IMAGE PROCESSING-BASED INDOOR POSITIONING SYSTEM

Abstract

The invention relates to a system that determines user location with metric-level accuracy in indoor environments within mobile communication networks, where GPS signals are insufficient in enclosed spaces.

Description

1 TARIFF IMAGE PROCESSING-BASED INDOOR POSITIONING SYSTEM Technical Area The invention provides for precise user location in indoor environments within mobile communication networks. To determine this, user 5 can be used in indoor areas where GPS signals are insufficient. Metrics are calculated using equipment, camera sensors, and image processing algorithms. It relates to a system that detects location with a high degree of accuracy. State of the Art Traditional GPS technology is problematic because satellite signals are blocked by building walls and ceilings. Due to this, it does not work indoors or has a sensitivity loss of 50-100 meters. It yields extremely inadequate results. GPS multipath error is dramatically worse indoors. This is increasing, and the reflected signals are causing predictions that are far from the actual location. Furthermore, floor level information cannot be obtained via GPS in any way. It cannot be determined. WiFi and 15 are commonly used in current indoor location tracking solutions. Bluetooth Low Energy (BLE) beacon-based fingerprinting methods pose many serious risks. This has disadvantages. Firstly, these systems require a comprehensive training phase. This requires that all indoor areas be scanned beforehand and signal strength maps be created. These maps need to be created. They must be kept constantly updated, and WiFi access is required. relocation of points, addition of new devices or internal building arrangements 20 If done this way, it requires further training and creates an operational burden. Secondly, the sensitivity generally remains at the 5-10 meter level, which is sensitive. It is insufficient for network optimization. Thirdly, in multi-story buildings, the floor... Detection cannot be reliably performed. Cell-ID, Timing Advance (TA), Enhanced Cell-ID (E-CID) and 25 used by operators Network-based positioning techniques such as OTDOA (Observed Time Difference of Arrival), Extremely coarse sensitivity in indoor environments at distances between 50-500 meters. These techniques allow different locations within the same small cell coverage area. They cannot distinguish between them, therefore, for sensitive network optimization It is not usable. Especially in dense urban and indoor hotspot scenarios, it is too much. 2 Creating overlapping coverage areas of numerous small cells, Cell-ID based This makes the positioning completely meaningless. Current network optimization systems are unaware of users' precise indoor locations. Therefore, it is unable to localize the coverage problems and optimize the handover parameters. They are unable to configure it in this way and are conducting trial-and-error small cell deployment strategies. 5 They have to develop these methods. Drive test and walk test activities are expensive, It is time-consuming and far from reflecting the real-time network status. It shows which corridor users are in. Because it's unknown which floor or room it is on, this is the root cause of the coverage problems. Root cause analysis cannot be performed. In E911 / E112 emergency calls, current systems require the user to have sensitive indoor 10 unable to determine its location, which prevents emergency response teams from reaching the correct location in time. It prevents access. FCC (Federal Communications Commission) E911 Phase Regulation II specifies 50 meters horizontal and 3 meters vertical in indoor environments. It requires precision, but current technologies cannot meet these requirements. It cannot meet the demand. 15 Small cell (femto, pico, micro cell) layout planning in enclosed spaces allows users It cannot be done optimally because the actual distribution and movement patterns are unknown. Small cells are being placed in random locations, not in areas with poor coverage. this also limits capacity expansion and reduces ROI (Return on Investment) value. It reduces it. Without location information, it is not possible to determine which areas have high user traffic. It is not possible to objectively determine where coverage problems are occurring. Traditional handover algorithms determine the user's physical position and direction of movement. Because he doesn't know, he uses terms like ping-pong handover, too late handover, and early handover. They frequently experience problems with (too early handover). In indoor environments, Since it is unknown which corridor the user passed through or which floor they went to, proactive handover 25 Decisions cannot be made and service interruptions are occurring. Characteristics of the indoor area where the user is located (high traffic density) (area, quiet office space, lobby, corridor, meeting room, hospital intensive care unit, etc.) The same QoS policies are applied to all users regardless of location. Adaptive QoS mechanisms are not available. 30 3 In conclusion, due to the negative aspects described above and the current solutions being the subject of discussion... Due to its shortcomings, an improvement is needed in the relevant technical field. It has been made. Purpose of the Invention The invention was created by drawing inspiration from existing situations and addressing the aforementioned drawbacks. 5 It aims to solve the problem. The main purpose of the invention is to enable mobile communication networks in indoor environments. environments) precise user location determination and using this location information to network The goal is to ensure that its performance is optimized. The invention is particularly suitable for indoor environments where GPS (Global Positioning System) signals are insufficient. in areas (shopping malls, office buildings, airports, metro stations, stadiums, hospitals), user equipment (UE) camera sensors and Location detection with metric-level accuracy using image processing algorithms. for this purpose and for this location information to be used in GSM / LTE / 5G network optimization. It offers an improved integrated system and method. 15 The system described in the invention uses data obtained from user devices or indoor cameras. image data, computer vision techniques, deep learning learning) algorithms and visual Simultaneous Localization and Mapping (vSLAM) By processing the data using various methods, it determines the user's exact location (x, y, z coordinates and floor level). It determines the location. The determined location information is integrated with network management systems in real time. by optimizing small cell deployment and planning coverage area. planning), handover parameter adjustment, load balancing, position-based Quality of Service (QoS) prioritization and emergency services. (E911 / E112) improvement and many other network optimization applications It is used. 25 The invention is not limited to location determination alone, but also integrates the determined location data into a network. Correlation analysis was performed using performance metrics (RSRP, RSRQ, SINR, throughput). by keeping track of the poor indoor coverage heatmaps, Identifying coverage holes and localizing sources of interference. and strategic network planning such as determining optimal small cell placement points 30 It also includes its functions. 4 The system described in the invention uses image processing and computer vision algorithms to detect internal objects. Horizontal accuracy of 1-3 meters and floor level detection in indoor environments. It provides performance that GPS cannot offer. SIFT (Scale-Invariant Feature Transform) ORB (Oriented FAST and Rotated BRIEF), CNN (Convolutional Neural Network) based Location determination with 95%+ accuracy using feature extraction and visual matching algorithms. It is done. The invention does not require additional infrastructure such as WiFi / BLE beacons. It utilizes existing user devices. Implementation can be done using cameras or indoor security cameras. Quickly deployable for new buildings, easy to update building regulations. It is possible. Coverage can be continuously expanded using a crowd-sourced mapping approach. 