A multi-dimensional security monitoring system for parking lot using infrared sensing technology

Abstract

The application relates to the technical field of security monitoring, and discloses a parking lot multidimensional security monitoring system using infrared sensing technology; the system comprises a temperature drift compensation module, a hot target area screening module, a target type identification module and a security decision mechanism establishment module; first, parking lot complementary sensor data is collected, and baseline temperature drift compensation is carried out based on environmental adaptability; second, a self-adaptive threshold segmentation algorithm and a target tracking algorithm are used to screen out a hot target area, and multidimensional characteristics of the hot target area are obtained; third, a target classification model is established based on a double-channel hybrid neural network, behavior mode recognition is carried out after the target classification model is trained and a classification label is input, and a parking lot target type is output; finally, the behavior characteristics of the parking lot target type are recognized, a risk level is set, and a three-level response security decision mechanism is established; the application realizes the purpose of multidimensional security monitoring of a parking lot by processing and modeling multidimensional data of the parking lot, and the method is objective and accurate.

Description

Technical Field

[0001] This invention relates to the technical field of security monitoring, specifically to a multi-dimensional security monitoring system for parking lots that utilizes infrared sensing technology. Background Technology

[0002] With the acceleration of urbanization and the surge in car ownership, parking lots are developing towards larger scale, underground locations, and multi-level structures. Their increasingly complex structures, relatively enclosed spaces, high pedestrian and vehicle traffic, and poorly managed environments not only make them high-risk areas for vehicle theft, scratches, and vandalism, but also pose fire hazards and security risks such as pedestrian congestion. Simultaneously, the widespread adoption of new energy vehicles has led to a surge in demand for charging safety monitoring. In recent years, cases of vehicle theft and damage to parking facilities have increased year by year, urgently requiring more intelligent, precise, and reliable 24 / 7 security monitoring methods to improve management efficiency and security levels.

[0003] Traditional parking lot security monitoring methods mainly use closed-circuit television (CCTV) surveillance, deploying cameras at entrances, exits, passages, and corners to monitor in real time or record and play back videos. Some systems integrate license plate recognition and behavior analysis functions, while parking space detectors are installed on the ground of parking spaces to determine the occupancy status. Security personnel conduct regular patrols, and manual alarm devices are set up in key locations.

[0004] Traditional parking lot security monitoring methods have serious limitations in video surveillance, including blind spots caused by light, pillar obstruction, rain and fog interference, and lens damage. They also have poor environmental adaptability, making it difficult to identify potential threats. Furthermore, they mainly rely on single visual dimension detection and lack the ability to deeply and intelligently recognize the presence of pedestrians and abnormal temperatures, resulting in a low level of intelligence. Summary of the Invention

[0005] To address the problems in related technologies, this invention provides a multi-dimensional security monitoring system for parking lots that utilizes infrared sensing technology, thereby overcoming the aforementioned technical problems in existing related technologies.

[0006] To solve the aforementioned technical problem, the present invention is achieved through the following technical solution:

[0007] This invention is a multi-dimensional security monitoring system for parking lots that utilizes infrared sensing technology, specifically including: a temperature drift compensation module, a thermal target area screening module, a target type identification module, and a security decision-making mechanism establishment module;

[0008] The temperature drift compensation module is used to collect complementary sensor data of the parking lot, perform baseline temperature drift compensation based on environmental adaptability, and obtain a processed multi-dimensional data set of the parking lot.

[0009] The hot target area filtering module is used to divide the parking lot into static and dynamic areas based on the processed multi-dimensional parking lot data set, and use an adaptive threshold segmentation algorithm and a target tracking algorithm to filter out the hot target areas and obtain the multi-dimensional features of the hot target areas.

[0010] The target type recognition module is used to set classification labels corresponding to multi-dimensional features of hot target areas, establish a target classification model, and after training, input the classification labels to perform behavior pattern recognition and output the target type of the parking lot.

[0011] The security decision-making mechanism module is used to identify the behavioral characteristics of parking lot target types, set risk levels, and establish a three-level response security decision-making mechanism.

[0012] Preferably, the collection of complementary sensor data in the parking lot includes:

[0013] Infrared sensors are installed in the security monitoring area to form an infrared sensor grid covering the parking lot, and infrared thermal images are collected. Complementary sensors are also deployed in the parking lot to collect data from the complementary sensors simultaneously, forming an initial multi-dimensional data set of the parking lot.

[0014] Preferably, the baseline temperature drift compensation based on environmental adaptability includes:

[0015] The infrared heat map in the initial parking lot multi-dimensional data set is replaced with an infrared heat map matrix;

[0016] Establish a temperature calibration table, obtain the reference temperature value, and calculate the temperature correction value of the pixel in the infrared thermal image matrix;

[0017] Once a sudden temperature change scenario is detected, the temperature correction values ​​of pixels in the infrared thermal image matrix are compensated, the final temperature values ​​of pixels in the infrared thermal image matrix are calculated, a compensated infrared thermal image is generated, and a processed multi-dimensional data set of the parking lot is obtained.

