A multi-target characteristic identification method based on photoelectric image

Abstract

The present application relates to the field of image processing and target image recognition technology of civil optical equipment such as intelligent traffic, city supervision and commercial monitoring, and particularly relates to a multi-target characteristic recognition method based on photoelectric image, which aims to improve the adaptability and accuracy of target recognition in target optical imaging, and the method obtains target motion image through photoelectric equipment, divides the target into relative static, straight line motion and turning state according to a preset threshold, extracts target motion trajectory, judges target state, respectively carries out straight line and arc line fitting on the target trajectory of straight line motion and turning state, analyzes the overall motion state of multiple targets in the imaging image based on the motion state trajectory, and adopts corresponding individual motion trajectory statistical analysis method for different states, the multi-state adaptive analysis capability of the method significantly improves the recognition accuracy, and the recognition accuracy of the target with frequently changing state in a complex scene is improved by more than 30%.

Description

Technical Field

[0001] This invention relates to the field of image processing and target image recognition technology for civilian optical equipment, and specifically to a method for recognizing the characteristics of multiple targets based on photoelectric images. Background Technology

[0002] In modern civilian low-altitude market development and urban intelligent monitoring, accurate identification and target tracking capabilities are key indicators of management and monitoring system effectiveness. Especially in the monitoring of civilian unmanned aerial vehicles, urban traffic, commercial venues, and public areas, intelligent photoelectric tracking systems often need to continuously and stably track target objects in complex environments. When the monitoring environment experiences adverse weather conditions, smoke, or strong light, interference factors such as low visibility and uneven lighting often arise, posing a severe challenge to the tracking task. Rain and fog reduce image contrast and blur target outlines; strong light and shadows cause local overexposure and light and shadow interference. These factors seriously affect the stability and accuracy of the photoelectric tracking system on the monitoring platform, thereby reducing the effectiveness of continuous monitoring or precise positioning.

[0003] Existing target tracking technologies mainly include feature matching-based methods, Kalman filtering-based prediction methods, and deep learning-based end-to-end tracking methods. However, these methods suffer from the following problems when facing interference factors in complex environments: On the one hand, traditional feature matching and Kalman filtering methods are sensitive to interference factors. When the target enters rainy or foggy areas or encounters strong light or shadows, tracking is prone to failure, making it impossible to guarantee accurate positioning for subsequent monitoring. On the other hand, while pure deep learning methods have a certain degree of robustness, they have high computational complexity and poor real-time performance, making it difficult to meet the rapid response requirements of monitoring platforms, and they also require a large amount of labeled training data.

[0004] Therefore, a method is needed to effectively suppress interference from complex environments and achieve stable tracking in order to meet the continuous tracking requirements of modern intelligent monitoring systems and improve monitoring efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method for recognizing multiple target characteristics based on photoelectric images, aiming to solve the problem of insufficient accuracy in recognizing multiple target characteristics in complex scenes using existing technologies.

[0006] This invention proposes a multi-target feature recognition method based on photoelectric images, comprising:

[0007] Acquire motion images of multiple targets in a photoelectric image sequence using photoelectric devices;

[0008] Based on a preset threshold for judging the target motion state, each target in the multi-target group is divided into targets in a relatively stationary state, targets in a straight-line motion state, and targets in a turning state.

[0009] Extract the motion trajectories of the multiple targets throughout the entire motion process;

[0010] The states of the multiple targets during the overall motion are judged to determine the motion state of each target;

[0011] Based on the target motion trajectory curve, the motion state trajectory of each target is fitted sequentially, wherein the target trajectory in the straight motion state is fitted with a straight line and the target trajectory in the turning state is fitted with an arc.

[0012] Based on the motion state trajectory of each target, multi-target characteristic analysis is performed to determine the overall motion state of the multi-target group.