10 The location information determined in the invention is integrated with network management systems in real time. By doing this, it provides dynamic network optimization. User location and network performance Real-time correlation analysis between metrics (RSRP, RSRQ, SINR, throughput) By doing this, coverage problems are localized and immediate action is taken. The invention combines location and signal quality data collected from all users to create 15 (crowd-sourced data aggregation), detailed interior coverage maps of buildings. Coverage heatmaps are created. Weak coverage areas (coverage holes) are identified, and high coverage areas are identified. Areas (high interference zones) and optimal signal quality regions are visualized. Maps enable network planning teams to make strategic decisions. The invention addresses the actual distribution of users, movement patterns, and coverage problems in 20 By utilizing their locations, small cells are placed in the most optimal positions. Traffic congestion is predicted using machine learning algorithms, thus increasing capacity. The regions where the expansion is most needed are identified. Small cell deployment ROI 40-60%. It is improved by a certain percentage. The invention enables proactive handover 25 by knowing the user's current position, direction of movement, and speed. Decisions are made. As the user passes through a corridor, the target cell at the end of the corridor... It is prepared in advance. For floor changes (using elevator / stairs), vertical handover. It is optimized. The ping-pong handover rate is reduced by 70-80%, and the handover success rate is 95%+. It is raised to that level. The invention features adaptive QoS 30 based on the characteristics of the indoor environment in which the user is located. Policies are implemented. In hospital operating rooms, train stations, stadium VIP areas. Special QoS guarantees are provided for priority users in these sections. High traffic Traffic shaping and admission control strategies in high-density areas are dynamic. It is adjustable. For E911 / E112 emergency calls, the user's precise indoor location (building name, floor, room) is required. (number, coordinates) automatically PSAP (Public Safety Answering Point) 5 It is transmitted to their systems. Full compliance with FCC regulations is ensured. Emergency response teams The time to reach the correct location is reduced by 60-70%, significantly increasing the life-saving rate. The invention combines users' location information and SINR measurements to determine the physical location of their resources. Locations are identified. Neighboring cell (inter-cell interference), co-channel, adjacent By identifying the floors and rooms where channel sources are located, targeted reduction strategies can be implemented. 10 (ICIC - Inter-Cell Interference Coordination, eICIC - enhanced ICIC, FeICIC - Further (Enhanced ICIC) is applied. The invention examines users' indoor movement patterns, the areas they visit, and the time they spend there. Time periods are analyzed using machine learning to predict future movement and traffic patterns. It is predicted. Proactive network resource allocation and traffic engineering are performed. Special event 15 During peak times (concerts, matches, conferences), the traffic density inside the building can be predicted in advance. A temporary capacity increase is planned. The system described in the invention uses all collected location and performance data to perform in-depth analysis. It continuously improves itself through learning models. Location accuracy increases over time. Network optimization strategies are fine-tuned, and anomaly detection capabilities improve. 20 Self-learning and self-optimizing features minimize the need for manual intervention. It is done. In addition to camera data, the invention incorporates IMU (Inertial Measurement Unit), WiFi, BLE, and UWB. (Ultra-Wideband), barometric pressure sensor, magnetometer, and multiple other sensors. The incoming data is combined using Kalman filtering and particle filtering to determine location. 25 Reliability and robustness are increased. When a sensor fails or unreliable data is obtained When it fails to provide the necessary signal, other sensors take over, ensuring uninterrupted service. The structural and characteristic features and all the advantages of the invention are given in the figures below. This becomes clearer thanks to the detailed explanation written with references to these figures. This will be understood, and therefore the evaluation should also take these forms and detailed explanations into account. 30 It needs to be done by taking precautions. 6 Figures that will help understand the invention. Figure 1 shows a representative block diagram of the system described in the invention. Explanation of Part References 1. Image capture interface 2. Image preprocessing module 5 3. Visual feature extraction engine 4. Visual database and mapping system 5. Location estimation engine 6. Sensor fusion module 7. Real-time tracking module 10 8. Indoor map storage 9. Network measurement correlation engine 10. Indoor coverage analysis module 11. Small cell layout optimization unit 12. Transfer parameter optimization engine 15 13. Service quality policy implementation module 14. Network configuration interface 15. Performance analytics dashboard Detailed Description of the Invention This detailed description explains the image processing-assisted indoor location 20 that is the subject of the invention. The identification system's preferred structures contribute not only to a better understanding of the subject. It is explained in this regard. Image capture interface (1), user equipment (smartphone, tablet, AR glasses) video streams from rear / front cameras or indoor security cameras It captures streams. Images in H.264 / H.265 video encoder (codec) formats. It decodes (solves). It performs frame extraction. (typically 1-5 frames per second). Video streamed via RTSP, HTTP, and WebRTC protocols. It receives the data. 7 Image preprocessing module (2) prepares raw images for processing: noise reduction (Gaussian blur, median filtering), contrast enhancement (histogram equalization, CLAHE), perspective correction, motion blur removal, lighting normalization (illumination normalization). Image enhancement in low light conditions. It applies (low-light enhancement) algorithms. 5 The visual feature extraction engine (3) extracts distinguishing visual features from images (visual SIFT (Scale-Invariant Feature Transform) extracts features. unaffected by scale and rotation changes using the Transform algorithm. Keypoints are obtained. ORB (Directed Fast and Rotated) BRIEF - Oriented FAST and Rotated BRIEF) algorithm with fast binary 10 identifiers, SURF (Speeded Up Robust Features) With this algorithm, robust visual features are extracted. Visual feature extraction Within the scope of the deep learning approach via the engine (3), pre-trained CNN Higher-level models (ResNet, VGG, and MobileNet) are used for Convolutional Neural Network (CNN) analysis. High-level semantic features are obtained. Typically, 500 for each image. Between 2000 and 2000 key points are identified. Visual database and mapping system (4) pre-mapped building interiors It stores and manages the visual database. Each reference location (reference location) obtained from multiple viewpoints Images and their attribute vectors are indexed and stored. The visual database and 20 In the mapping system (4), reference similar to query image The KD-tree was created so that images could be found within milliseconds. Hierarchical k-means and Locality-Sensitive Hash (LSH) Fast lookup data structures such as Sensitive Hashing are used. Approximately 10,000 m² For an interior space of this size, between 50,000 and 100,000 reference images. It is stored. The position estimation engine (5) uses features extracted from the query image. It matches reference attributes found in the database (feature matching). RANSAC Outliers were identified using the Random Sample Consensus algorithm. Values ​​(outliers) are filtered and geometric validation is performed. PnP (Perspective-n-30) Point – Perspective-n-Point) or EPnP (Efficient Perspective-n-Point) three-dimensional–two-dimensional (3D-2D) point matching using algorithms (correspondences) The camera's exposure (position and orientation) is calculated. Triangulation 8 (triangulation) and bundle adjustment processes are applied to reposition The accuracy of the determination is increased. As a result, the user's (x, y, z) coordinates and floor level are accurately determined. Floor level and heading angle are determined. The sensor fusion module (6) enables the merging of multiple sensor data (multi-sensor fusion). This is accomplished by; visual position estimation obtained from the camera, IMU 5 Accelerometer and gyroscope provided by the Inertial Measurement Unit. (gyroscope) and magnetometer data, WiFi RSSI fingerprint Fingerprinting information, proximity data of BLE beacons, barometric data. Floor level detection is performed using a pressure sensor and UWB (Ultra Wide) Band) distance measurement (ranging) data are evaluated together. Extended Kalman Filter 10 (Extended Kalman Filter - EKF) or Unscented Kalman Filter (Unscented Kalman Optimal state estimation is performed using a filter (UKF). In addition, reliability scores are calculated for each sensor. Adaptive weighting is applied. The real-time tracking module (7) continuously monitors the user's instantaneous location. 