Method for optimizing control strategy of pedestrian recognition and motion trail analysis system of monitoring video

Abstract

The invention provides a method for optimizing a control strategy of a pedestrian recognition and motion trail analysis system based on a surveillance video. The method comprises the following steps: acquiring experimental data; an initial analysis model is established for the pedestrian analysis subsystem, a verified analysis model is obtained, behavior characteristics of the pedestrian analysis subsystem are simulated, and a first simulation result is output and obtained; based on a deep learning algorithm, an initial behavior dimension reduction model is established for the pedestrian analysis subsystem; training the initial behavior dimension reduction model by using the first simulation result; verifying the trained behavior dimension reduction model by using second experimental data to obtain a verified behavior dimension reduction model; setting optimization parameters and a control strategy optimization target; and adopting a machine learning-simulation optimization algorithm to generate optimization parameters corresponding to the control strategy, and inputting the verified behavior dimension reduction model to obtain a prediction result. According to the invention, the identification accuracy and prediction efficiency of the monitoring system can be improved, and the adaptability and practicability of the system are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent video surveillance systems, and more specifically, to a method for optimizing control strategies of a surveillance video pedestrian recognition and motion trajectory analysis system. Background Art

[0002] Existing surveillance video systems face some challenges in pedestrian recognition and motion trajectory analysis. With the acceleration of urbanization, the improvement of public safety needs, and the development of artificial intelligence technology, the requirements for surveillance video systems are getting higher and higher. Traditional surveillance systems mainly rely on simple image processing techniques to identify and track pedestrians, but these methods often have difficulty in providing accurate recognition and analysis results in complex environments, such as low light, high dynamics, or multi-pedestrian scenes. In addition, existing systems have low computational efficiency when processing large-scale data and are difficult to adapt to changing environmental factors and pedestrian behavior patterns. In terms of technical principles, many existing systems use feature-based recognition methods, which rely on manually extracted features such as shape, color, and texture. These features are easily disturbed in complex environments, resulting in reduced recognition accuracy. At the same time, some systems try to use machine learning methods to improve recognition rates, but these methods often require a large amount of labeled data and have limitations in model generalization capabilities.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the existing system is not adaptable enough in complex environments, making it difficult to accurately capture and analyze the movement trajectory of pedestrians; there is a lack of effective models to comprehensively consider the coupling relationship between pedestrian behavior and environmental factors; and there are bottlenecks in large-scale data processing and real-time requirements, which limits the practicality and efficiency of the monitoring system. Summary of the invention

[0004] The present invention provides a method for optimizing control strategy of a monitoring video pedestrian recognition and motion trajectory analysis system, comprising:

[0005] Step S1. Conducting experimental tests on the video acquisition subsystem to obtain experimental data; wherein the working conditions of the experimental tests cover a first test working condition set and a second test working condition set; the experimental data include first experimental data corresponding to the first test working condition set and second experimental data corresponding to the second test working condition set;

[0006] Step S2. Based on the pedestrian-environment coupling relationship of the pedestrian behavior analysis model, an initial analysis model is established for the pedestrian analysis subsystem, and a verified analysis model is obtained through verification with the first experimental data; the behavior characteristics of the pedestrian analysis subsystem are simulated using the verified analysis model, and a first simulation result is obtained by output;

[0007] Step S3. Based on the deep learning algorithm, an initial behavior dimensionality reduction model is established for the pedestrian analysis subsystem; the initial behavior dimensionality reduction model is trained using the first simulation result to obtain a trained behavior dimensionality reduction model;

[0008] Step S4. Using the second experimental data to verify the trained behavior dimensionality reduction model to obtain a verified behavior dimensionality reduction model;

[0009] Step S5. Setting optimization parameters and control strategy optimization targets; wherein the control strategy optimization targets include expected behavior distribution, expected maximum speed, expected minimum speed, and expected speed inconsistency within the pedestrian analysis subsystem;

[0010] Step S6. Use machine learning-simulation optimization algorithm to generate optimization parameters corresponding to the control strategy, and input the verified behavior dimensionality reduction model to obtain prediction results; continuously update the optimization parameters and input the verified behavior dimensionality reduction model to obtain corresponding prediction results until the corresponding prediction results reach the control strategy optimization target set in step S5.

[0011] Furthermore, the intersection of the first test condition set and the second test condition set is an empty set.

[0012] Furthermore, the experimental data include actual position values ​​of each pedestrian in the pedestrian analysis subsystem, actual speed values ​​collected by the speed sensor layout, actual behavior distribution, actual maximum speed, actual minimum speed and actual speed inconsistency.