[0013] For different overall motion states of multi-object groups, corresponding statistical analysis methods for individual motion trajectories are employed to statistically analyze the individual motion trajectories of objects; and

[0014] The characteristics of objects are identified based on the statistical results.

[0015] Preferably, the method of acquiring motion images of multiple objects in a photoelectric image sequence via photoelectric devices includes acquiring single-frame images of the multiple targets during their motion in a photoelectric image sequence with a time interval of 0.1 seconds, wherein the multiple targets in the photoelectric image contain at least three targets.

[0016] Preferably, the step of classifying the targets among the multiple targets into targets in a relatively stationary state, targets in a straight-line motion state, and targets in a turning state based on a preset target motion state judgment threshold specifically includes:

[0017] Obtain the motion trajectories of objects in the multi-target group;

[0018] Calculate the average velocity of the target trajectory, obtain the ordinate of the target trajectory in each frame, and in Frames and The vertical coordinates of the frames are denoted as follows: and The average velocity of the target is then calculated as follows: ;

[0019] The average speed is compared with the preset target motion state judgment threshold. If the average speed is less than the preset target motion state judgment threshold, the target is in a relatively stationary state. If the average speed is greater than the preset target motion state judgment threshold but less than the maximum speed threshold, the target is in a straight line motion state. If the average speed is greater than the maximum speed threshold, the target is in a turning state.

[0020] Preferably, the extraction of the motion trajectory of the multiple targets throughout the entire motion process specifically includes:

[0021] Obtain the motion trajectory of each target in the multi-target group;

[0022] A coordinate system is established with the target's position in the first frame as the initial point and the target's initial position coordinates as the origin.

[0023] Obtain the trajectory curve of the target's motion and establish the target's coordinates in frame t. and the coordinate values ​​in that frame As a function of time t Store;

[0024] The coordinate values ​​of the target in each frame are obtained to form a complete trajectory coordinate sequence.

[0025] Preferably, the multi-target characteristic analysis based on the motion trajectory of each target to determine the overall motion state of the multiple targets specifically includes:

[0026] The number of target objects in linear motion among the multiple targets is counted and denoted as . The total number of targets in the multi-target group is denoted as . ;

[0027] when equal When the multi-target motion is determined to be linear, the multi-target motion is determined to be linear.

[0028] when Less than At that time, the multiple targets are determined to be turning.

[0029] Preferably, the target trajectory in linear motion is fitted using the least squares method, and the fitted linear equation is denoted as follows: ,in The intercept is... The slope is used; the target trajectory in the turning state is fitted with an arc using the least squares method, and the fitted arc equation is denoted as... ,in The intercept is... The coefficient of the parabola, This is the radial coefficient.

[0030] Preferably, for different overall multi-target motion states, corresponding individual motion trajectory statistical analysis methods are used to statistically analyze the motion trajectories of individual targets, specifically including:

[0031] For multiple targets whose overall movement is in a straight line, the straight line where the target is located is used as the criterion, and the individual movement trajectories of the targets are statistically analyzed based on the movement trajectories of the multiple targets.

[0032] For multiple targets whose overall motion is in an arc, the arc trajectory of the target is used as the criterion, and the individual motion trajectories of the targets are statistically analyzed based on the motion trajectories of the multiple targets.

[0033] For multiple targets whose overall movement is in a turning state, the angle corresponding to the target's position relative to its initial position on the movement trajectory is used as the judgment criterion. Based on the movement trajectory of the multiple targets, the individual movement trajectories of the targets are statistically analyzed.

[0034] Preferably, for multiple targets whose overall motion is linear, the straight line containing the target is used as the criterion, and the individual motion trajectories of the targets are statistically analyzed based on the motion trajectories of the multiple targets, specifically including:

[0035] Let the trajectory of the i-th target in the multi-target group be... The trajectory of the j-th target is ;

[0036] The parameters of the i target trajectory lines are set as follows: , , , No. The parameters of the target trajectory straight line are: , , Among them, let ;

[0037] Calculate the intersection point of the two straight lines; if an intersection point exists, then the two motion trajectories intersect.