15 Nonlinear and non-Gaussian Particle Filter (Particle Filter) is used to ensure reliable monitoring in various environments. The Sequential Monte Carlo method is used. Visual odometry and IMU are also employed. dead reckoning techniques based on frame-to-frame sequential image processes Movement estimation is performed between frames. Loop closure detection (loop closure 20 position deviations (drifts) that occur during long-term tracking are detected by applying (detection) method. It is corrected. The system typically operates at a tracking update rate of 5-10 Hz. Also, obfuscation... (occlusion) management and re-initiation of follow-up after loss of follow-up It also includes (tracking recovery) mechanisms. The indoor map repository (8) contains floor plans of buildings, three-dimensional (3D) 25 models and semantic information (room types, hallways, staircases, elevators, and emergency exits) (such as locations) stores. CAD (Computer-Aided Design) format (DWG, DXF) It parses (aligns) floor plans to the building coordinate system. Geographic Information It provides integration with the system (GIS - Geographic Information System) data. Additionally... building address, building height, number of floors and type of use (office, shopping mall, hospital 30 It stores and manages building metadata (such as building information, etc.). Network measurement correlation engine (9), from the determined user location and user equipment (UE) RSRP within the RRC (Radio Resource Control) measurement reports received. 9 (Reference Signal Received Power), RSRQ (Reference Signal Received Quality), RSSI (Received Signal Strength Indicator) and SINR (Signal-to-Interference-plus-Noise Ratio) It performs correlation analysis between the values. For each location point, the average is calculated. Minimum and maximum signal level values ​​are calculated between the measurement points. Signal quality in the remaining regions, kriging and inverse distance weighting (Inverse 5) Spatial interpolation methods such as Distance Weighting (IDW) It is estimated using time-series analysis. Changes in indoor coverage performance over time are monitored and analyzed. Indoor coverage analysis module (10), collected location and measurement data Indoor coverage heatmaps are created using this method. Guide 10 Within the framework of grid-based representation, for example, 1 × 1 meter Coverage quality for each cell is determined using grid cells of a specific size. The coverage quality score is calculated. Based on the calculated values, the regions are categorized as follows: coverage holes (RSRP < -110 dBm), weak coverage areas (-110 dBm) 15 as dBm < RSRP < -95 dBm**) and good coverage areas (RSRP > -95 dBm). They are classified. Additionally, cells are divided into dominant server cells and other cell types. Cell overlap areas are identified. Points with high interference intensity. Interference hotspots are located based on the SINR < 0 dB condition. Small cell layout optimization unit (11), optimum small cell It determines the settlement points. Coverage holes, high traffic 20 high traffic density zones, areas with low SINR levels, and Multi-objective optimization (MOO) taking user density distribution into account. Optimization is carried out. The optimization goals are to improve coverage. (improvement) and maximizing capacity increase [maximize(coverage_improvement, capacity_increase)], intercellular interference (inter-25 minimizing cell interference and deployment costs The genetic algorithm is structured as follows: [minimize(inter-cell_interference, deployment_cost)]. (Genetic Algorithm), simulated annealing, or particle swarm Particle Swarm Optimization (PSO) methods are used for candidate placement. points, three-dimensional ray-tracing propagation models 30 It is simulated using [method name]. The output is prioritized small cell placement locations. Estimated coverage and capacity improvement values ​​are obtained. The transfer parameter optimization engine (12) uses the user's position, direction of movement and speed. using location-aware handover parameters It performs optimization. Each handover boundary location (handover boundary) optimum hysteresis, time-to-trigger (TTT) and trigger time for (location) Calculates A3 offset values. Critical areas like corridors, staircases, and elevators. Proactive handover triggering mechanisms in mobility zones Definitions. Specifically designed for vertical handover operations related to floor changes. It establishes handover policies. Frequent and unnecessary handovers are referred to as ping-pong handovers. By identifying regions of cell change, it increases hysteresis values ​​in these regions. Using machine learning methods, the probability of successful handover is 10. A handover success prediction is made. Service quality policy implementation module (13), indoor location of the user semantic information related to the region (for example; hospital operating room, stadium) (VIP area, airport business class waiting lounge, conference room) can be adapted depending on the situation. It implements Quality of Service (QoS) policies. Location-based QCI (QoS Class Identifier - 15) QoS Class Descriptor) mapping, GBR (Guaranteed Bit Rate) Allocation and ARP (Allocation and Retention Priority) It performs the adjustment. In areas defined by geo-fencing. It enforces custom QoS rules. Policy and Pricing Rules Function (PCRF - It provides integration with the Policy and Charging Rules Function. 20 Network configuration module (14) optimizes network parameters based on eNodeB / gNB. It applies configuration management processes to its stations. TR-069 performs the handover via NETCONF and SNMP protocols. parameters, QoS policies and admission control threshold values (thresholds) updates. SON (Self-Organizing Network) 25 It provides coordination with its functions and in this context, load balancing (MLB - Mobility) Load Balancing), Mobility Robustness Optimization (MRO) Optimization), RACH (Random Access Channel) optimization and energy saving. It supports (energy saving) functions. Adjacent cells via X2 / Xn interfaces. It performs parameter coordination with (neighbor cells). 30 The performance analytics dashboard (15) monitors, analyzes and performs system performance. reports. KPI (Key Performance Indicator) Within the scope of the calculations; positioning accuracy, 11 positioning availability, coverage improvement rate (coverage improvement rate), handover success rate, ping-pong ping-pong ratio, user throughput improvement performance improvements such as small cell offload rate. It evaluates the metrics. Interactive 3D visualization (interactive 3D 5) Indoor coverage maps are created using visualization (for the user). distribution heatmaps (user distribution heatmaps) and handover flow diagrams (Handover flow diagrams) are used for visualization. Machine learning. anomaly detection using methods, predictive maintenance Maintenance and capacity forecasting are performed. Grafana, Kibana 10 and performance indicators are provided through integration with Microsoft Power BI platforms. Dashboards are created. Image Capture Interface (1), from the rear camera of the user equipment (UE) It receives video streams at 1920×1080 resolution. Using an H.264 codec. The encoded video stream is decoded in real time at 3-5 15-second intervals. The image is transmitted to the pre-processing module (2) at a rate of 3-5 frames per second (fps). In low light conditions automatic exposure adjustment and frame rate adaptation Adaptive frame rate (AMP) operations are performed. The image preprocessing module (2) performs sequential preprocessing on each received image frame. performs operations: 20 (a) High-frequency noise elimination For this purpose, Gaussian blurring is applied (σ=1.5). (b) CLAHE for local contrast enhancement (Contrast Limited Adaptive Histogram Equalization) The Adaptive Histogram Equalization method is applied (clip limit=2.0). 25 (c) Perspective when the camera tilt angle is greater than 15° Perspective distortion correction is performed. (d) Motion blur detection and Wiener inverse convolution Filtering (Wiener deconvolution filtering) is applied. Preprocessing latency is 15-25 ms per image frame. 30 12 Pre-processed images are input data to the visual feature extraction engine (3). The information is transferred. The engine runs three different feature extraction algorithms in parallel: (a) SIFT (Scale-Invariant Feature Transform) (Transformation) sensor using an average of 800-1200 per image frame. 128 keypoints are identified, and for each keypoint, 5-dimensional grids of 128 dimensions are created. The descriptor is calculated. (b) Calculation using ORB (Oriented FAST and Rotated BRIEF) sensor For efficiency purposes, 500-1000 binary keypoints and 256-bit An identifier is created. (c) Pre-trained ResNet-50 CNN (Convolutional Neural Network - 10 From the conv5 layer of the Convolutional Neural Network (CNN) model, a general feature of size 2048 is obtained. The vector (global feature vector) is extracted. The total feature extraction time is between 80-120 ms. The extracted features are already in the visual database and mapping system (4). It is matched with calculated reference properties. In the database, PostgreSQL + PostGIS 15 75,000 references were spatially indexed using [method name]. The image shows an office building of approximately 15,000 m². Hierarchical keywords for Approximate Nearest Neighbor Search The term "Hierarchical Vocabulary Tree" is used:  Branching factor: 10 20  Depth: 6 The recovery time for the most similar Top-K (typically K=20) reference image is 30-50 ms. It is among them. The position estimation engine (5) uses the obtained reference images and the query image (query RANSAC (Random Sample 25) performs geometric validation between images. Outlier feature matching using the Consensus (Random Sample Agreement) algorithm. (outlier feature matches) are filtered by:  Number of iterations: 1000  Inlier threshold: 5 pixels  The inlier ratio is typically in the 40-60% range. 