[0018] Preferably, dividing the parking lot into static and dynamic areas based on the processed multi-dimensional parking lot data set includes:

[0019] The compensated infrared thermal image is meshed to generate an infrared thermal image grid, which is then converted into a final infrared thermal image matrix. The temperature variance of each pixel is calculated, and the infrared thermal image grid is marked as dynamic and static areas to obtain a partitioned infrared thermal image.

[0020] Preferably, the step of using an adaptive threshold segmentation algorithm and a target tracking algorithm to filter out hot target regions includes:

[0021] The dynamic area in the partitioned infrared thermal image is recorded as the infrared thermal image to be detected, and converted into the infrared thermal image matrix to be detected. Hot pixels are marked to form a thermally connected region. Then, ghost image filtering is performed to obtain the final thermally connected region.

[0022] The millimeter-wave radar signals in the processed multi-dimensional parking lot data set are converted into millimeter-wave radar point cloud data.

[0023] Clustering is performed on the millimeter-wave radar point cloud data to generate moving target clusters;

[0024] Determine the center position of the moving target cluster and the center position of the final thermally connected region, and set a distance threshold. When the distance between the center position of the moving target cluster and the center position of the final thermally connected region is less than the distance threshold, the corresponding thermally connected region forms a thermal target region.

[0025] Preferably, the obtained multi-dimensional features of the thermal target region include:

[0026] Infrared morphological features, temperature distribution features, and radar motion features of the thermal target area are extracted to obtain multi-dimensional features of the thermal target area.

[0027] Preferably, establishing the target classification model includes:

[0028] Based on the multi-dimensional features of the thermal target area, classification labels are obtained, and infrared thermal images and motion features to be identified are generated.

[0029] The target classification model is defined as having a thermal imaging feature channel and a spatiotemporal motion channel. The thermal imaging feature channel includes an input layer, a convolutional neural network, a max pooling layer, and an output layer. The spatiotemporal motion channel includes an input layer, a two-layer long short-term memory neural network, and an output layer. The output results of the thermal imaging feature channel and the spatiotemporal motion channel are combined and upscaled through dual-channel feature concatenation, and then passed through a fully connected layer to obtain the final output result, thus establishing the target classification model.

[0030] Preferably, the step of inputting classification labels after training for behavior pattern recognition includes:

[0031] Infrared thermal image samples and motion feature samples are acquired and divided into a sample training set and a sample test set. These are then input into the target classification model for training to obtain the final target classification model.

[0032] The infrared thermal image and motion features to be identified are input into the final target classification model for behavior pattern recognition, and the parking lot target type is output.

[0033] Preferably, the establishment of the three-level response security decision-making mechanism includes:

[0034] Based on the parking lot target type, identify suspected theft behavior characteristics, illegal intrusion behavior characteristics, personnel falling to the ground behavior characteristics, and vehicle fire behavior characteristics to obtain a behavior feature set;

[0035] Risk levels are set based on the behavioral feature set, and a three-level response security decision-making mechanism is established.

[0036] The present invention has the following beneficial effects:

[0037] 1. This invention deploys an infrared sensor array grid to collect complementary sensor data from parking lots, and then performs baseline temperature drift compensation based on environmental adaptability. It uses a two-level temperature compensation mechanism (basic calibration + abrupt change compensation) to reduce thermal imaging temperature drift errors in rainstorm / scorching sun scenarios. It also uses multi-source data collaboration to overcome monitoring failures caused by environmental interference, providing highly reliable security protection for the future.

[0038] 2. This invention divides the parking lot into static and dynamic zones, processes only dynamic zone data, eliminates static thermal interference through thermal imaging temperature variance partitioning, reduces computational load, uses adaptive threshold segmentation and target tracking algorithms to filter out thermal target areas, and employs infrared thermal image compensation + millimeter-wave radar fusion to solve the false alarm problem of traditional solutions in rain, fog, obstruction and thermal interference scenarios, breaks through the physical limitations of single video surveillance, and achieves accurate target perception in complex environments.

[0039] 3. This invention sets classification labels corresponding to multi-dimensional features of thermal target areas, establishes a target classification model based on a dual-channel hybrid neural network, and significantly improves classification accuracy through a dual-channel feature fusion mechanism (infrared morphology + motion sequence). It effectively solves the problem of misjudgment caused by changes in light and heat source interference in traditional methods. It has strong deep intelligent recognition capabilities for the presence of pedestrians and abnormal temperatures, and reduces the cost of manual security.

[0040] 4. This invention identifies the behavioral characteristics of parking lot target types and sets risk levels, establishing a three-level response security decision-making mechanism. It breaks through the limitations of traditional passive security response, intelligently matches the three risk levels (high / medium / low), realizes on-demand allocation of response resources, avoids resource waste, and achieves automated collaborative control of security monitoring, fire protection facilities, access control systems, and lighting equipment.