[0013] Furthermore, the initial analysis model in step S2 includes a pedestrian behavior sub-model and a pedestrian motion sub-model; wherein the pedestrian behavior sub-model is expressed as:

[0014]

[0015] Among them, H i represents the behavior description of the i-th pedestrian; n is the number of factors in the environment that affect the pedestrian's behavior; w ij is the weight of the correlation between the ith pedestrian and the jth factor; P ij is the position parameter of the i-th pedestrian; E j is the jth environmental factor; g is the behavior analysis function that comprehensively considers location and environmental factors;

[0016] The pedestrian motion submodel is expressed as:

[0017]

[0018] Among them, V ij is the speed of the i-th pedestrian under the j-th environmental factor; v base is the basic speed; m is the number of behavioral features that affect the speed; h is ijkis the activity level of pedestrian i with feature k under environmental factor j; S ijk is the adjustment value of feature k to speed, according to pedestrian behavior H i and environmental factors j Sure.

[0019] Furthermore, the input of the initial analysis model in step S2 is the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the first test condition set, and the output is the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem, wherein the simulation behavior distribution B i Determined by the following formula:

[0020]

[0021] Among them, t0 and t1 are the start and end time of the simulation respectively.

[0022] Furthermore, the verified analysis model obtained through verification of the first experimental data in step S2 specifically includes: performing a first-stage comparison of the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem with the first experimental data; if the error of the first-stage comparison is less than a preset first threshold, retaining the initial analysis model as the verified analysis model; otherwise, adjusting the parameters of the initial analysis model and performing a first-stage comparison until the error of the first-stage comparison is less than the preset first threshold.

[0023] Furthermore, the verified behavior dimensionality reduction model obtained by verifying the trained behavior dimensionality reduction model using the second experimental data in step S4 specifically includes: inputting the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the second test condition set into the trained behavior dimensionality reduction model, outputting a second simulation result for a second-stage comparison with the second experimental data, and if the error of the second-stage comparison is less than a preset second threshold, executing step S5; otherwise, repeating steps S3 and S4 until the error of the second-stage comparison is less than the preset second threshold.

[0024] Furthermore, the optimization parameters in step S5 are the acquisition frequency and resolution of the video acquisition subsystem.

[0025] Furthermore, the first test condition set includes a low-light environment test and a high-dynamic environment test; the second test condition set includes a multi-pedestrian environment test and a complex background environment test.

[0026] Furthermore, the initial behavior dimensionality reduction model described in step S3 is a fusion model of a deep learning convolutional neural network and a recurrent neural network; wherein, the number of convolutional neural network layers is set to 3, each convolutional layer is followed by a maximum pooling layer, the recurrent neural network uses a long short-term memory network LSTM structure, the number of LSTM layers is set to 2, the learning rate is set to 0.0001, the number of iterations is set to 5000 times, the split ratio of the validation set and the training set obtained by the first simulation result is 8:2, the length of a single extracted data segment is 15, and the training target residual is set to 0.00005.

[0027] According to the above-mentioned embodiment of the present invention, at least the following beneficial effects are achieved: the control strategy optimization method of the pedestrian recognition and motion trajectory analysis system of the monitoring video can improve the accuracy of pedestrian recognition and the precision of motion trajectory analysis. Through the experimental data obtained by experimental testing, combined with the pedestrian behavior analysis model, the system can establish a more accurate initial analysis model, and further optimize the behavior dimension reduction model through the deep learning algorithm. This method can not only enhance the adaptability of the system to complex environments, but also improve the prediction and analysis capabilities of pedestrian behavior. In addition, the method can dynamically adjust the optimization parameters and control strategies through the machine learning-simulation optimization algorithm to achieve precise control of the expected behavior distribution, speed and other parameters in the pedestrian analysis subsystem. This continuous optimization process can ensure that the system can achieve the expected performance goals under different test conditions, thereby providing a more stable and reliable monitoring effect in practical applications. Through this method, the system can better serve the fields of public safety, traffic management, etc., and provide more scientific and effective data support for decision makers. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which:

[0029] Figure 1 A flowchart of a method for optimizing a control strategy for a surveillance video pedestrian recognition and motion trajectory analysis system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present invention, and are not intended to limit the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to fully convey the scope of the present invention to those skilled in the art.