30 13 Using an EPnP (Efficient Perspective-n-Point) solver, a minimum of 4 intrinsic elements can be identified. The camera exposure [R|t] is calculated from the inlier correspondence. This exposure is calculated based on rotation. It consists of a rotation matrix and a translation vector. When the camera satisfies the condition of reprojection error < 2 pixels The pose is considered valid. 5 The calculated camera position is converted to the building's global coordinate system and used as a reference. It is aligned with the reference coordinate system. The results obtained are as follows:  User's coordinates in meters (x, y, z) (for example; x=23.5 m, y=12.8 m, z=7.2 m), 10  Floor level is determined by deriving it from the z-coordinate.  Heading angle: 𝜙 = 𝑎𝑡𝑎𝑛2(𝑅, [1, 0]𝑅[0, 0]) It is calculated as follows: 15 Position estimation uncertainty, covariance matrix. Calculated using a covariance matrix. Typical values:  Horizontal position accuracy: σxy = 1.2 - 2.5 meters  Vertical position accuracy: 20 σz = 0.5 - 1.5 meters The Sensor Fusion Module (6) combines visual-based position estimation with other sensor measurements. fusion Dead cells are detected using accelerometer data provided by the IMU (Inertial Measurement Unit). Dead reckoning is performed. 25  Sampling rate: 100 Hz The change in position is calculated as follows: 14 Δ𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 = ∫ (𝑎𝑐𝑐𝑒𝑙𝑒𝑟𝑎𝑡𝑖𝑜𝑛 − 𝑔𝑟𝑎𝑣𝑖𝑡𝑦)𝑑𝑡 Heading angle drift is corrected using gyroscope data. Absolute heading reference using magnetometer data. is obtained. 5 Using WiFi RSSI values ​​(typically 5-10 access points are detected), a weighted average can be obtained. Auxiliary positioning using the weighted centroid localization method. The estimate is made. BLE beacons provide proximity information by estimating distance from RSSI values: d = 10 ( ) / ( ∗ ) Here:  d: estimated distance,  TxPower: beacon transmission power,  RSSI: Received signal level, 15  n: ambient coefficient (n=2-3) It is used as. Within the Extended Kalman Filter (EKF) structure, the state vector is continuously represented. Updated: 𝑋 = [𝑥, 𝑦, 𝑧, 𝑣, 𝑥, 𝑣, 𝑦, 𝑣𝑧𝜙𝜔] Here:  x,y,z: location information,  vx, vy, vz: velocity components,  φ: orientation angle, 25  ω: angular velocity. Prediction step: 𝑋 ∣ = 𝐹 ⋅ 𝑋 + 𝐵 ⋅ 𝑢 Here:  F: state transition matrix,  uₖ: These are IMU measurements. In the Update Step, the following measurement models are used: 5  Visual-based location information: 𝑖𝑛𝑛𝑜𝑣𝑎𝑡𝑖𝑜𝑛 = 𝑧 − 𝐻 ⋅ 𝑋  WiFi location measurements,  BLE proximity constraints. 10 Sensor reliability scores are adjusted dynamically:  Visual positioning reliability: 𝜎 = 𝑓(𝑛, 𝑢𝑚𝑏𝑒𝑟_𝑜𝑓_𝑖𝑛𝑙𝑖𝑒𝑟𝑠𝑟 𝑒𝑝𝑟𝑜𝑗𝑒𝑐𝑡𝑖𝑜𝑛_𝑒𝑟𝑟𝑜𝑟)  WiFi reliability: 15 𝜎 = 𝑓(𝑛, 𝑢𝑚𝑏𝑒𝑟_𝑜𝑓_𝐴𝑃𝑠𝑅𝑆𝑆𝐼_𝑣𝑎𝑟𝑖𝑎𝑛𝑐𝑒) EKF update frequency: 10 Hz. The real-time tracking module (7) uses the EKF output to track the user's movement. It follows its trajectory. Modeling multimodal probability distributions 20 For this purpose, a particle filter is used. Parameters:  Number of particles: N=500 Each particle represents a possible position hypothesis. Resampling In this stage, low-weight particles are eliminated. Visual odometry 25 Relative motion estimation between successive images is performed using: [[Δx, Δy, Δz, Δφ] 16 This process:  Lucas-Kanade optical flow method,  Feature tracking algorithms It is carried out using. Place recognition for loop closure detection 5 It is applied. In this context:  The Bag-of-Words model is used.  If a location previously visited by the user is detected again, the accumulated location is displayed. The drift error is corrected. 10  Pose graph optimization is applied. The specified user location is determined by timestamp, UE ID, x, y, and z. with coordinates, floor information and accuracy value, indoor map repository (8) aligns. The user's location (room, corridor, or area) is determined semantically. 15 For this process:  Spatial point-in polygon testing on floor plan polygons (polygon test) is performed.  The geometric region in which the user's coordinates are located is determined. Example of the result obtained: 20 "User: Building_A, Floor_3, Conference_Room_301, coordinates: (23,5; 12,8; 7.2), accuracy: ±1.8 m This semantic location information is then presented in JSON format. The data is transferred to the downstream modules. In parallel, RRC measurement reports (RRC: 25) from the user equipment (UE) MeasurementReport messages are collected by the network and fed into the network measurement correlation engine. (9) is directed. 17 The measurement report includes the following parameters:  Pertaining to the serving cell: RSRP (Reference Signal Received Power) (Power), o RSRQ (Reference Signal Received Quality) Quality),  Relating to neighboring cells: o RSRP, o RSRQ, o RSSI (Received Signal Strength Indicator), 10  Timing Advance value,  CQI (Channel Quality Indicator). The measurement reporting period is typically 480 ms - 1920 ms, depending on the configuration. It is within the range. The collected measurements are synchronized using timestamp synchronization. 15 It is applied and matched with location data: (location,RSRP,RSRQ,SINR,CQI,timestamp) The network measurement correlation engine (9) stores the collected location and measurement pairs in the following format: It stores in a time-series database: 20 (location, measurements) Data is stored in InfluxDB. The following spatial tags are assigned to each measurement record:  Building ID, 25  floor ID  room ID  Grid cell ID. The following analyses are performed by running continuous queries: (a) The following RSRP statistics are calculated for each grid cell: 30  Average value (mean),  Standard deviation (std),  Minimum value (min),  Maximum value (max). (b) By applying temporal analysis: 35 18  Time-of-day variations,  Weekly usage patterns (day-of-week patterns) It is removed. (c) Dominant serving cells and cell overlap regions (Cell overlap zones) are detected. 5 For example: 𝑅𝑆𝑅𝑃 − 𝑅𝑆𝑅𝑃 < 3𝑑𝐵 If this occurs, the area in question is designated as an overlap zone. Indoor coverage analysis module (10), indoor coverage from aggregated measurement statistics 10 It creates indoor coverage heatmaps. The building floor area is divided into grid cells of size 1m × 1m (discretization). The coverage quality score for each grid cell is as follows: 𝑆𝑐𝑜𝑟𝑒 = 𝑤 × 𝑅𝑆𝑅𝑃 + 140 60 + 𝑤 × 𝑅𝑆𝑅𝑄 + 𝑤 × 𝑆𝐼𝑁𝑅 15 Weight values:  w1 = 0.5  w2 = 0.3  w3 = 0.2 It is determined in such a way as to be. 20 The score is normalized between 0 and 100. Coverage classification: Scoreboard Visualization <30 Poor coverage (Red) 30-60 Fair coverage (Yellow) 60-80 Good coverage (Green) >80 Excellent coverage coverage) Dark green A WebGL-based 3D rendering engine is used for heat map visualization. 19 The coverage gap detection algorithm is run. Adjacent sets of grid cells satisfying the following conditions:  Coverage quality score < 30,  Cluster area > 25 m², If present, it is defined as a coverage hole. 5 The following information is reported for each coverage gap:  Location information (centroid coordinates),  Area size (m²),  Severity level (average coverage score),  Number of affected users (number of unique UEs in the last 7 days), 10  Suggested action: that small cell deployment, antenna tilt adjustment, that power boost. Interference hotspots are identified. 15 SINR value: 𝑆𝐼𝑁𝑅 < 0𝑑𝐵 The affected regions are localized, and the dominant interfering cells are identified. is determined. 