[0041] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0043] Figure 1 This invention provides a flowchart of a multi-dimensional security monitoring system for parking lots utilizing infrared sensing technology;

[0044] Figure 2 This invention provides a flowchart illustrating a multi-dimensional security monitoring method for parking lots utilizing infrared sensing technology. Detailed Implementation

[0045] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] Traditional parking lot security monitoring methods have serious limitations in video surveillance, including blind spots caused by light, pillar obstruction, rain and fog interference, and lens damage. They also have poor environmental adaptability, making it difficult to identify potential threats. Furthermore, they mainly rely on single visual dimension detection and lack the ability to deeply and intelligently recognize the presence of pedestrians and abnormal temperatures, resulting in a low level of intelligence.

[0047] To solve the above technical problems, such as Figure 1 As shown in the figure, this embodiment of the invention provides a multi-dimensional security monitoring system for parking lots using infrared sensing technology, specifically including: a temperature drift compensation module, a hot target area screening module, a target type identification module, and a security decision-making mechanism establishment module; the temperature drift compensation module is used to collect complementary sensor data of the parking lot, perform baseline temperature drift compensation based on environmental adaptability, and obtain a processed multi-dimensional data set of the parking lot; the hot target area screening module is used to divide the parking lot into static and dynamic areas according to the processed multi-dimensional data set of the parking lot, and use an adaptive threshold segmentation algorithm and a target tracking algorithm to screen out hot target areas, and obtain multi-dimensional features of hot target areas; the target type identification module is used to set classification labels corresponding to the multi-dimensional features of hot target areas, establish a target classification model, and after training, input the classification labels to perform behavior pattern recognition and output the parking lot target type; the security decision-making mechanism establishment module is used to identify the behavioral characteristics of parking lot target types, set risk levels, and establish a three-level response security decision-making mechanism.

[0048] In a specific embodiment, 80 infrared sensors (grid density: 1 node per 25㎡) are installed on the top of an underground parking lot. Complementary sensors are added to cover key locations in the lanes and parking areas: 4 millimeter-wave radars are deployed at the entrance / corner (to detect the speed / trajectory of moving targets), 6 temperature sensors are installed on the pillars (to collect ambient temperature in real time), and 3 light sensors are installed above the lanes (to determine day / night and lighting conditions). All sensors achieve synchronous data collection with a 1-second cycle through a LoRa gateway, providing a good data foundation for the embodiment of the present invention.

[0049] In the specific implementation of the above embodiments, firstly, by deploying an infrared sensor array grid, complementary sensor data of the parking lot is collected. Then, baseline temperature drift compensation is performed based on environmental adaptability to obtain a processed multi-dimensional data set of the parking lot. This method employs a two-level temperature compensation mechanism (basic calibration + abrupt change compensation) to reduce thermal imaging temperature drift errors in rainstorm / scorching sun scenarios. It also uses multi-source data collaboration to overcome monitoring failures caused by environmental interference, providing highly reliable security assurance for subsequent operations. Secondly, based on the processed multi-dimensional data set of the parking lot, the parking lot is divided into static and dynamic areas. An adaptive threshold segmentation algorithm and a target tracking algorithm are used to filter out hot target areas, obtaining multi-dimensional features of the hot target areas. This method only processes dynamic area data, and thermal imaging temperature variance partitioning excludes static thermal interference, reducing computational load. Furthermore, it uses infrared thermal image compensation + millimeter-wave radar fusion to solve the false alarm problem of traditional solutions in rain, fog, obstruction, and thermal interference scenarios. This algorithm overcomes the physical limitations of single video surveillance, enabling accurate target perception in complex environments. It then sets classification labels corresponding to multi-dimensional features of thermal target areas and establishes a target classification model based on a dual-channel hybrid neural network. After training, the input classification labels are used for behavior pattern recognition, outputting the target type in the parking lot. This algorithm significantly improves classification accuracy through a dual-channel feature fusion mechanism (infrared morphology + motion timing), effectively solving the misjudgment problem caused by changes in light and heat source interference in traditional methods. It has strong deep intelligent recognition capabilities for pedestrian presence and temperature anomalies, reducing manual security costs. Finally, it identifies the behavioral characteristics of parking lot target types and sets risk levels, establishing a three-level response security decision-making mechanism. This overcomes the limitations of traditional passive security responses, intelligently matching three risk levels (high / medium / low) to achieve on-demand allocation of response resources, avoiding resource waste, and realizing automated collaborative control of security monitoring, fire protection facilities, access control systems, and lighting equipment.

[0050] Furthermore, to better illustrate the technical solutions of the embodiments of the present invention, such as... Figure 2 As shown, this paper describes a multi-dimensional security monitoring method for parking lots using infrared sensing technology, and provides a detailed explanation of a multi-dimensional security monitoring system for parking lots using infrared sensing technology, including the following:

[0051] S1. Deploy an infrared sensor array grid, collect complementary sensor data from the parking lot, form an initial multi-dimensional data set of the parking lot, perform baseline temperature drift compensation based on environmental adaptability, and obtain a processed multi-dimensional data set of the parking lot.