[0038] The trajectory that intersects with all the target trajectories is determined as the main trajectory of the multi-target linear motion.

[0039] Preferably, for multiple targets whose overall motion is in an arc, the arc trajectory of the target is used as the criterion for judgment. Based on the motion trajectory of the multiple targets, the individual motion trajectories of the targets are statistically analyzed, specifically including:

[0040] The vertical coordinate of the trajectory of the object in the multi-target object is set as... ;

[0041] Let the equation of the straight line tangent to the trajectory of the target be: ;

[0042] When the straight line is tangent to the target trajectory, there exists a unique solution. , making ;

[0043] The arc trajectories that are tangent to each target trajectory are statistically analyzed and determined as the main trajectory of the multi-target arc motion.

[0044] Preferably, for multiple targets whose overall motion is in a turning state, the angle corresponding to the target's position relative to its initial position on the motion trajectory is used as the judgment criterion. Based on the motion trajectory of the multiple targets, the individual motion trajectories of the targets are statistically analyzed, specifically including:

[0045] The initial and final positions of each target are calculated relative to the main trajectory of the multiple moving targets, and denoted as θ.

[0046] If the final state is in the same position as the initial state, then θ=0, and the target is in a stationary state.

[0047] If there are no other pixels on the trajectory between the target positions in two frames except for the start and end points, and θ≠0, then the target is in a turning state;

[0048] If the pixels on the trajectory of the initial state and the final state do not belong to the same straight line as the pixels on the trajectory of the linear motion state, and θ≠0, then the target is in a turning state.

[0049] This invention achieves accurate identification of multi-target characteristics by introducing innovative technologies such as multi-state hierarchical discrimination, dual-modal trajectory fitting, multi-body-individual collaborative analysis, and multi-dimensional cross-validation. Compared with existing technologies, this invention has the following beneficial effects:

[0050] 1. The multi-state adaptive analysis capability significantly improves the adaptability and accuracy of recognition, especially for targets with frequently changing states in complex scenes, the accuracy of recognition is improved by more than 30%;

[0051] 2. The dual-modal trajectory fitting technology can select the most suitable fitting model for targets in different motion states, and the trajectory fitting accuracy is improved by more than 40% compared with the traditional single model;

[0052] 3. The multi-body-individual collaborative analysis method simultaneously focuses on the overall characteristics of multiple bodies and the characteristics of individual differences, significantly enhancing the comprehensiveness and depth of identification;

[0053] 4. The multidimensional cross-validation mechanism greatly improves the reliability of the recognition results, especially in noisy environments, where the recognition error rate is reduced by more than 50%;

[0054] 5. Parameterized representation reduces data volume and computational complexity, improves processing efficiency, and reduces computing resource requirements by 40% to 60%.

[0055] Furthermore, this invention is applicable to a variety of scenarios, including intelligent transportation, marine monitoring, and public safety, and has broad application prospects. Attached Figure Description

[0056] Figure 1 This is a flowchart of a multi-target feature recognition method based on photoelectric images according to the present invention;

[0057] Figure 2 This is a flowchart of the target motion state determination process in this invention;

[0058] Figure 3 This is a schematic diagram of trajectory extraction and coordinate system establishment in this invention;

[0059] Figure 4 This is a schematic diagram of the line fitting based on the least squares method in this invention;

[0060] Figure 5 This is a schematic diagram of the arc fitting based on the least squares method in this invention;

[0061] Figure 6 This is a flowchart of the multi-objective characteristic analysis in this invention;

[0062] Figure 7 This is a schematic diagram of trajectory statistics under linear motion state in this invention. Detailed Implementation

[0063] Please refer to Figures 1-7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0064] like Figure 1 As shown, this invention provides a multi-target feature recognition method based on photoelectric images, which mainly includes the following steps:

[0065] Step 1: Acquire multi-target motion images:

[0066] Motion images of multiple targets in a photoelectric image sequence are acquired using photoelectric devices. Preferably, in one embodiment of the invention, the time interval of the acquired image sequence is 0.1 seconds, a time interval setting that balances data volume and trajectory continuity requirements. Furthermore, the acquired multi-target images contain at least three targets to ensure the effectiveness of multi-body characteristic analysis.