20 Small cell placement optimization unit (11), coverage gaps, high traffic density regions (user count heatmap) and areas with low data transfer rates (low Candidate small cell location points using throughput areas as input data. It creates. The genetic algorithm-based optimization process consists of the following stages: 25 (a) Establishing the initial population  N=100 randomly generated candidate placement scenarios are created. (b) Calculation of the fitness function 𝑓𝑖𝑡𝑛𝑒𝑠𝑠 = 𝛼 × 𝑐𝑜𝑣𝑒𝑟𝑎𝑔𝑒_𝑖𝑚𝑝𝑟𝑜𝑣𝑒𝑚𝑒𝑛𝑡 + 𝛽 × 𝑐𝑎𝑝𝑎𝑐𝑖𝑡𝑦_𝑔𝑎𝑖𝑛 − 𝛾 × 𝑖𝑛𝑡𝑒𝑟𝑓𝑒𝑟𝑒𝑛𝑐𝑒_𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒 − 𝛿 × 𝑐𝑜𝑠𝑡 Here:  coverage_improvement: coverage improvement rate,  capacity_gain: capacity increase,  interference_increase: increase in interference,  cost: installation cost 5 It is considered as such. Weighting coefficients are adjusted experimentally. (c) Genetic operations The following procedures are applied:  Selection, 10  Crossover,  Mutation. Optimization:  This is carried out over 200 generations. (d) Pareto front face extraction 15 Multi-objective optimization results in different performance-cost balances. Optimal solutions are determined. For each candidate small cell location, three-dimensional ray tracing (3D ray- Tracing-based propagation simulation is performed. The three-dimensional model of the building was created taking into account the walls, floor structures, and material properties. It is created. The material parameters used are, for example, as follows:  Relative dielectric constant for concrete: εr = 5  Relative dielectric constant for wood: εr = 2.5 Radio wave propagation calculations are performed in the following steps: (a) The transmitted signal parameters emitted from the small cell antenna are determined: 25  EIRP (Effective Isotropic Radiated Power),  Frequency information,  Antenna radiation pattern. (b) Ray paths are tracked taking into account the following physical effects:  Reflection, 30  Diffraction, 21  Scattering. (c) The received signal level is estimated for each grid point:  RSRP estimation is performed. (d) SINR is calculated taking into account interference from other cells. Estimated and measured RSRP values ​​for simulation accuracy verification: 5 The RMSE (Root Mean Square Error) is calculated between them. Typically: 𝑅𝑀𝑆𝐸 < 6𝑑𝐵 Verification is provided in this way. 10 Prioritized small cell placement points by the optimization unit, and An output list is generated that includes estimated performance metrics. Example output: Location_1: (x=35.2 m, y=28.5 m, z=3.5 m) 15 Estimated coverage improvement: 85% Capacity increase: 2.5 Gbps Interference increase: 2 dB Estimated installation cost: $8,000 Return on investment (ROI): 18 months Network planners review the generated proposals and small cell It makes installation decisions. The transfer parameter optimization engine (12) uses the user's movement trajectory history. Analyzes (user trajectory history). Areas where frequent handover occurs are identified using the following criteria: 25  Location sets with a handover rate > 2 handovers / minute. The following information is recorded in the database for each handover event:  Source cell,  Target cell,  Trigger reason: 30 22 that A3 incident, that A5 incident,  Timestamp,  Location information,  Success / failure status. 5 Location-specific handover parameter optimization is performed. Especially in areas of critical mobility:  Corridor intersections,  Elevator entrance areas, The parameters are adapted accordingly. 10 Example:  Hysteresis value: 3dB → 5dB It is increased as follows: 15  TTT (Time-To-Trigger) value: 40mS → 160mS It is set as follows. This procedure is implemented to reduce ping-pong handover incidents. 20 A special control logic is applied for vertical movement scenarios involving floor changes. Elevator or stair movement is determined according to the following condition: d dt > 0.5m / s If this occurs, vertical mobility is detected. 25 Predictive handover preparation is carried out. Example: User:  If moving from Floor 2 to Floor 3, 23 System:  Pre-determines the target cell on floor 3,  X2AP initiates the handover preparation with a Handover Request message. Thanks to this early preparation:  Handover delay is reduced by 40-50%. 5 Typical value: 150mS → 80mS A machine learning-based handover success prediction model is trained. Training data: 10  Past handover events. Features used:  Source cell RSRP value,  Target cell RSRP value,  RSRQ, 15  SINR,  User speed,  Handover type,  Location characteristics. Label: 20  Successful handover,  Failed handover. The Random Forest classifier model is used: Parameters:  Number of trees: 500 25  Maximum depth: 15 Model accuracy is approximately [level not specified]. If the estimated handover success probability is below 0.7:  Handover parameters are adjusted,  Or the handover process is postponed until more suitable radio conditions arise. 30 This reduces unnecessary handover attempts and increases network continuity. 24 Service quality policy implementation module (13), user semantic location information It evaluates semantic location and applies location-based QoS rules. The rules defined in the policy database include the following examples: Rule 1: For hospital operating room environments If user location = Hospital_Operating Room, then 5 QCI = 1 (GBR voice service), Priority = 2 (high), ARP is assigned as 1. Rule 2: For the stadium VIP area If the user's location is Stadium_VIP_Zone, then it's 10. A dedicated bearer is created. GBR is assigned as 10 Mbps. Rule 3: For general use areas If User location = Public_Area The default QoS policy applies. 15 QoS policy implementation, PCRF (Policy and Charging Rules Function) The pricing rules function is implemented via the interface. PCC (Policy and Charging Control) rules are implemented using Diameter Gx / Gxx messages. It is transferred to the PGW / PCEF components. In an emergency call scenario, when an E911 / E112 call is detected, 20 The user's sensitive location information is automatically sent to PSAP (Public Safety Answering Point) - It is forwarded to the Public Security Response Point unit. Location information is generated in the following format: Within the PIDF-LO (Presence Information Data Format - Location Object) XML structure:  Building name, 25  Floor information,  Room information,  Civic address information And  Latitude, 30  Longitude,  Altitude Geodetic coordinates containing this information are found. Location information is included in the SIP INVITE message using the Geolocation header. It is movable. Example information provided to the PSAP operator: "Emergency call: Westfield Mall, 2nd Floor, near Entrance C, Coordinates: 37.7749, - 122.4194, Accuracy: ±2.5 m" 5 Network configuration interface (14), optimized network parameters based on eNodeB / gNB It transfers them to their stations. Configuration processes are carried out via the NETCONF protocol (RFC 6241). is carried out. Configuration messages are sent encoded in XML format. 10 Updated parameters:  Handover A3 offset value,  Hysteresis value,  TTT (Time-To-Trigger) values,  Parameters related to neighboring cell relationships, 15  Acceptance control thresholds: Maximum number of RRC connections, Maximum PRB usage rate,  QoS policy identifiers. Configuration changes are implemented as atomic transactions. 20 In this context:  The process is complete if all changes are successful.  If an error occurs, the transaction will be completely reversed. The rollback mechanism detects deterioration in KPI performance. In this case, it automatically reverts to the previous configuration. 25 The performance analytics dashboard (15) displays real-time and historical KPI data. It visualizes and analyzes. Key performance indicators monitored: Positioning performance  Positioning success rate: 30 26 𝑆𝑢𝑐𝑐𝑒𝑠𝑠𝑓𝑢𝑙 𝑃𝑜𝑠𝑖𝑡𝑖𝑜𝑛𝑠 / 𝑇𝑜𝑡𝑎𝑙 𝑃𝑜𝑠𝑖𝑡𝑖𝑜𝑛 𝑅𝑒𝑞𝑢𝑒𝑠𝑡𝑠  Positioning delay: o p50, o p95, 5 those p99 percentile values. Typical value:  p50 = 180 ms  p95 = 450 ms Network performance metrics 10 Coverage improvement:  Before / after comparison:  15% increase in area coverage. Handover success rate: 95.2% → 98.7% Ping-pong handover rate: 12% → 3% User data transfer speed: 20  Median value: 18𝑀𝑏𝑝𝑠 → 25𝑀𝑏𝑝𝑠 Small cell traffic transfer rate: 32% → 48% The machine learning-based anomaly detection module continuously monitors system operations. It monitors. The Isolation Forest algorithm is used to create a normal system behavior profile. Parameter:  Contamination value: Set to 0.01. 30 Examples of anomalies detected: 27 1. Deterioration in positional accuracy. Possible cause:  Image database corruption,  Reference image error. 2. Increased risk of sudden handover failure. Possible cause:  Network configuration error,  Incorrect setting of handover parameters. 3. Unexpected coverage gap formation. Possible cause:  eNodeB failure,  Antenna or power adjustment problem. The system generates an alarm after detecting an anomaly and performs root cause analysis (root cause 15 It generates recommendations for (analysis). Example output: "Anomaly detected: Reduced coverage in Zone_A." Possible cause: The power of the eNodeB_123 transmitter may have been reduced. Suggested action: Check the eNodeB_123 configuration." 20 How the System Works The system can determine the precise positions of users in indoor environments and track these positions. multi-layered design to use its knowledge in optimizing mobile communication networks. It adopts a working approach. The working principle is; computer vision, multi-sensor. fusion, integrated use of machine learning and network optimization techniques 25 It is based on its use. Positioning Phase In the first stage, the image capture interface (1) is connected to the camera of the user equipment or It receives image streams from cameras located inside the building. Noise reduction and contrast enhancement of images by image preprocessing module (2) 30 and undergoes perspective correction procedures. 