[0031] Those skilled in the art will appreciate that the embodiments of the present invention may be implemented as a system, device, apparatus, method or computer program product. Therefore, the present invention may be implemented in the following forms, namely: complete hardware, complete software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0032] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0033] Reference below Figure 1 , Figure 1 The flowchart of the control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system provided by one embodiment of the present invention is as follows. Figure 1 As shown, a control strategy optimization method 100 for a monitoring video pedestrian recognition and motion trajectory analysis system includes:

[0034] Step S1, conducting an experimental test on the video acquisition subsystem to obtain experimental data; wherein the working conditions of the experimental test cover a first test working condition set and a second test working condition set; the experimental data include first experimental data corresponding to the first test working condition set and second experimental data corresponding to the second test working condition set;

[0035] Step S2, based on the pedestrian-environment coupling relationship of the pedestrian behavior analysis model, establish an initial analysis model for the pedestrian analysis subsystem, obtain a verified analysis model through first experimental data verification; use the verified analysis model to simulate the behavior characteristics of the pedestrian analysis subsystem, and output a first simulation result;

[0036] Step S3, based on the deep learning algorithm, establish an initial behavior dimensionality reduction model for the pedestrian analysis subsystem; use the first simulation result to train the initial behavior dimensionality reduction model to obtain a trained behavior dimensionality reduction model;

[0037] Step S4, using the second experimental data to verify the trained behavior dimensionality reduction model to obtain a verified behavior dimensionality reduction model;

[0038] Step S5, setting optimization parameters and control strategy optimization targets; wherein the control strategy optimization targets include expected behavior distribution, expected maximum speed, expected minimum speed, and expected speed inconsistency within the pedestrian analysis subsystem;

[0039] Step S6, use the machine learning-simulation optimization algorithm to generate the optimization parameters corresponding to the control strategy, input the verified behavior dimensionality reduction model to obtain the prediction results; continuously update the optimization parameters and input the verified behavior dimensionality reduction model to obtain the corresponding prediction results, until the corresponding prediction results reach the control strategy optimization goal set in step S5.

[0040] It should be noted that the control strategy optimization method for the monitoring video pedestrian recognition and motion trajectory analysis system is a systematic solution that integrates video acquisition, pedestrian behavior analysis, deep learning algorithm and machine learning optimization algorithm. Through experimental testing and data analysis, the method continuously optimizes the pedestrian recognition and motion trajectory analysis model to adapt to different monitoring environments and needs.

[0041] Specifically, the nouns involved in this method include video acquisition subsystem, pedestrian analysis subsystem, pedestrian behavior analysis model, deep learning algorithm, behavior dimensionality reduction model, machine learning-simulation optimization algorithm, etc. During the implementation process, it is necessary to set up a set of experimental test conditions, including a first test condition set and a second test condition set. These condition sets may include low-light, high-dynamic, multiple pedestrians, and complex background environments. At the same time, it is necessary to define the type of experimental data, such as pedestrian position, speed, behavior distribution, etc., and select appropriate deep learning models and machine learning algorithms for training and optimization.

[0042] Preferably, when implementing step S1, multi-camera synchronous acquisition technology can be used to ensure the accuracy and completeness of the experimental data. In step S2, the establishment of the initial analysis model can be based on the existing pedestrian behavior database and environmental parameters, and preliminary model construction can be performed through an expert system. In step S3, the fusion model of the deep learning convolutional neural network and the recurrent neural network can be further refined. For example, the number of layers and parameters of the convolutional neural network can be adjusted according to actual data to improve the recognition ability of the model. In step S5, the setting of the optimization parameters may include but is not limited to the frame rate, resolution, encoding format, etc. of the video acquisition to adapt to different monitoring scenarios and requirements. More specifically, an adaptive adjustment mechanism can be used to dynamically adjust these parameters according to the real-time monitoring effect to achieve the best monitoring effect.

[0043] In some embodiments, the intersection of the first test condition set and the second test condition set is an empty set.

[0044] It should be noted that the intersection of the first test condition set and the second test condition set is an empty set, which means that the two test condition sets are completely independent and have no overlap. Such a design can ensure that the system can be effectively tested and verified in different and unique environments, thereby improving the adaptability and robustness of the system in diverse scenarios.

[0045] Specifically, the first test condition set may include low-light environment testing and high-dynamic environment testing, while the second test condition set may include multi-pedestrian environment testing and complex background environment testing. These test condition sets cover various challenges that pedestrian recognition and motion trajectory analysis systems may encounter, such as recognition in low-light or fast-moving pedestrians, and trajectory analysis in the presence of multiple people or in complex and changing backgrounds.