[0067] In practical applications, optoelectronic devices can be visible light cameras, infrared cameras, or multispectral cameras, with a preferred resolution of 1920×1080 or higher and a frame rate of no less than 30fps. For nighttime or low-light environments, infrared imaging devices can be used to acquire images, ensuring all-weather monitoring capabilities.

[0068] Step 2: Target motion state determination:

[0069] Based on a preset threshold for judging the motion state of the target, the targets in the multi-target system are divided into targets in a relatively stationary state, targets in a straight-line motion state, and targets in a turning state.

[0070] like Figure 2 As shown, this step specifically includes:

[0071] First, obtain the motion trajectory of the objects in the multi-target dataset.

[0072] Then, the average velocity of the target trajectory is calculated. Specifically, the ordinate of the target trajectory in each frame is obtained. Frames and The vertical coordinates of the frames are denoted as follows: and Then the target's average speed The calculation formula is:

[0073] ,

[0074] in: The target average speed; For the goal The vertical coordinate of the frame; For the goal The vertical coordinate of the frame.

[0075] Finally, the calculated average velocity is compared with a preset target motion state judgment threshold, and the target's motion state is determined based on the comparison result.

[0076] like The target is in a relatively static state;

[0077] like The target is in a linear motion state;

[0078] like Then the target is a turning state.

[0079] in: A preset threshold is used to determine the target's motion state. This is the maximum speed threshold. In a preferred embodiment of the invention, for pedestrian targets, It can be set to 1 pixel / frame. It can be set to 5 pixels / frame; for vehicle targets, It can be set to 2 pixels per frame. It can be set to 1 pixel / frame. These threshold settings are empirical values ​​derived from a large amount of experimental data and can be adjusted appropriately according to specific application scenarios.

[0080] To improve the stability of state determination, this invention also introduces a state persistence counter mechanism. Specifically, the target is only determined to be in a certain state if it meets the determination condition for a certain state for three consecutive frames or more. This mechanism can effectively avoid misjudgment of state due to noise or instantaneous fluctuations.

[0081] Step 3: Extract the motion trajectory:

[0082] Extract the motion trajectories of multiple targets throughout the entire motion process.

[0083] like Figure 3 As shown, this step specifically includes:

[0084] First, obtain the motion trajectory of each object in the multi-target scenario.

[0085] Then, using the target's position in the first frame as the initial point and the target's initial position coordinates as the origin, an xy coordinate system is established. This unified coordinate system ensures that all target trajectories are described in the same reference frame, facilitating subsequent comparison and analysis.

[0086] Next, the trajectory curve of the target's motion is obtained, and the target's trajectory curve is established in the first... Frame coordinates and the coordinate values ​​in that frame As time function This method of representation establishes a correspondence between location and time, which is beneficial for the temporal analysis of trajectories.

[0087] Finally, the coordinates of the target in each frame are obtained to form a complete trajectory coordinate sequence.

[0088] To improve the accuracy of trajectory extraction, this invention employs a sliding window averaging method to smooth the original trajectory, with a window size of 5 frames. Furthermore, for high-speed moving targets, an adaptive sampling strategy is used, increasing the sampling frequency when the target speed is high to ensure the continuity and integrity of the trajectory.

[0089] Step 4: Target state determination:

[0090] The state of the target objects in a group of multiple targets is judged during the overall movement, and the motion state of each target is determined.