28 Then the visual feature extraction engine (3) extracts SIFT (Scale- from the processed images. Invariant Feature Transform), ORB (Oriented FAST and Rotated BRIEF) Deep learning-based Convolutional Neural Network (CNN) combined with computer vision algorithms It extracts distinctive visual features using its models. The extracted features are already in the visual database and mapping system (4) 5 The generated features are matched with those of reference images. Then, position estimation is performed. engine (5), geometric verification and Perspective-n-Point (PnP) using algorithms to determine the user's three-dimensional position coordinates and orientation accounts. Sensor fusion module (6), camera-based position estimation; Inertial Measurement Unit 10 (IMU), WiFi, Bluetooth Low Energy (BLE) markers and barometric pressure sensor Data obtained from auxiliary sensors such as these are used to perform an Extended Kalman Filter. Kalman Filter (EKF) performs optimum condition estimation by combining components within its structure. Then the real-time tracking module (7), particle filter and visual Using odometry (visual odometry) techniques, the user's movement trajectory is continuously tracked. It follows as follows. Location–Network Correlation Phase The determined location information is semantically matched with the indoor map repository (8). The building, floor, room, or other indoor area where the user is located is determined. Then the network measurement correlation engine (9) took RRC measurements from user equipment 20 radio performance metrics such as RSRP, RSRQ, and SINR included in the reports It performs correlation analysis by associating the determined location information. Time-series datasets consist of location, measurement values, and timestamps. It is stored in the database and statistically analyzed using continuous queries. is carried out. 25 Network Optimization Phase The indoor coverage analysis module (10) uses the collected location and network measurement data. It creates grid-based indoor coverage heat maps. It also provides coverage. 29 gaps (coverage holes), areas of poor coverage, and points of intense interference It identifies (interference hotspots). Small cell placement optimization unit (11), multi-objective genetic algorithms and three using three-dimensional ray tracing (3D) based radio propagation models for optimal performance It determines the location points of small cell settlements. 5 In addition, the transfer parameter optimization engine (12), position-dependent handover by identifying the regions and determining the optimum hysteresis and triggering time for each region. It calculates Time-to-Trigger (TTT) parameters. It also uses machine learning methods. Handover success prediction is performed using this method, and proactive measures are taken accordingly. Handover decisions are made. 10 Service quality policy implementation module (13) is the semantic context in which the user is located. depending on the location (e.g., operating room, stadium VIP area, or emergency zone) It implements adaptive Quality of Service (QoS) policies. In emergency calls, the user's sensitive location information is automatically shared with the Public. It is forwarded to the Security Response Points (PSAP). 15 Finally, network parameters optimized via the network configuration interface (14), Using NETCONF and SNMP protocols, to eNodeB and gNodeB base stations It is applied. Continuous Improvement Cycle 20 Performance analytics dashboard (15) displays system performance key performance It continuously monitors and reports through indicators (KPIs). During the operation, anomaly detection algorithms based on machine learning were used. Potential performance problems are identified. Thanks to the feedback mechanism; 25  System parameters are continuously fine-tuned,  Location determination models are retrained,  Network optimization strategies are updated and improved. In this way, the system becomes a learning structure that improves its performance over time. They meet. 30 Key Advantages The proposed system includes GPS, WiFi fingerprinting, and network-based technology. By eliminating the limitations of positioning methods in enclosed spaces, the meter It achieves precise indoor positioning at this level. Using the precise location information obtained; 5  coverage optimization,  small cell layout planning,  handover optimization,  Quality of Service (QoS) management,  Emergency location services 10 It is carried out in real time and proactively. In addition, the system is self-learning and self-optimizing. It has a (self-optimizing) structure and reduces operating costs (Operational Expenditure - (OPEX) is reduced, mobile communication network performance is improved, and end user It significantly improves the experience. 15 This invention enables modern telecommunications operators to manage and orchestrate networks. One of the critical components of Network Management and Orchestration (NMO) infrastructure. It is positioned as such, and specifically the Indoor Network Optimization Package (Indoor It plays a central role within the Network Optimization Suite. Integration within the Network Management System Hierarchy 20 Element Management Layer (EMS) The invention is within the Element Management Layer (EMS);  Small Cell Gateway  Distributed Antenna System Controller (DAS) Controller), 25  Wi-Fi Controller It communicates directly with indoor communication infrastructure components such as these. System;  Measurement data from eNodeB base stations via S1 and X2 interfaces receiving, 30  Network configuration via TR-069 and NETCONF management protocols It is implementing its changes. 31 Domain Management Layer (DMS) At the Domain Management Layer level, the invention is integrated with the following systems: He is working. Self-Organizing Network (SON) Domain Manager It works in coordination with the following SON functions: 5  Mobility Load Balancing (MLB)  Mobility Robustness Optimization MRO)  Coverage and Capacity Optimization CCO) 10 Radio Access Network Optimization Manager (Manager) In line with general optimization strategies for radio access network parameters It enables it to function. • Location-Based Services Platform (LBS) 15 Indoor positioning services via application programming interfaces (APIs) It offers third-party applications. • Network Planning and Design Tools Two-way data exchange is carried out using the following systems.  Coverage planning simulation software, 20  Small cell layout planning systems. Network Management Layer (NMS) Meeting at the highest management level;  Network General Analytics Platform,  Big Data Repository, 25  Manager Dashboards It provides data for this purpose. In addition, the following key performance indicators (KPIs) at the operator level It nourishes.  The nationwide indoor coverage rate is 30  Small cell efficiency,  Response time of emergency services. 32 Data Flow Architecture and Integration Points Southbound Interfaces • S1-MME Interface Through this interface;  User equipment context information, 5  RRC measurement reports,  Handover signaling,  Emergency call notifications is being received. • X2 Interface 10 This interface;  Handover coordination between eNodeBs,  Sharing cargo information It is used for this purpose. • TR-069 / NETCONF 15 Through these protocols;  Small cell structures,  Software updates,  Alarm notifications It is managed. 20 • Gx / Gxx Diameter Interface In order to implement Dynamic Quality of Service (QoS) policies, the Policy and It communicates with the Pricing Rules Function (PCRF). • User Equipment Application Programming Interface (UE Application API) Subject to user consent; 25  Location queries from smartphone applications,  Uploading camera footage to the system is being carried out. Northbound Interfaces • RESTful API 30 33 Indoor communication with third-party applications using JSON data format over HTTPS. Location services are provided. • NGSI-LD (Next Generation Service Interface) Sharing contextual information with smart building and Internet of Things (IoT) platforms. is being carried out. 5 • MQTT Message Broker Publish / subscribe for real-time location updates. Its architecture is used. • GraphQL Interface Flexible querying of coverage analyses and performance data is possible. It provides opportunities. • SNMP and Syslog To network management systems;  alarm,  event, 15  performance Notifications are being sent. Horizontal Integration Interfaces (East-West Interfaces) • FINAL Coordination Protocol Among the SON systems from different manufacturers; 20  policy reconciliation,  conflict resolution is being carried out. • Location Information Server (LIS) Emergency services were connected using the HELD (HTTP Enabled Location Delivery) protocol on 25 days. Integration is provided. • Radio Environment Map (REM) Database The following information is being shared.  Spectrum detection data,  Entrepreneurship maps. 30 34 • Geographic Information System (GIS) Integration The following data is updated simultaneously.  