[0046] Preferably, we can set specific parameters and concepts. For example, in low-light environment testing, the system's sensitivity and exposure time can be set to adapt to different lighting conditions. In high-dynamic environment testing, the system's frame rate and dynamic range can be adjusted to capture fast-moving pedestrians. In multi-pedestrian environment testing, more advanced pedestrian detection algorithms can be used to distinguish and track multiple pedestrians. In complex background environment testing, background subtraction technology can be introduced to reduce the impact of background interference on pedestrian recognition. In addition, specific thresholds can be set for each test condition set to evaluate the system's performance in a specific environment.

[0047] In some embodiments, the experimental data includes actual position values ​​of each pedestrian in the pedestrian analysis subsystem, actual speed values ​​collected by the speed sensor layout, actual behavior distribution, actual maximum speed, actual minimum speed and actual speed inconsistency.

[0048] It should be noted that this method emphasizes the comprehensiveness and uniqueness of the experimental data. The intersection of the first test condition set and the second test condition set is an empty set, which means that the two sets of test conditions are completely independent and are designed for different environmental conditions and scenarios to ensure the robustness and adaptability of the system under diverse conditions.

[0049] Specifically, the experimental data covers the actual position values ​​of each pedestrian in the pedestrian analysis subsystem, the actual speed values ​​collected by the speed sensor layout, the actual behavior distribution, the actual maximum speed, the actual minimum speed, and the actual speed inconsistency. These parameters are key indicators for evaluating system performance and ensure the accuracy of model training and verification. For example, the actual position value can be extracted from the surveillance video by the video analysis software, while the actual speed value is provided by the speed sensors arranged in a specific area.

[0050] Preferably, the experimental data can be collected through an automated data acquisition system that can monitor and record the location and speed information of pedestrians in real time. In addition, in order to improve the quality and availability of data, data cleaning and preprocessing techniques can be used, such as removing outliers and noise, to ensure that the data input to the model is accurate and reliable. In terms of parameter setting, the density and distribution of sensors, as well as the frequency and resolution of data acquisition can be adjusted according to the complexity of the actual scene and monitoring requirements.

[0051] In some embodiments, the initial analysis model in step S2 includes a pedestrian behavior sub-model and a pedestrian motion sub-model; wherein the pedestrian behavior sub-model is represented as:

[0052]

[0053] Among them, H i represents the behavior description of the i-th pedestrian; n is the number of factors in the environment that affect the pedestrian's behavior; w ij is the weight of the correlation between the ith pedestrian and the jth factor; P ij is the position parameter of the i-th pedestrian; E j is the jth environmental factor; g is the behavior analysis function that comprehensively considers location and environmental factors;

[0054] The pedestrian motion submodel is expressed as:

[0055]

[0056] Among them, V ij is the speed of the i-th pedestrian under the j-th environmental factor; v base is the basic speed; m is the number of behavioral features that affect the speed; h is ijk is the activity level of pedestrian i with feature k under environmental factor j; S ijk is the adjustment value of feature k to speed, according to pedestrian behavior H i and environmental factors j Sure.

[0057] It should be noted that the initial analysis model is the core component of the pedestrian recognition and motion trajectory analysis system. It consists of a pedestrian behavior sub-model and a pedestrian motion sub-model, which work together to simulate and analyze the behavior and motion of pedestrians under different environmental factors.

[0058] Specifically, the behavior description in the pedestrian behavior sub-model represents the behavior characteristics of pedestrians in the environment, while the number and correlation weights of factors affecting pedestrian behavior in the environment determine the diversity and complexity of pedestrian behavior. The pedestrian motion sub-model considers the impact of basic speed and behavior characteristics on speed, where the activity and adjustment values ​​are dynamically calculated based on pedestrian behavior and environmental factors.

[0059] Preferably, the input parameters of the initial analysis model can be adjusted according to the actual monitoring environment to adapt to different monitoring scenarios. For example, in a low-light environment, the weight of light sensitivity can be increased, while in a high-dynamic environment, the recognition ability of fast-moving objects can be improved. In addition, the behavior analysis function in the pedestrian behavior sub-model can adopt a variety of algorithms, such as support vector machines (SVM) or neural networks, to improve the generalization ability of the model. The basic speed in the pedestrian motion sub-model can be preset according to the general walking speed of pedestrians, while the activity and adjustment values ​​can be learned from historical data through machine learning algorithms.

[0060] In some embodiments, the input of the initial analysis model in step S2 is the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the first test condition set, and the output is the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem, wherein the simulation behavior distribution B i Determined by the following formula:

[0061]

[0062] Among them, t0 and t1 are the start and end time of the simulation respectively.