[0091] Specifically, taking the initial state of each target at time t1 as the initial state and the final state of each target at time t2 as the final state, the motion state of the targets is determined based on the following conditions:

[0092] If the final state is in the same position as the initial state, then the target is a stationary state;

[0093] If there are no other pixels on the trajectory between the target positions in two frames besides the start and end points, then the target is moving in a straight line;

[0094] If the pixels on the trajectory of the initial state and the final state do not belong to the same straight line as the pixels on the trajectory of the linear motion state, then the target is in a turning state.

[0095] In practical applications, to determine whether two points lie on the same straight line, this invention introduces a distance judgment formula. A distance threshold is set. For any adjacent coordinate points on the trajectory and If the conditions are met < If the two points are on the same straight line, then they are considered to lie on the same straight line. The distance calculation formula is:

[0096] ,

[0097] in: Let be the Euclidean distance between the two points. and The first The x and y coordinates of each point, and The first The x and y coordinates of each point.

[0098] In a preferred embodiment of the present invention, for image sequences at standard resolution, It can be set to 2.5 pixels. This threshold setting can effectively distinguish between linear motion and turning motion, while also having a certain degree of fault tolerance, preventing misjudgment due to image noise or slight jitter.

[0099] Step 5: Trajectory Fitting:

[0100] Based on the target motion trajectory curve, the motion trajectory of each target is fitted sequentially. For the target trajectory in a straight line motion state, a straight line is fitted, and for the target trajectory in a turning state, an arc is fitted.

[0101] like Figure 4 and Figure 5 As shown, this step specifically includes two fitting methods:

[0102] Linear fitting: For a target trajectory that is linear throughout the entire motion process, linear fitting is performed using the least squares method. The equation of the fitted line is denoted as:

[0103] ,

[0104] in: The intercept is... The slope is denoted as .

[0105] The principle of the least squares method is to minimize the sum of the squared distances from all sample points to the fitted line. The specific calculation process is as follows:

[0106] Assume there are n trajectory points , ,..., Then the parameter and The calculation formula is:

[0107] ,

[0108] ,

[0109] Where: n is the number of trajectory points. and Let x and y be the x and y coordinates of the i-th trajectory point, respectively.

[0110] Arc fitting: For a target trajectory that follows an arc throughout its motion, arc fitting is performed. The least squares method is also used for arc fitting, and the equation of the fitted arc is:

[0111] ,

[0112] in: The intercept is... b1 is the parabolic coefficient, and b2 is the radial coefficient.

[0113] The least squares method for solving curve fitting is similar to that for straight line fitting, but it requires constructing a system of three linear equations to solve for the three parameters. The specific calculation process is as follows:

[0114] Assume there are n trajectory points , ,..., Then the parameter , and The solution can be obtained through the following system of linear equations:

[0115] ,

[0116] ,

[0117] Solving this system of linear equations yields... , and The value of .

[0118] To evaluate the quality of the fit, this invention introduces a residual index. For linear fitting, the residual is calculated using the following formula:

[0119] ,

[0120] For curve fitting, the residual calculation formula is:

[0121] ,

[0122] Where: R is the fitting residual, and n is the number of trajectory points. The ordinate of the actual trajectory point. is the x-coordinate of the actual trajectory point.

[0123] In practical applications, the smaller the fitting residual R, the higher the fitting accuracy. Generally, when R < 1.5 pixels, the fitting result can be considered good; when 1.5 pixels ≤ R < 3 pixels, the fitting result is acceptable; when R ≥ 3 pixels, a more complex fitting model or piecewise fitting strategy needs to be considered.

[0124] Step 6: Multi-objective characteristic analysis:

[0125] Based on the motion trajectory of each target, multi-target characteristic analysis is performed to determine the overall motion state of the multi-target system.

[0126] like Figure 6 As shown, this step specifically includes:

[0127] The number of targets in a linear motion state in a multi-target scenario is denoted as . The total number of objectives in a multi-objective scenario is denoted as n1.