[0052] S1 includes the following steps:

[0053] S11. Select a security monitoring area in the parking lot, and set up infrared sensors in the security monitoring area to form an infrared sensor grid covering the parking lot, resulting in an infrared sensor array grid. The infrared nodes in the infrared sensor array grid represent infrared sensors, and an infrared thermal image is collected. Then, deploy complementary sensors for the parking lot. The complementary sensing sensors include millimeter-wave radar sensors, temperature sensors, and light sensors. Collect complementary sensor data synchronously at a fixed period. The complementary sensor data includes millimeter-wave radar signals, temperature data, and light status. Combined with the infrared thermal image, an initial multi-dimensional data set of the parking lot is formed.

[0054] S12. Based on environmental adaptability, baseline temperature drift compensation is performed on the initial parking lot multi-dimensional data set to obtain a processed parking lot multi-dimensional data set. The specific steps are as follows:

[0055] S121. Set sample points on the infrared heat map in the initial parking lot multi-dimensional data set. The sample points are pixels. Quantize the sample points, convert the infrared heat map into an infrared heat map matrix, and label the temperature values ​​of the pixels in the infrared heat map matrix.

[0056] S122. Set the ambient temperature gradient and blackbody radiation source. Within the range of the ambient temperature gradient, use a temperature sensor to measure the temperature value of the blackbody radiation source and establish a temperature calibration table. Set the sensor temperature drift coefficient, obtain the reference temperature value according to the temperature calibration table, obtain the temperature data in the initial multi-dimensional data set of the parking lot, and calculate the pixel temperature correction value in the infrared thermal image matrix. Where T represents the pixel temperature value in the infrared thermal image matrix, and k represents the sensor temperature drift coefficient. Representing temperature data, Indicates the reference temperature value;

[0057] S123. Based on the temperature data in the initial parking lot multi-dimensional data set, calculate the temperature change rate within a fixed time period, set a temperature threshold, and when the temperature change rate exceeds the temperature threshold, enter the temperature sudden change scenario. Introduce a compensation weight coefficient to compensate the pixel temperature correction values ​​in the infrared thermal image matrix again, and obtain the final pixel temperature values ​​in the infrared thermal image matrix. ,in The temperature change rate is represented by t, which represents a fixed time interval. Indicates the compensation weighting coefficient;

[0058] S124. The final temperature values ​​of the pixels in the infrared thermal image matrix are used to generate a compensated infrared thermal image, resulting in a processed multi-dimensional data set of the parking lot.

[0059] In this embodiment, by deploying an infrared sensor array grid, complementary sensor data from the parking lot is collected, and baseline temperature drift compensation is performed based on environmental adaptability to obtain a processed multi-dimensional parking lot data set. This method employs a two-stage temperature compensation mechanism (basic calibration + abrupt change compensation) to reduce thermal imaging temperature drift errors in heavy rain / scorching sun scenarios. The use of multi-source data collaboration overcomes monitoring failures caused by environmental interference, providing highly reliable security assurance for subsequent operations. Specifically, an infrared sensor grid covering the parking lot is deployed, an initial multi-dimensional parking lot data set is collected, and a blackbody radiation source experiment is conducted. At different ambient temperatures (e.g., -10℃, 0℃, 25℃, 40℃), the sensor output values ​​to the blackbody radiation source are measured, and temperature-voltage response curves are constructed. Blackbody calibration table; for example, during a summer afternoon downpour, the actual temperature at the entrance of an underground parking lot drops sharply from 38℃ to 26℃ within 15 minutes. The blackbody calibration table obtains a baseline value of 25.8℃, the sensor temperature drift coefficient is 0.35, and the sampling period is 1 second. The heat map matrix points are then corrected. Historical temperature data at the entrance of the underground parking lot is collected, the rate of change distribution is analyzed, and a histogram of the change per minute is created. The temperature change rate is ≤0.5℃ / min for 95% of the time period, so a temperature threshold of 0.5℃ / min is set, and secondary compensation is performed: a temperature change rate of -0.8℃ / min is detected (exceeding the temperature threshold of 0.5℃ / min). The compensation weight coefficient is then determined to be equal to the difference between the actual temperature and the baseline value divided by the product of the temperature threshold and the sampling period, i.e. Calculate the final temperature value of each pixel to eliminate thermal image distortion caused by sudden temperature drops;

[0060] S2. Based on the processed multi-dimensional parking lot data set, the parking lot is divided into static and dynamic areas. The hot target areas are selected by using an adaptive threshold segmentation algorithm and a target tracking algorithm to obtain the multi-dimensional features of the hot target areas.