[0063] It should be noted that the output results of the initial analysis model include the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem. These simulation results are based on the simulation of pedestrian behavior and movement by the initial analysis model, and are the basis for further verification and optimization of the model.

[0064] Specifically, the simulated behavior distribution predicts the possible action paths and behavior patterns of pedestrians by comprehensively analyzing the behavior patterns of pedestrians in the monitoring scene, including but not limited to walking direction, speed changes, etc. The integral in the formula represents the integration of the probability density function of pedestrian behavior between the start and end time of the simulation to obtain the expected results of the behavior distribution. In addition, the simulated maximum speed, minimum speed and speed inconsistency predict the maximum speed, minimum speed and speed change inconsistency that pedestrians may reach based on pedestrian behavior and environmental factors.

[0065] Preferably, in order to improve the accuracy and practicality of the simulation, the following settings can be adopted: the calculation of the simulated behavior distribution can be based on a pedestrian behavior analysis function, which takes into account pedestrian position parameters and environmental factors, and trains a large amount of historical data through a machine learning algorithm to obtain weight parameters. At the same time, the maximum speed and minimum speed of the simulation can be adjusted according to the activity of pedestrian behavior characteristics and environmental factors, which can be determined by data collected by environmental sensors. More specifically, the start time of the simulation can be set to the moment when monitoring starts, and the end time can be set to a preset monitoring period, such as 30 seconds or 1 minute. In addition, in order to verify the accuracy of the model, the simulation results can be compared and analyzed with the actual collected data. If the error is within an acceptable range, the model can be considered to be valid. If the error is large, it is necessary to adjust the model parameters or retrain the model until the accuracy requirements are met.

[0066] In some embodiments, the verified analysis model obtained through verification of the first experimental data in step S2 specifically includes: performing a first-stage comparison of the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem with the first experimental data; if the error of the first-stage comparison is less than a preset first threshold, retaining the initial analysis model as the verified analysis model; otherwise, adjusting the parameters of the initial analysis model and performing a first-stage comparison until the error of the first-stage comparison is less than the preset first threshold.

[0067] It should be noted that the first-stage comparison of the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency within the pedestrian analysis subsystem with the first experimental data means that the output of the simulation model needs to be compared with the actual collected experimental data to evaluate the accuracy and reliability of the model.

[0068] Specifically, the first initial simulation behavior distribution refers to the distribution of pedestrian behavior patterns within a specific time period predicted by the simulation model, the first initial simulation maximum speed and the first initial simulation minimum speed refer to the maximum and minimum moving speeds of pedestrians in the monitoring area predicted by the simulation model, and the first initial simulation speed inconsistency refers to the irregularity of pedestrian speed changes. The parameter settings of these concepts will be determined based on the characteristics of the experimental data and the requirements of the system design.

[0069] Preferably, when performing the first stage comparison, a statistical method can be used to quantify the error, such as using the mean square error (MSE) or the mean absolute error (MAE) as an evaluation indicator. If the error of the first stage comparison is less than the preset first threshold, it means that the simulation model has a high degree of consistency with the experimental data and can be retained as a verified analysis model. If the error exceeds the threshold, the parameters of the simulation model need to be adjusted, which may include adjusting the weights of the pedestrian behavior analysis function, the basic speed, the feature activity, etc., to improve the prediction accuracy of the model. In addition, it is also possible to consider introducing an adaptive learning mechanism so that the model can automatically adjust parameters according to new experimental data to better adapt to changes in the actual environment.

[0070] In some embodiments, the verified behavior dimensionality reduction model obtained by verifying the trained behavior dimensionality reduction model using the second experimental data in step S4 specifically includes: inputting the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the second test condition set into the trained behavior dimensionality reduction model, outputting a second simulation result for a second-stage comparison with the second experimental data, and if the error of the second-stage comparison is less than a preset second threshold, executing step S5; otherwise, repeating steps S3 and S4 until the error of the second-stage comparison is less than the preset second threshold.

[0071] It should be noted that the core of using the second experimental data to verify the trained behavioral dimensionality reduction model is to ensure the effectiveness and accuracy of the model in practical applications. This process involves inputting data collected in the actual environment into the trained model and comparing it with the experimental data to evaluate the performance of the model.