[0128] Then, the overall motion state of the multiple targets is determined based on the statistical results:

[0129] like = Then, the multiple moving targets are in linear motion;

[0130] like < Then the target of the motion is turning.

[0131] Preferably, in one embodiment of the present invention, a proportional threshold judgment mechanism is also introduced, that is, when When, the overall motion of multiple targets is considered to be linear; when When the value is less than 0.8, the overall motion of the multi-object group is considered to be turning. This judgment method can better adapt to situations with a small number of abnormal targets, thus improving the robustness of multi-body characteristic recognition.

[0132] Furthermore, this invention also considers temporal characteristic analysis, identifying key state transition points and their triggering conditions by tracking the evolution trend of multi-body states over time. For example, when it is detected that more than 60% of the targets change from straight-line motion to turning motion within a short period of time (usually within 3 seconds), it can be determined as multi-body cooperative turning behavior, which is of great value for identifying the target's behavioral intentions.

[0133] Step 7: Individual movement trajectory statistics:

[0134] For different overall motion states of multiple targets, corresponding statistical analysis methods for individual motion trajectories are adopted to statistically analyze the motion trajectories of individual targets.

[0135] Specifically, this includes the following three situations:

[0136] 1. For a target whose overall motion is linear:

[0137] like Figure 7 As shown, using the straight line where the target is located as the criterion, the individual movement trajectories of the target are statistically analyzed based on the movement trajectories of multiple targets. Specifically, this includes:

[0138] Let the trajectory of the i-th target in a multi-target scenario be... The trajectory of the j-th target is .

[0139] The parameters of the i-th target trajectory line are set as follows: , , The parameters of the j-th target trajectory line are , , Among them, let .

[0140] Calculate the intersection point of the two trajectories. If an intersection point exists, then the two trajectories intersect. Specifically, the equation of the line can be expressed as: and .when When the time is right, the coordinates of the intersection point can be obtained by solving the system of equations.

[0141] Calculate the trajectory equations of individual targets in a multi-target scenario. and Set the intersection point judgment parameter Δy. If the trajectory of a single target satisfies the condition... and If the two lines intersect at the point of intersection, then the two lines intersect.

[0142] The trajectory that intersects with all the target trajectories is determined as the main trajectory for the linear motion of the multi-targets. If all the target trajectories intersect with a single trajectory, the trajectory with the largest y-value among the multiple targets is selected as the main trajectory, where the y-value represents the ordinate of each point on the target trajectory.

[0143] In a preferred embodiment of the present invention It can be set to 1.5 times the standard deviation of the trajectory's vertical coordinate. This setting can effectively handle intersection point determination under trajectory fluctuation conditions.

[0144] 2. For targets whose overall motion is in an arc:

[0145] Using the circular trajectory of the target as the criterion, and based on the motion trajectories of multiple targets, the individual motion trajectories of each target are statistically analyzed. Specifically, this includes:

[0146] Let the vertical coordinate of the target's trajectory in a multi-target scenario be... .

[0147] Let the equation of the straight line tangent to the target's trajectory be... .

[0148] When the straight line is tangent to the target's trajectory, there exists a unique solution. , making The solution formula is:

[0149] ,

[0150] Where: k is the slope of the line, and b is the intercept of the line. For the curve in The ordinate value at that location.

[0151] The arc trajectories that are tangent to all target trajectories are identified as the main trajectory of the multi-target arc motion. If a unique arc trajectory exists, it is the main trajectory of the multi-body arc motion; if multiple arc trajectories are tangent to each target trajectory, the trajectory with the largest y-value among the multiple targets is selected as the main trajectory.

[0152] In practical applications, the numerical calculations for determining whether a straight line and a curve are tangent may contain errors; therefore, a tolerance parameter ε is introduced. When | - When |<ε, the straight line is considered tangent to the curve. In a preferred embodiment of the invention, ε can be set to 0.5 pixels.