[0061] S2 includes the following steps:

[0062] S21. Based on the compensated infrared heat map in the processed multi-dimensional parking lot data set, the compensated infrared heat map is gridded to generate an infrared heat map grid, which is then converted into a final infrared heat map matrix. The infrared heat map grid is marked on the final infrared heat map matrix. The temperature variance of all pixels within the infrared heat map grid is calculated to obtain the pixel temperature variance. A variance threshold is set. When the pixel temperature variance is greater than the variance threshold, the corresponding infrared heat map grid is marked as a dynamic area; otherwise, the corresponding infrared heat map grid is marked as a static area. The dynamic and static areas are classified as region attributes, and the region attributes are marked in the infrared heat map grid to obtain a partitioned infrared heat map.

[0063] S22. The dynamic region in the partitioned infrared thermal image is denoted as the infrared thermal image to be detected. The infrared thermal image to be detected is converted into an infrared thermal image matrix. An adaptive threshold segmentation algorithm is used to perform preliminary screening of the thermal target region to obtain the final thermally connected region. The specific steps are as follows:

[0064] S221. Calculate the average and standard deviation of the temperature values ​​of the pixels in the infrared thermal image matrix to be detected, and calculate the adaptive threshold. ,in This represents the average value of the pixel temperature. The standard deviation of pixel temperature values ​​is represented. When the pixel temperature value in the infrared thermal image matrix to be detected is greater than the adaptive threshold, the corresponding pixel is recorded as a hot pixel, forming a thermally connected region.

[0065] S222. Set an error region, which includes a small target error region and a large target error region. Filter the regions in the hot-connected region that are smaller than the small target error region and filter the regions in the hot-connected region that are larger than the large target error region to complete the ghost image filtering and obtain the final hot-connected region.

[0066] S23. Using the millimeter-wave radar signal from the processed multi-dimensional parking lot data set, perform a range-dimensional Fourier transform on the millimeter-wave radar signal to calculate the peak frequency and obtain the target distance; perform a Doppler-dimensional Fourier transform on the same target distance to determine the peak position and obtain the target velocity; the millimeter-wave radar signal forms a peak in the angular dimension, and the peak position corresponds to the target angle; combine the target distance, target velocity, and target angle to form a three-dimensional array of radar coordinates; map the three-dimensional array of radar coordinates to a spatial coordinate system to form point cloud data, and obtain millimeter-wave radar point cloud data;

[0067] The millimeter-wave radar point cloud data is clustered to generate moving target clusters. Then, a target tracking algorithm is used to filter out hot target regions from the final thermally connected regions. The specific steps are as follows:

[0068] S231. Based on the illumination status in the processed multi-dimensional parking lot data set, the illumination status of the parking lot is divided into normal status and interference status, and the millimeter-wave radar point cloud data is divided into normal millimeter-wave radar point cloud data and interference millimeter-wave radar point cloud data.

[0069] For normal and interference states, a neighborhood radius and a minimum number of neighborhood points are set respectively. Normal millimeter-wave radar point cloud data and interference millimeter-wave radar point cloud data form clusters within the neighborhood radius. When the number of normal millimeter-wave radar point cloud data and the number of interference millimeter-wave radar point cloud data within the neighborhood radius are greater than or equal to the minimum number of neighborhood points, the corresponding millimeter-wave radar point cloud data are merged into the cluster to generate a moving target cluster.

[0070] S232. Determine the center position of the moving target cluster and the center position of the final thermally connected region, and set a distance threshold. When the distance between the center position of the moving target cluster and the center position of the final thermally connected region is less than the distance threshold, the corresponding thermally connected region forms a thermal target region.

[0071] S24. Extract multi-dimensional features of the thermal target region from the thermal target region, calculate the area, aspect ratio, and circularity of the thermal target region to obtain infrared morphological features; calculate the mean temperature and temperature variance of the thermal target region to obtain temperature distribution features; then calculate the absolute value of velocity and acceleration of the thermal target region, and collect the motion sequence to obtain radar motion features; combine the infrared morphological features, temperature distribution features, and radar motion features to obtain multi-dimensional features of the thermal target region.