[0072] Specifically, the second experimental data in step S4 refers to the data on pedestrian behavior and motion trajectory collected under the second test condition set. This includes but is not limited to information such as the location, speed, and behavior pattern of the pedestrian. The trained behavior dimensionality reduction model is a model optimized by a deep learning algorithm, which can convert high-dimensional pedestrian behavior data into a low-dimensional representation while retaining key information. At this stage, the input parameters and environmental parameters of the model need to be set according to the actual situation to ensure that the model can adapt to different test conditions.

[0073] Preferably, in order to further improve the validation effect of the model, the output results of the model can be analyzed in more detail. For example, specific performance indicators such as prediction accuracy and recall rate can be set to quantify the performance of the model. In addition, it is also possible to consider introducing additional regularization terms to prevent the model from overfitting, or to use cross-validation methods to evaluate the generalization ability of the model. In terms of parameter setting, the network structure can be adjusted according to the characteristics of the experimental data, such as increasing or decreasing the number of convolutional layers or LSTM layers, or adjusting the learning rate and number of iterations to achieve the best training effect.

[0074] In some embodiments, the optimization parameters in step S5 are the acquisition frequency and resolution of the video acquisition subsystem.

[0075] It should be noted that the optimization parameters are the acquisition frequency and resolution of the video acquisition subsystem, which are related to the quality and speed of the system's acquisition of video data. The acquisition frequency determines the smoothness of the video, while the resolution determines the clarity of the video. The optimization of these two parameters can improve the system's recognition accuracy of pedestrian behavior and the accuracy of motion trajectory analysis.

[0076] Specifically, the acquisition frequency of the video acquisition subsystem is usually measured in frames per second (FPS), while the resolution is defined by the number of pixels, such as 1920x1080 for high-definition video. In the implementation process, these parameters can be set according to the needs of the actual monitoring environment. For example, in scenes that require quick response, a higher acquisition frequency can be selected; in scenes that require high definition to identify details, a higher resolution can be selected. In addition, these parameters can also be adjusted according to the processing power of the pedestrian analysis subsystem to ensure the real-time nature of the data and the accuracy of the analysis.

[0077] Preferably, in order to further optimize the performance of the video acquisition subsystem, an adaptive adjustment strategy can be adopted. For example, the system can automatically adjust the acquisition frequency and resolution according to the dynamic changes of the current monitoring scene. In scenes with a small number of pedestrians and simpler behaviors, the system can reduce the acquisition frequency to reduce the amount of data processing; while in scenes with dense pedestrians or complex behaviors, the acquisition frequency and resolution can be increased to obtain more detailed information. In addition, image quality enhancement techniques such as noise reduction and sharpening can be combined to further improve the quality of video data, thereby providing more reliable input data for pedestrian recognition and motion trajectory analysis.

[0078] In some embodiments, the first test condition set includes a low-light environment test and a high-dynamic environment test; the second test condition set includes a multi-pedestrian environment test and a complex background environment test.

[0079] It should be noted that the first test condition set and the second test condition set are designed to ensure that the system can maintain efficient and accurate performance in different environments. The first test condition set includes low-light environment tests and high-dynamic environment tests, which help evaluate the system's recognition capabilities in low-light or rapidly changing scenes. The second test condition set includes multi-pedestrian environment tests and complex background environment tests, which help evaluate the system's recognition capabilities in scenes with dense crowds or cluttered backgrounds.

[0080] Specifically, the low-light environment test of the first test condition set can simulate conditions at night or indoors with insufficient light, and test the performance of the system under conditions with almost no natural light or artificial light. The high-dynamic environment test can simulate fast-moving objects or rapidly changing scenes, and test the stability and accuracy of the system when dealing with fast-moving pedestrians. The multi-pedestrian environment test of the second test condition set can simulate places with large traffic such as shopping malls and stations, and test the system's recognition ability when dealing with multiple pedestrians appearing at the same time. The complex background environment test can simulate scenes with rich background elements, diverse colors and textures, such as markets and festival celebrations, to test the robustness of the system under complex visual interference.

[0081] Preferably, for the implementation of the first test condition set, a specific test scene can be set, such as using specific lighting equipment in a low-light environment to simulate night conditions, and adjusting the exposure parameters of the video acquisition subsystem to adapt to light changes. In high-dynamic environment testing, rapidly changing scenes can be captured by setting fast-moving simulated pedestrians or using high-speed cameras. For the second test condition set, multiple simulated pedestrians can be arranged in the test site, and the system can be tested using complex background patterns or actual scene recordings. In addition, the behavior analysis function and motion sub-model parameters of the pedestrian analysis subsystem can be adjusted according to different test conditions to adapt to different environmental conditions and pedestrian behavior patterns.