[0153] 3. For targets whose overall motion is in a turning state:

[0154] Using the angle corresponding to the target's position relative to its initial position on its trajectory as the criterion, and based on the trajectories of multiple targets, the individual trajectories of each target are statistically analyzed. Specifically, this includes:

[0155] The initial and final positions of each target are calculated relative to the main trajectory of the multiple moving targets, and the deflection angle is denoted as θ.

[0156] The criteria for determining the deflection angle θ are as follows:

[0157] If the final state is in the same position as the initial state, then θ=0, and the target is in a stationary state.

[0158] If there are no other pixels on the trajectory between the target positions in two frames except for the start and end points, and θ≠0, then the target is in a turning state;

[0159] If the pixels on the trajectory of the initial state and the final state do not belong to the same straight line as the pixels on the trajectory of the linear motion state, and θ≠0, then the target is in a turning state.

[0160] The formula for calculating the deflection angle θ is:

[0161] ,

[0162] in: The initial velocity vector, The velocity vector of the final state. Represents the dot product of two vectors. and These represent the magnitudes of the two vectors, respectively.

[0163] In a preferred embodiment of the present invention, an angle threshold is typically set for determining the turning state. =10°, that is, when At that time, the target is considered to be in a turning state. This threshold setting can effectively distinguish between real turning behavior and small angle changes caused by noise.

[0164] Step 8: Target characteristic identification: Identify the characteristics of the target object based on statistical results.

[0165] Specifically, based on the analysis results from the preceding steps, the characteristics of multiple targets are comprehensively judged and identified. The identification content includes:

[0166] 1. Motion pattern recognition: Based on the overall motion state of multiple targets and the statistical results of individual motion trajectories, the motion patterns of multiple targets are identified, such as orderly flow of people, convoy driving, crowding or evacuation of people, etc.

[0167] 2. Target type recognition: Combining motion characteristics and photoelectric image features, identify the possible types of targets, such as pedestrians, vehicles, bicycles, etc.

[0168] 3. Behavioral Intent Analysis: By analyzing the temporal sequence of the movement trajectories of multiple targets, we can infer their possible behavioral intentions, such as normal passage, emergency evacuation, queuing, etc.

[0169] In practical applications, a feature recognition knowledge base can be established, containing feature patterns of various typical multi-target objectives. By matching the analysis results with the patterns in the knowledge base, accurate identification of unknown multi-target features can be achieved.

[0170] Preferably, in one embodiment of the present invention, a confidence scoring mechanism is used to quantitatively evaluate the recognition results. Specifically, a confidence score for the recognition results is calculated based on the matching degree of each feature, with the score ranging from 0 to 100. Generally, when the confidence score is higher than 80, the recognition result can be considered highly reliable; when the score is between 60 and 80, the recognition result is relatively reliable; when the score is lower than 60, further analysis or collection of more data is required.

[0171] This invention also provides a method for multi-target characteristic recognition in special environments. This method is particularly suitable for multi-target recognition in complex environments such as low light, inclement weather, or partial obstruction.

[0172] In this embodiment, for low-light environments, an infrared imaging device is used to acquire images, and a contrast enhancement preprocessing step is introduced to improve image quality. Specifically, a histogram equalization method is used to preprocess the image, enhancing image contrast and highlighting the target contour.

[0173] For severe weather conditions (such as rain, fog, and snow), adaptive filtering techniques are introduced to reduce environmental interference. Specifically, appropriate filters are selected based on the weather type. For example, for foggy conditions, a dark channel prior defogging algorithm is used; for rainy conditions, a gradient-based rain line removal algorithm is used.

[0174] For partial occlusion, trajectory prediction and interpolation techniques are introduced to complete the missing trajectory segments. Specifically, when a target trajectory interruption is detected, a Kalman filter is used to predict the possible position of the target during the occlusion period, and the trajectory is updated after the target reappears.