[0072] In this embodiment, based on the processed multi-dimensional parking lot data set, the parking lot is divided into static and dynamic areas. An adaptive threshold segmentation algorithm and a target tracking algorithm are used to filter out hot target areas, obtaining multi-dimensional features of these areas. This method only processes dynamic area data, and thermal imaging temperature variance partitioning eliminates static thermal interference, reducing computational load. Furthermore, infrared thermal image compensation combined with millimeter-wave radar fusion solves the false alarm problem of traditional solutions in rain, fog, obstruction, and thermal interference scenarios, overcoming the physical limitations of single video surveillance and achieving accurate target perception in complex environments. Specifically, for example, a 1m×1m grid coordinate system is established, and area attributes are labeled for each grid. By measuring the temperature variance of walls, columns, etc., the calculated temperature variance is less than 5℃. For example, the grid containing parking space B3 is divided into dynamic areas. The parking area is divided into two zones: a static zone (representing walls, fixed facilities, etc.) and a dynamic zone (representing lanes, parking spaces, etc.). The average temperature of the dynamic zone is calculated to be 28.7℃ with a standard deviation of 3.2℃. Three pixel blocks in region B3 (39.1℃, 38.5℃, 40.2℃) exceed the threshold of 38.3℃, forming a thermally connected region. Small targets (such as flying insects) with an area <0.01㎡ and large targets (such as heat waves) with an area >10㎡ are measured to complete the initial screening of thermal targets, and region B3 is retained. Millimeter-wave radar fusion: range-dimensional FFT (Fourier transform) → target distance (5.2m), Doppler-dimensional FFT → target velocity (3m / s), angle-dimensional FFT → target angle (azimuth 23°). The radar polar coordinates are mapped to the Cartesian coordinates of the parking lot. Specifically, the Cartesian coordinates (x, y) of the parking lot are respectively... , , where l represents the target distance. The target angle is represented, and point cloud data is generated. Point cloud clustering processing is performed: normal millimeter-wave radar point cloud data (neighborhood radius 0.5m, minimum number of points 3) and interference millimeter-wave radar point cloud data (neighborhood radius 0.8m, minimum number of points 4) are used to form a moving target cluster in parking space B3. The center coordinates (35.0, 18.8) are calculated, and then the center coordinates of the hot target (35.2, 18.6) are calculated. A distance threshold of 0.5m is set. At this point, the moving target cluster and the final thermally connected area are considered to coincide. The distance error is 0.28m < 0.5m, and they are confirmed as the same target, thus obtaining the hot target area.

[0073] S3. Set classification labels corresponding to the multi-dimensional features of the hot target area, establish a target classification model based on a dual-channel hybrid neural network, and after training, input the classification labels to perform behavior pattern recognition and output the parking lot target type.

[0074] S3 includes the following steps:

[0075] S31. Set pedestrian features, vehicle features, animal features, and stationary heat source features. Based on the infrared morphological features and temperature distribution features in the multi-dimensional features of the thermal target area, distinguish pedestrians, vehicles, animals, and stationary heat sources to obtain classification labels. Mark the classification labels in the thermal target area to obtain an infrared thermal image to be identified. Then, record the radar motion features in the multi-dimensional features of the thermal target area as the motion features to be identified.

[0076] S32. Establish a target classification model based on a dual-channel hybrid neural network, acquire sample data for training, perform behavior pattern recognition, and output the target type in the parking lot. The specific steps are as follows:

[0077] S321. The target classification model is set to include a thermal imaging feature channel and a spatiotemporal motion channel. The thermal imaging feature channel includes an input layer, a convolutional neural network, a max pooling layer, and an output layer. The spatiotemporal motion channel includes an input layer, a two-layer long short-term memory neural network, and an output layer. The output results of the thermal imaging feature channel and the spatiotemporal motion channel are combined and upscaled through dual-channel feature concatenation, and then the final output result is obtained through a fully connected layer to establish the target classification model.

[0078] S322. Collect multi-dimensional data of the parking lot again, and after processing with baseline temperature drift compensation, adaptive threshold segmentation algorithm and target tracking algorithm, label the data with classification tags to obtain infrared thermal image samples and motion feature samples, and divide them into sample training set and sample test set.

[0079] The sample training set is input into the target classification model for training. Infrared thermal image samples are input into the thermal imaging feature channel, and motion feature samples are input into the spatiotemporal motion channel. A maximum number of training rounds is set, and training is stopped when the maximum number of training rounds is reached, thus obtaining a trained target classification model.

[0080] S323. Then input the sample test set into the trained target classification model, set the accuracy threshold, and when the accuracy of the model output result is greater than the accuracy threshold, the final target classification model is obtained; otherwise, adjust the model weights until the accuracy of the model output result is greater than the accuracy threshold.

[0081] The infrared thermal image and motion features to be identified are input into the final target classification model for behavior pattern recognition, and the parking lot target type is output. The parking lot target type includes parking lot pedestrians, parking lot vehicles, parking lot animals and parking lot stationary heat sources.