Building database,  Floor plans,  Geographic information data. 5 Cloud-Native Deployment Architecture The system operates using a microservice-based cloud-native architecture. Thus High scalability, fault tolerance, and uninterrupted software deployment are ensured. Microservices Image Processing Service 10  Container-based computer vision applications,  GPU acceleration (CUDA and TensorRT),  Horizontal scaling based on image processing intensity. Location Calculation Service  Stateless computing architecture, 15  Automatic scaling based on location demand density,  Developed using the C++ language for low latency. Database Service  PostgreSQL cluster: those floor plans, 20 that reference image metadata.  InfluxDB: that time series measurement data.  Redis: Caching frequently accessed information. 25 Machine Learning Training Service  Distributed training based on TensorFlow or PyTorch,  Daily or weekly model retraining,  Model version management,  A / B testing infrastructure. 30 Network Integration Service It includes customization modules for network equipment from different manufacturers. For example;  Ericsson, 35 Nokia,  Huawei. Moreover;  S1-AP,  X2-AP, 5  NETCONF It performs conversion operations between protocols and configures backup / restore. It performs loading functions. Analytical Service  Large-scale data processing with Apache Spark, 10  Real-time data stream processing with Apache Kafka Streams,  Dashboard infrastructure based on Node.js and Express. Orchestration and Direction Within the scope of system administration;  Container orchestration with Kubernetes, 15  Service discovery with Consul or etcd,  Load balancing with Nginx Ingress,  System monitoring with Prometheus and Grafana,  Managing logs with Elasticsearch, Logstash, and Kibana (ELK),  Distributed transaction monitoring with Jaeger 20 is being carried out. Security Architecture Identity Verification and Authorization System;  OAuth 2.0, 25  OpenID Connect It performs user authentication using its protocols. API access;  Role-Based Access Control (RBAC) It is managed by 30. Communication between microservices is as follows:  mutual TLS (mTLS) 36 This is guaranteed. Data Privacy The collected image data; • Face blurring,  Plate Masking 5 It is anonymized through various processes. System;  User consent management under GDPR,  data retention policies,  The right to be forgotten 10 It supports the requirements. Location data;  During storage, the AES-256 algorithm,  During transmission, TLS 1.3 protocol is used. It is encrypted using 15. Network Security In the system;  DMZ architecture for services accessible from the outside,  Allow-list based firewall rules, 20  Intrusion Detection System (IDS),  regular security audits,  penetration tests Network security is ensured by implementing these measures. 25 Different Application Areas of the Invention 1. Location-Aware Slice Selection for 5G Network Slicing Within the scope of the invention, the user's indoor location is taken into account in 5G communication networks. The most suitable network slice is selected accordingly. For example;  factory production line, 30  hospital operating room, 37  stadium spectator area depending on different areas of use, such as;  Enhanced Mobile Broadband (eMBB),  Ultra Reliable Low Latency Communication (URLLC),  Multiple Machine-Type Communication (mMTC) 5 The most suitable network slot is automatically selected. In this context, Network Slice Selection Assistance Information Information (NSSAI) parameters are assigned dynamically. For users in critical mission areas, a very low URLLC network slice is selected. Low latency and high communication reliability are ensured. 10 2. Spatial Design for Augmented Reality (AR) and Mixed Reality (MR) Applications Fixing The invention combines augmented reality (AR) and mixed reality (MI). It provides precise spatial anchoring for MRI applications. Thanks to the visual positioning system, virtual objects are 15 times closer to physical world coordinates. It can become permanently attached. In this context;  Augmented reality-supported indoor navigation,  Shared design visualization in meeting rooms,  Augmented reality applications in museum exhibitions 20 It can be accomplished. 3. Smart Building and Internet of Things (IoT) Device Management The system tracks the precise locations of Internet of Things (IoT) devices within the building. It enables the determination of this. Examples of positionable devices include: 25  air conditioning systems (HVAC),  lighting systems,  access control systems It can be given. 38 Moreover;  Indoor signage for maintenance personnel,  equipment tracking,  Automatic location of faulty sensors The functions are being performed. 5 Example output: "A malfunction has been detected in the temperature sensor located in Room 412, 4th Floor, Building B." The system also takes user density into account;  heating,  cooling, 10  energy management It performs optimization. 4. Indoor Navigation for Robotic Systems and Autonomous Vehicles Invention;  Automated Guided Vehicles (AGVs), 15  Autonomous Mobile Robots (Autonomous Mobile Robot - AMR) It provides precise indoor navigation. Using a Visual SLAM (Visual Real-Time Positioning and Mapping) system;  dynamic obstacle detection,  optimal route planning, 20 • Multiple robot coordination is being carried out. This system;  warehouse automation,  hospital logistics robots, 25  Airport baggage handling systems,  Retail inventory management robots It can be used in applications such as these. 5. Emergency and Security Procedures In case of emergencies such as fire, earthquake, etc., the system; 30 39 Using users' real-time location information, the most suitable evacuation routes It calculates and sends redirection notifications to mobile devices. Moreover; 1. The location of security personnel inside the building in real time. can be monitored, 5 2. When the panic button is activated, location information can be automatically transmitted. In addition;  missing child,  elderly individual,  10 people in need of care Search activities are being conducted using the last known location information during the search. is being accelerated. 6. Retail Analytics and Customer Behavior Analysis By analyzing customer movements in shopping malls and department stores;  The most frequently used corridors, 15  most visited stores,  Waiting times in product sections,  customer density maps is being created. Depending on the user's location; 20  personalized campaigns,  discount notifications They can be sent to mobile devices. Waiting times at the checkout counters are analyzed to optimize staff allocation. 7. Healthcare Facilities and In-Hospital Positioning 25 The system is used in healthcare facilities;  patient monitoring,  personnel tracking,  medical equipment management It can be used for this purpose. 30 For example; 40 If Alzheimer's or dementia patients go outside the permitted areas An automatic alarm is generated. High-cost medical devices;  infusion pumps,  patient monitors, 5  wheelchairs Their real-time locations are being tracked. Moreover;  Response times to nurse calls,  Equipment transportation processes 10 It is being optimized. Contact tracing can be carried out during epidemic situations. 8. Special 5G and Campus Network Optimization Invention;  corporate campus networks, 15  Industrial IoT applications,  smart factories It is used in the planning and optimization of dedicated 5G / LTE networks. For critical machinery in production areas;  unique small cell arrangement, 20  Quality of Service (QoS) is provided. In addition, seamless mobility management between open and closed areas. is being carried out. 9. Satellite-Terrestrial Hybrid Positioning 25 At building entrances, window edges, and similar transition areas;  Global Positioning System (GPS)  visual location determination They are used together. 41 Where a GPS signal is available, GPS data provides visual positioning. It is used as a reference for the calibration of the system. In environments where there is no GPS signal, the system relies entirely on visual location determination. It is switching to this method. 10. Indoor Navigation and Warehouse Inspection with Unmanned Aerial Vehicles 5 The system uses unmanned aerial vehicles that operate in warehouses and enclosed spaces where GPS signals are unavailable. It supports autonomous movement of vehicles. Using the visual SLAM method;  avoiding obstacles,  precise flight control, 10  Reading barcodes and RFID tags,  inventory audit is being carried out. In addition, unmanned aerial vehicles are used in building inspections;  structural damage assessment, 15  thermal imaging,  precise three-dimensional positioning They can be implemented together. Thanks to these application areas, the invention is not only relevant to mobile communication networks. not in optimization; 5G communication systems, smart cities, smart buildings, 20 industrial automation, healthcare technologies, robotic systems, augmented reality, security applications and numerous different technological fields such as the Internet of Things (IoT) It offers a comprehensive solution that can be used.