[0082] In some embodiments, the initial behavior dimensionality reduction model in step S3 is a fusion model of a deep learning convolutional neural network and a recurrent neural network; wherein, the number of convolutional neural network layers is set to 3, each convolutional layer is followed by a maximum pooling layer, the recurrent neural network uses a long short-term memory network LSTM structure, the number of LSTM layers is set to 2, the learning rate is set to 0.0001, the number of iterations is set to 5000 times, the segmentation ratio of the validation set and the training set obtained by the first simulation result is 8:2, the length of a single extracted data segment is 15, and the training target residual is set to 0.00005.

[0083] It should be noted that the first test condition set includes low-light environment testing and high-dynamic environment testing, which indicates that the system needs to be tested under different lighting and dynamic conditions to ensure its robustness. In extension, low-light environment testing may involve nighttime or low-light scenes, while high-dynamic environment testing may include fast-moving objects or rapidly changing lighting conditions.

[0084] Specifically, low-light environment testing may include video acquisition at night or in low-light conditions indoors to test the system's ability to recognize pedestrians under these conditions. High-dynamic environment testing may involve testing the system's performance under strong or rapidly changing lighting conditions, such as direct sunlight or reflected conditions. In these tests, the system's video acquisition subsystem needs to adjust its acquisition frequency and resolution to adapt to different lighting conditions.

[0085] Preferably, for low-light environment testing, the system's video acquisition subsystem can use a high-sensitivity camera and adjust the acquisition frequency to reduce noise while increasing resolution to maintain image clarity. For high-dynamic environment testing, a camera with a wide dynamic range (WDR) can be used, and algorithm optimization can be used to balance the details of different lighting areas to ensure accurate capture of pedestrian features. In addition, the system can use an adaptive algorithm to automatically adjust parameters according to changes in ambient light to achieve the best monitoring effect.

[0086] The above-mentioned embodiments of the present invention have the following beneficial effects: The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system of the present invention can significantly improve the recognition accuracy and analysis efficiency of the system in complex environments. This method can ensure that the system can achieve highly accurate pedestrian behavior prediction and trajectory analysis under different test conditions through comprehensive experimental testing, pedestrian behavior analysis, deep learning algorithms and machine learning optimization algorithms. Specifically, this method can adapt to changing environmental factors and pedestrian behavior patterns, improve the system's understanding and prediction capabilities of pedestrian behavior, and thus enhance the practicality and reliability of the monitoring system.

[0087] Furthermore, by optimizing the control strategy, the system can precisely adjust the parameters of the video acquisition subsystem, such as acquisition frequency and resolution, to meet different monitoring needs. In the tests of the first test condition set and the second test condition set, the system was able to be specially optimized for environments such as low light, high dynamics, multiple pedestrians and complex backgrounds, ensuring that high-quality monitoring data can be provided under various conditions. In addition, the application of the fusion model of deep learning convolutional neural network and recurrent neural network can further enhance the system's ability to recognize pedestrian behavior characteristics, making the system more efficient in processing large-scale data, and can provide strong technical support for fields such as public safety and traffic management.

[0088] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0089] The above descriptions are only some preferred embodiments of the present invention and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with (but not limited to) technical features with similar functions disclosed in the embodiments of the present invention to form a technical solution.

Claims

1. A method for optimizing control strategy of a surveillance video pedestrian recognition and motion trajectory analysis system, the system comprising a video acquisition subsystem and a pedestrian analysis subsystem, characterized in that: The method comprises the following steps: Step S1. Conducting experimental tests on the video acquisition subsystem to obtain experimental data; wherein the working conditions of the experimental tests cover a first test working condition set and a second test working condition set; the experimental data include first experimental data corresponding to the first test working condition set and second experimental data corresponding to the second test working condition set; Step S2. Based on the pedestrian-environment coupling relationship of the pedestrian behavior analysis model, an initial analysis model is established for the pedestrian analysis subsystem, and a verified analysis model is obtained through verification with the first experimental data; the behavior characteristics of the pedestrian analysis subsystem are simulated using the verified analysis model, and a first simulation result is obtained by output; Step S3. Based on the deep learning algorithm, an initial behavior dimensionality reduction model is established for the pedestrian analysis subsystem; the initial behavior dimensionality reduction model is trained using the first simulation result to obtain a trained behavior dimensionality reduction model; Step S4. Using the second experimental data to verify the trained behavior dimensionality reduction model to obtain a verified behavior dimensionality reduction model; Step S5. Setting optimization parameters and control strategy optimization targets; wherein the control strategy optimization targets include expected behavior distribution, expected maximum speed, expected minimum speed, and expected speed inconsistency within the pedestrian analysis subsystem; Step S6. Use machine learning-simulation optimization algorithm to generate optimization parameters corresponding to the control strategy, and input the verified behavior dimensionality reduction model to obtain prediction results; continuously update the optimization parameters and input the verified behavior dimensionality reduction model to obtain corresponding prediction results until the corresponding prediction results reach the control strategy optimization target set in step S5.

2. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 1 is characterized in that: The intersection of the first test condition set and the second test condition set is an empty set.

3. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 2 is characterized in that: The experimental data include the actual position values ​​of each pedestrian in the pedestrian analysis subsystem, the actual speed values ​​collected by the speed sensor layout, the actual behavior distribution, the actual maximum speed, the actual minimum speed and the actual speed inconsistency.

4. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 3 is characterized in that: The initial analysis model in step S2 includes a pedestrian behavior sub-model and a pedestrian motion sub-model; wherein the pedestrian behavior sub-model is expressed as: Among them, H i represents the behavior description of the i-th pedestrian; n is the number of factors in the environment that affect the pedestrian's behavior; w ij is the weight of the correlation between the ith pedestrian and the jth factor; P ij is the position parameter of the i-th pedestrian; E j is the jth environmental factor; g is the behavior analysis function that comprehensively considers location and environmental factors; The pedestrian motion submodel is expressed as: Among them, V ij is the speed of the i-th pedestrian under the j-th environmental factor; v base is the basic speed; m is the number of behavioral features that affect the speed; h is ijk is the activity level of pedestrian i with feature k under environmental factor j; S ijk is the adjustment value of feature k to speed, according to pedestrian behavior H i and environmental factors j Sure.

5. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 4 is characterized in that: The input of the initial analysis model in step S2 is the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the first test condition set, and the output is the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem, where the simulation behavior distribution B i Determined by the following formula: Among them, t0 and t1 are the start and end time of the simulation respectively.

6. The control strategy optimization method for monitoring video pedestrian recognition and motion trajectory analysis system according to claim 5 is characterized in that: The verified analysis model obtained by the first experimental data verification in step S2 specifically includes: performing a first-stage comparison of the first initial simulation behavior distribution, the first initial simulation maximum speed, the first initial simulation minimum speed and the first initial simulation speed inconsistency in the pedestrian analysis subsystem with the first experimental data; if the error of the first-stage comparison is less than the preset first threshold, retaining the initial analysis model as the verified analysis model; otherwise, adjusting the parameters of the initial analysis model and performing a first-stage comparison until the error of the first-stage comparison is less than the preset first threshold.

7. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 6 is characterized in that: The verified behavior dimensionality reduction model obtained by verifying the trained behavior dimensionality reduction model using the second experimental data in step S4 specifically includes: inputting the actual input parameters and actual environmental parameters of the pedestrian analysis subsystem in the second test condition set into the trained behavior dimensionality reduction model, outputting a second simulation result for a second-stage comparison with the second experimental data, and if the error of the second-stage comparison is less than the preset second threshold, executing step S5; otherwise, repeating steps S3 and S4 until the error of the second-stage comparison is less than the preset second threshold.

8. The control strategy optimization method for monitoring video pedestrian recognition and motion trajectory analysis system according to claim 7 is characterized in that: The optimization parameters in step S5 are the acquisition frequency and resolution of the video acquisition subsystem.

9. The control strategy optimization method for monitoring video pedestrian recognition and motion trajectory analysis system according to claim 8 is characterized in that: The first test condition set includes low-light environment test and high-dynamic environment test; The second test condition set includes multi-pedestrian environment testing and complex background environment testing.

10. The control strategy optimization method of the monitoring video pedestrian recognition and motion trajectory analysis system according to claim 9 is characterized in that: The initial behavior dimensionality reduction model described in step S3 is a fusion model of a deep learning convolutional neural network and a recurrent neural network; wherein, the number of convolutional neural network layers is set to 3, each convolutional layer is followed by a maximum pooling layer, the recurrent neural network uses a long short-term memory network LSTM structure, the number of LSTM layers is set to 2, the learning rate is set to 0.0001, the number of iterations is set to 5000 times, the segmentation ratio of the validation set and the training set obtained by the first simulation result is 8:2, the length of a single extracted data segment is 15, and the training target residual is set to 0.00005.

Images