporal continuity and the spatial relationship between individuals and to generate a risk warning based on the detection of only a single movement characteristic. It is not considered sufficient.

17. According to Claim 16, it is an artificial intelligence-based image processing system, the characteristic of which is: a predetermined definition of an activity that is defined as normal physical activity If movement and object characteristics are identified, then those characteristics pose a risk. It is used to reduce the score or suppress the risk warning. According to Claim 18, it is an artificial intelligence-based image processing system, the feature of which is edge-to-edge. The processing unit's image stream includes at least person and object detection, person and object one or more of the following processes: monitoring, feature extraction, and risk assessment to be carried out in the physical environment where the imaging infrastructure is located It is structured.

19. According to Claim 18, it is an artificial intelligence-based image processing system, the feature of which is edge-to-edge. The entire raw image stream of the processing unit is continuously transmitted to a remote server. to perform risk assessment without the need for transfer It is structured.

20. According to Claim 1, it is an artificial intelligence-based image processing system whose feature is image processing. to verify the integrity of data or data extracted from image data It is the implementation of a data integrity mechanism.

21. According to claim 20, it is an artificial intelligence-based image processing system whose feature is data. the integrity mechanism involves generating a hash value for the image data. It includes.

22. According to Claim 1, it is an artificial intelligence-based image processing system whose feature is image processing. restricting access to data based on user authorization and providing risk warnings. 19 If created, the relevant image will only be accessible to authorized users. It is made accessible from the terminal. According to Claim 23, 1, it is an artificial intelligence-based image processing system, the feature of which is early The alert includes the monitoring unit or area where the event occurred, the time of the event, and the risk level. and it must be created to include at least one of the following event types.

24. According to claim 23, it is an artificial intelligence-based image processing system, the feature of which is early The warning is displayed in a user panel, and the user panel shows the risk of events. will allow them to be viewed, sorted, and examined according to their level It is structured in this way.

25. According to claim 24, it is an artificial intelligence-based image processing system, the characteristic of which is: The authorized user can review the early warning generated by the system. access to relevant image data and human assessment of the event It is structured in a way that will provide this.

26. Using image data obtained from an existing imaging infrastructure an AI-based imaging system that performs real-time risk detection and early warning. It is a processing method, and its characteristic is; a) receiving a live video stream from at least one display unit, b) processing of the received live video stream at a local edge processing unit, c) Detecting people and objects in the image stream, d) Tracking the identified persons and objects through successive image frames being done, e) the positional, temporal and movement characteristics of the tracked persons and objects removal, f) Evaluation of the extracted features in at least two different layers of analysis, g) joint evaluation of analysis outputs obtained from different analysis layers by creating a risk score, h) the generated risk score with at least one predetermined threshold value comparison, i) the imaging unit where the incident was detected if the risk score exceeds the threshold value or an early warning regarding regional information, time of event and risk level creation and j) transmitting the generated early warning to an authorized user terminal It is an artificial intelligence-based image processing method characterized by its inclusion of several steps.

27. According to claim 26, it is an artificial intelligence-based image processing method whose feature is the person and temporal movement for each person or object during the tracking of objects. the creation of its trajectory and the relationships between these trajectories of motion It is a comparison.

28. According to claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is risk the score is based on person movements, interpersonal positional relationships, object movements and the event It is formed by considering at least two of its temporal continuities together.

29. According to claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is: Risk score is calculated by applying different weighting coefficients to different analysis inputs. is the creation of.

30. According to claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is: If the first risk threshold is exceeded, then the first level and the second risk threshold are exceeded. In this case, a second-level early warning system is established.

31. According to Claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is: a predetermined person relating to an activity defined as normal physical activity and if the object's motion characteristics are identified, the risks associated with those characteristics It is used as a mitigating factor in the creation of the score.

32. According to claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is: During the analysis of the image stream, the raw image data is continuously transmitted remotely. an endpoint positioned close to the display infrastructure without being transmitted to the server. It is processed in the processing unit. 21 33. According to Claim 26, it is an artificial intelligence-based image processing method, the characteristic of which is risk After the alert is generated, the relevant event is handled by the authorized user. allowing for investigation and automatic punishment by the system or the failure to make a disciplinary decision. 22