Claims

42 REQUESTS 1. It is an image processing-assisted indoor positioning system, the feature of which is;  Capturing video streams from user equipment, images Image capture interface (1) that analyzes and performs frame extraction.  Noise reduction on the images it receives from the image capture interface (1), 5 contrast enhancement, perspective correction, blur removal, lighting image pre-processing that applies normalization and image enhancement algorithms processing module (2), • Extracts distinctive visual features from images, adjusts scale and rotation. Binary identifiers that obtain key points unaffected by changes 10 and a visual feature extraction engine that produces robust visual features (3),  storing a pre-mapped visual database of each of their interior spaces Images obtained from different perspectives for the reference position and a visual database that indexes and stores the attribute vectors belonging to these, and mapping system (4), 15  Attributes extracted from the query view and references found in the database By matching attributes, the user's coordinates, floor level, and orientation can be determined. Position estimation engine that determines the angle (5),  Sensor fusion module combining multiple sensor data (6),  Real-time tracking that continuously monitors the user's current location 20 module (7),  storing floor plans, three-dimensional models, and semantic information of buildings and interior spaces that align the floor plans with the building coordinate system by separating them. map repository (8),  Measurement reports obtained from the user's equipment based on the determined user location 25 performing correlation analysis between the signal values ​​within each Calculates signal level values ​​for location points, measurement points Network measurement correlation that estimates signal quality in the regions between them. engine (9),  Using the collected location and measurement data, indoor coverage maps 30 indoor coverage analysis module (10),  Small cell placement that determines optimum small cell placement points optimization unit (11), 43  Location using the user's location, direction of movement, and speed information. transfer that performs conscious handover parameter optimization parameter optimization engine (12),  based on semantic information related to the indoor environment in which the user is located Service quality policy 5 implementing adaptive service quality policies application module (13),  A network that applies optimized network parameters to base stations configuration module (14),  Performance analytics that monitor, analyze, and report system performance. It includes the instrument panel (15). 10 2. It is a position determination system according to claim 1, and its feature is an H.264 / H.265 video encoder. It includes an image capture interface (1) that decodes images in various formats.

3. It is a location determination system according to claim 1, and its features are; RTSP, HTTP and WebRTC. Image capture interface that receives image data via protocols (1) 15 It includes.

4. It is a position determination system according to claim 1, and its characteristic is that it uses the SIFT algorithm. A visual that obtains key points unaffected by scale and rotation changes. It includes a feature extraction engine (3).

5. It is a positioning system according to claim 1, and its feature is; fast binary 20 with ORB algorithm. It includes a visual feature extraction engine (3) that extracts identifiers.

6. It is a positioning system according to claim 1, and its feature is; robustness with SURF algorithm. It includes a visual feature extraction engine (3) that extracts visual features.

7. It is a positioning system according to Claim 1, and its characteristic is; a pre-trained Convolutional system. Visual feature 25 that extracts higher-level semantic features using neural network models It includes an extraction motor (3).

8. It is a location determination system according to claim 1, and its feature is similar to a query view. To find reference images, a KD-tree, hierarchical k-means, and A visual database and mapping system using locally sensitive hybrid data structures. (4) includes. 30 44 9. It is a position determination system according to claim 1, and its characteristic is the RANSAC algorithm. The position performs geometric validation by filtering outliers using this method. It includes a prediction engine (5).

10. It is a positioning system according to Claim 1, and its characteristic is; perspective point algorithms. Using three-dimensional–two-dimensional point mappings, the camera's exposure is adjusted to 5. It includes a position estimation engine (5) that calculates.

11. It is a position determination system according to Claim 1, and its features are triangulation and bundle. a positioning system that increases the accuracy of position determination by applying compensation processes. It includes a prediction engine (5).

12. It is a position determination system according to Claim 1, and its feature is; visual data obtained from the camera. 10 Position estimation is provided by the accelerometer, gyroscope, and Inertial Measurement Unit. Magnetometer data, fingerprint information, and barometric pressure sensor data are used. combining floor level detection data and distance measurement data It includes a sensor fusion module (6).

13. It is a position determination system according to claim 1, and its feature is; using kalman filters 15 It includes a sensor fusion module (6) that performs state estimation.

14. It is a position determination system according to Claim 1, and its characteristic is; reliability for each sensor. Sensor fusion module that applies adaptive weighting by calculating scores. (6) is included.

15. It is a position determination system according to claim 1, and its characteristic is that it is nonlinear and Gaussian. Particle filter for reliable monitoring in environments that do not conform to distribution standards. It includes a real-time tracking module (7) that implements the method.

16. It is a positioning system according to Claim 1, and its feature is that it is visual odometry and IMU based. Motion estimation between successive image frames using dead recency techniques It includes a real-time tracking module (7) which performs the following. 25 17. It is a position determination system according to Claim 1, and its feature is loop closure detection. By applying this method, it corrects positional deviations that occur during long-term follow-up. It includes a time tracking module (7).

18. It is a positioning system according to Claim 1, and its characteristic is that it uses Geographic Information System data. It includes an integrated indoor map repository (8). 30 45 19. It is a position determination system according to Claim 1, and its feature is that it receives data from the user's equipment. Correlation between RSRP, RSRQ, RSSI, and SINR in RRC measurement reports. It includes a network measurement correlation engine (9) that performs the analysis.

20. It is a position determination system according to Claim 1, and its characteristic is time series analysis. By applying this, we can analyze time-dependent changes in indoor coverage performance. 5 It includes a network measurement correlation engine (9).

21. It is a position determination system according to Claim 1, and its feature is guide-based representation. Within this approach, using grid cells, coverage quality is determined for each cell. an interior system that calculates the score and classifies the regions according to the calculated values. It includes the coverage analysis module (10). 10 22. It is a positioning system according to Claim 1, and its feature is; coverage gaps, high high traffic areas, areas with low SINR levels, and user density. Small cell performing multi-objective optimization taking distribution into account It includes a settlement optimization unit (11).

23. It is a positioning system according to Claim 1, and its features include; genetic algorithm, annealing 15 small cell simulation or particle swarm optimization methods It includes a settlement optimization unit (11).

24. It is a position determination system according to Claim 1, and its feature is; each handover boundary position. The transfer calculates the optimum hysteresis, trigger time, and A3 offset values. It includes a parameter optimization engine (12). 20 25. It is a positioning system according to Claim 1, and its characteristic is that it is located in critical mobility zones. Transfer parameters that define proactive handover triggering mechanisms It includes an optimization engine (12).

26. It is a location determination system according to Claim 1, and its feature is; Machine learning methods. Transfer parameter optimization 25 which estimates the probability of handover success. It includes the motor (12).

27. It is a positioning system according to Claim 1, and its characteristic is position-based QCI. Quality of service policy that performs matching, GBR allocation and ARP adjustment It includes application module (13). 46 28. It is a location determination system according to Claim 1, and its features include policy and pricing. Service quality policy implementation that integrates with the rules function. It includes module (13).

29. It is a location determination system according to Claim 1, and its feature is configuration management. Network 5 performs its operations via TR-069, NETCONF and SNMP protocols. It includes the configuration module (14).

30. It is a position determination system according to Claim 1, and its features include handover parameters and QoS. Network configuration module that updates policies and acceptance control threshold values. (14) is included.

31. It is a position determination system according to claim 1, and its feature is; SON functions and 10 Coordination and load balancing, mobility enhancement and optimization, Network supporting RACH optimization and energy saving functions. It includes the configuration module (14).

32. It is a positioning system according to Claim 1, and its characteristic is positioning accuracy. positioning availability, coverage improvement rate, handover success rate, 15 Ping-pong ratio user data transfer speed improvement and small cell traffic transfer Performance analytics dashboard (15) which evaluates the metrics related to the ratio It includes.

33. It is a positioning system according to Claim 1, and its feature is; interactive three-dimensional. Using visualization, indoor coverage maps, user distribution heat maps 20 and performance analytics dashboard visualizing handover flowcharts (15) It includes.

34. It is a location determination system according to Claim 1, and its characteristic feature is machine learning methods. using anomaly detection, predictive maintenance and capacity estimation It includes a performance analytics dashboard (15). 25

Images