[0082] In this embodiment, classification labels corresponding to multi-dimensional features of the thermal target area are set, and a target classification model is established based on a dual-channel hybrid neural network. After training, the classification labels are input for behavior pattern recognition, and the parking lot target type is output. This algorithm model significantly improves the classification accuracy through a dual-channel feature fusion mechanism (infrared morphology + motion time sequence), effectively solving the misjudgment problem caused by changes in light and heat source interference in traditional methods. It has strong deep intelligent recognition capabilities for pedestrian presence, temperature anomalies, etc., reducing the cost of manual security. Specifically, for example, classification labels are set: infrared morphological features distinguish people (tall, thin) The radar distinguishes between vehicles (rectangular) and animals (small circles) based on temperature distribution characteristics, differentiating between high-temperature vehicle engines and uniform human body heating; and between pedestrians moving intermittently and vehicles moving at a constant speed based on radar motion characteristics. The thermal imaging feature channel (processing infrared thermal images) receives 128×128 pixel infrared thermal images as input, uses a 3-layer convolutional neural network (32 / 64 / 128 filters), employs the ReLU activation function, uses a 2×2 window for max pooling, and outputs a 256-dimensional feature vector. The spatiotemporal motion channel (processing radar motion characteristics) receives time-series data (120 seconds long, 1 sample per second). The sample points (each containing velocity and acceleration) are analyzed using a dual-layer LSTM with 128 units per layer to handle temporal dependencies. The output layer outputs a 128-dimensional feature vector in a fully connected layer. Feature fusion involves concatenating the outputs from the two channels (256 + 128 = 384 dimensions), passing the concatenation through a fully connected layer (128 neurons) and a Softmax layer to output the probability distributions for four categories. Data collected from the same parking lot over six months includes 852 pedestrian samples (including walking, crouching, running, etc.), 1203 vehicle samples (including entering, leaving, reversing, etc.), and 16 animal samples. 7 cases (cats, dogs, etc.) and 736 cases of stationary heat source samples (unextinguished trains, charging piles, stray objects, etc.); the maximum number of training rounds was set to 50, the optimizer was Adam (learning rate 0.001), and the loss function was cross-entropy loss. The final target classification model was trained, showing an accuracy of 98% for pedestrian recognition, 97% for vehicle recognition, 93% for animal recognition, and 99% for stationary heat source recognition; when a 128×128 cropped image of region B3 plus a 120-second velocity / acceleration sequence were input, the model output probability was: pedestrian 92.3%, the target type in the parking lot is pedestrian;

[0083] S4. Identify the behavioral characteristics of the target type in the parking lot, set risk levels, establish a three-level response security decision-making mechanism, and realize multi-dimensional security monitoring of the parking lot;

[0084] S4 includes the following steps:

[0085] S41. Set vehicle distance thresholds and vehicle proximity time. When the distance between a pedestrian and a vehicle in the parking lot target type is less than the vehicle distance threshold and the duration exceeds the vehicle proximity time, a suspected theft behavior feature is triggered. When pedestrians and animals in the parking lot target type are located in the restricted area of ​​the parking lot, an illegal intrusion behavior feature is triggered. When the aspect ratio of pedestrians in the parking lot target type changes, a person falling to the ground behavior feature is triggered. Set a temperature threshold. When the average temperature of vehicles in the parking lot target type exceeds the temperature threshold, a vehicle fire behavior feature is triggered, and a set of behavioral features is identified.

[0086] S42. Based on the behavioral feature set, risk levels are set. When a suspected theft behavior is triggered, the corresponding infrared heat map is archived, and a low-risk alert is activated. When an illegal intrusion behavior is triggered, the behavior is reported to the security monitoring platform, and a medium-risk warning is activated. When a person falls to the ground or a vehicle is on fire, staff are notified to handle the situation on-site, and a high-risk alarm is activated. A three-level response security decision-making mechanism is established to achieve multi-dimensional security monitoring of the parking lot.

[0087] In this embodiment, behavioral characteristics of parking lot target types are identified, risk levels are set, and a three-level response security decision-making mechanism is established. This breaks through the limitations of traditional passive security responses, intelligently matching three risk levels (high / medium / low) to achieve on-demand allocation of response resources, avoiding resource waste, and realizing automated collaborative control of security monitoring, fire protection facilities, access control systems, and lighting equipment. Specifically, for example, a rule is set: if a pedestrian approaches a vehicle in the parking lot at a distance of more than 0.5 meters for more than 2 minutes, it is considered a suspected theft. Trigger conditions for different behavioral characteristics are set. In parking space 12 of area A, if the length-to-width ratio of a pedestrian in the parking lot changes from 1.2 to 0.3 (falling posture), the system determines that the person has fallen behavior characteristics and broadcasts to 12. People near parking space number 1, please assist; a passenger needs medical help. Next to the charging station in Zone B, the vehicle temperature is 142℃ (increased by 100℃ within 2 minutes). The system determines this as a vehicle fire (data on vehicle spontaneous combustion is obtained and analyzed; a fire occurs when the vehicle temperature exceeds 100℃ within 1 minute). The temperature exceeds the 100℃ combustion threshold, and the precise coordinates are sent to 119: B2 floor - Zone B - Charging station number 3. At the entrance to the equipment room, a pedestrian lingered in the restricted area for 120 seconds. The system determines this as an illegal intrusion and apprehends the person attempting to steal cables. At parking space number 7 in Zone C, a pedestrian remained 0.3m away from the car for 300 seconds. The system determines this as a suspected theft; after monitoring verification, it was confirmed to be a designated driver waiting for the car owner.

[0088] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0089] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A multi-dimensional security monitoring system for parking lots utilizing infrared sensing technology, characterized in that, include: The temperature drift compensation module is used to collect complementary sensor data of the parking lot, perform baseline temperature drift compensation based on environmental adaptability, and obtain a processed multi-dimensional data set of the parking lot. The hot target area filtering module is used to divide the parking lot into static and dynamic areas based on the processed multi-dimensional parking lot data set, and to filter out the hot target areas using an adaptive threshold segmentation algorithm and a target tracking algorithm to obtain the multi-dimensional features of the hot target areas. The target type recognition module is used to set classification labels corresponding to multi-dimensional features of hot target areas, establish a target classification model, and after training, input the classification labels to perform behavior pattern recognition and output the target type of the parking lot. The security decision-making mechanism module is used to identify the behavioral characteristics of target types in parking lots, set risk levels, and establish a three-level response security decision-making mechanism. The collected parking lot complementary sensor data includes: Infrared sensors are installed in the security monitoring area to form an infrared sensor grid covering the parking lot, and infrared thermal images are collected; complementary sensors are also deployed in the parking lot to collect data from the complementary sensors simultaneously, forming an initial multi-dimensional data set of the parking lot. The baseline temperature drift compensation based on environmental adaptability includes: The infrared heat map in the initial parking lot multi-dimensional data set is replaced with an infrared heat map matrix; Establish a temperature calibration table, obtain the reference temperature value, and calculate the temperature correction value of the pixel in the infrared thermal image matrix; Once a sudden temperature change scenario is detected, the temperature correction values ​​of pixels in the infrared thermal image matrix are compensated, the final temperature values ​​of pixels in the infrared thermal image matrix are calculated, a compensated infrared thermal image is generated, and a processed multi-dimensional data set of the parking lot is obtained. The step of dividing the parking lot into static and dynamic areas based on the processed multi-dimensional parking lot data set includes: The compensated infrared thermal image is meshed to generate an infrared thermal image grid, which is then converted into a final infrared thermal image matrix. The temperature variance of each pixel is calculated, and the infrared thermal image grid is marked as dynamic and static areas to obtain a partitioned infrared thermal image.

2. The parking lot multi-dimensional security monitoring system utilizing infrared sensing technology according to claim 1, characterized in that, The process of using an adaptive threshold segmentation algorithm and a target tracking algorithm to filter out hot target regions includes: The dynamic area in the partitioned infrared thermal image is recorded as the infrared thermal image to be detected, and converted into the infrared thermal image matrix to be detected. Hot pixels are marked to form a thermally connected region. Then, ghost image filtering is performed to obtain the final thermally connected region. The millimeter-wave radar signals in the processed multi-dimensional parking lot data set are converted into millimeter-wave radar point cloud data. Clustering is performed on the millimeter-wave radar point cloud data to generate moving target clusters; Determine the center position of the moving target cluster and the center position of the final thermally connected region, and set a distance threshold. When the distance between the center position of the moving target cluster and the center position of the final thermally connected region is less than the distance threshold, the corresponding thermally connected region forms a thermal target region.

3. A parking lot multi-dimensional security monitoring system utilizing infrared sensing technology according to claim 2, characterized in that, The obtained multi-dimensional features of the thermal target region include: Infrared morphological features, temperature distribution features, and radar motion features of the thermal target area are extracted to obtain multi-dimensional features of the thermal target area.

4. A multi-dimensional security monitoring system for parking lots utilizing infrared sensing technology according to claim 3, characterized in that, The establishment of the target classification model includes: Based on the multi-dimensional features of the thermal target area, classification labels are obtained, and infrared thermal images and motion features to be identified are generated. The target classification model is defined as having a thermal imaging feature channel and a spatiotemporal motion channel. The thermal imaging feature channel includes an input layer, a convolutional neural network, a max pooling layer, and an output layer. The spatiotemporal motion channel includes an input layer, a two-layer long short-term memory neural network, and an output layer. The output results of the thermal imaging feature channel and the spatiotemporal motion channel are combined and upscaled through dual-channel feature concatenation, and then passed through a fully connected layer to obtain the final output result, thus establishing the target classification model.

5. A multi-dimensional security monitoring system for parking lots utilizing infrared sensing technology according to claim 4, characterized in that, The process of recognizing behavioral patterns by inputting classification labels after training includes: Infrared thermal image samples and motion feature samples are acquired and divided into a sample training set and a sample test set. These are then input into the target classification model for training to obtain the final target classification model. The infrared thermal image and motion features to be identified are input into the final target classification model for behavior pattern recognition, and the parking lot target type is output.

6. A multi-dimensional security monitoring system for parking lots utilizing infrared sensing technology according to claim 5, characterized in that, The establishment of the three-level response security decision-making mechanism includes: Based on the parking lot target type, identify suspected theft behavior characteristics, illegal intrusion behavior characteristics, personnel falling to the ground behavior characteristics, and vehicle fire behavior characteristics to obtain a behavior feature set; Risk levels are set based on the behavioral feature set, and a three-level response security decision-making mechanism is established.

Images