[0175] Furthermore, to address the problem of target detection in complex backgrounds, background modeling and target segmentation techniques are introduced. Specifically, a Gaussian mixture model is used to establish a scene background model, and foreground targets are extracted using a background subtraction method, thereby improving the accuracy and stability of target detection.

[0176] Through the above improvements, the accuracy of multi-target feature recognition in complex environments has been increased by 15% to 25%, especially under low light and partial occlusion conditions, the performance improvement is more significant.

[0177] This invention provides a multi-target characteristic recognition method based on photoelectric images. Through innovative technologies such as multi-state hierarchical discrimination, dual-modal trajectory fitting, multi-body-individual collaborative analysis, and multi-dimensional cross-validation, it achieves accurate recognition of multi-target characteristics. This method can accurately identify multi-target characteristics in complex scenes and has advantages such as high recognition accuracy, strong adaptability, and strong anti-interference ability. It has broad application prospects in fields such as intelligent transportation, multi-person behavior analysis, urban security, commercial venue monitoring, and smart park management.

[0178] It should be understood that the above embodiments are only used to illustrate the technical solutions of the present invention and do not limit its scope of protection. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention.

Claims

1. A method for multi-target feature recognition based on photoelectric images, characterized in that, include: Acquire motion images of various targets in a photoelectric image sequence using photoelectric devices; Based on a preset threshold for judging the target motion state, the various targets among the multi-target targets are classified into targets in a relatively stationary state, targets in a straight-line motion state, and targets in a turning state. Extract the motion trajectories of the multiple targets throughout the entire motion process; The states of the multiple targets during the overall motion are judged to determine the motion state of each target; Based on the target motion trajectory curve, the motion state trajectory of each target is fitted sequentially, wherein the target trajectory in the straight motion state is fitted with a straight line and the target trajectory in the turning state is fitted with an arc. Based on the motion state trajectory of each target, multi-target characteristic analysis is performed to determine the overall motion state of the multi-target group. For different overall motion states of multi-target groups, corresponding individual motion trajectory statistical analysis methods are adopted to statistically analyze the individual motion trajectories of the targets; as well as The target characteristics are identified based on the statistical results.

2. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: The method of acquiring motion images of multiple targets in an optoelectronic image sequence via optoelectronic devices includes acquiring single-frame images of the multiple targets during their motion in an optoelectronic image sequence with a time interval of 0.1 seconds, wherein the multiple targets in the optoelectronic image contain at least three targets.

3. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: The step of classifying the targets in the multi-target group into targets in a relatively stationary state, targets in a straight-line motion state, and targets in a turning state based on a preset target motion state judgment threshold specifically includes: Obtain the motion trajectory of the target itself among the multiple targets; Calculate the average velocity of the target trajectory, obtain the ordinate of the target trajectory in each frame, and in Frames and The vertical coordinates of the frames are denoted as follows: and The average velocity of the target is then calculated as follows: ; The average speed is compared with the preset target motion state judgment threshold. If the average speed is less than the preset target motion state judgment threshold, the target is in a relatively stationary state. If the average speed is greater than the preset target motion state judgment threshold but less than the maximum speed threshold, the target is in a straight line motion state. If the average speed is greater than the maximum speed threshold, the target is in a turning state.

4. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: The extraction of the motion trajectory of the multiple targets throughout the entire motion process specifically includes: Obtain the motion trajectory of each target in the multi-target group; A coordinate system is established with the target's position in the first frame as the initial point and the target's initial position coordinates as the origin. Obtain the trajectory curve of the target's motion and establish the target's coordinates in frame t. and the coordinate values ​​in that frame As a function of time t Store; The coordinate values ​​of the target in each frame are obtained to form a complete trajectory coordinate sequence.

5. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: The process of performing multi-target characteristic analysis based on the motion trajectory of each target to determine the overall motion state of the multiple targets specifically includes: The number of targets in linear motion among the multiple targets is counted and denoted as . The total number of targets in the multi-target group is denoted as . ; when equal When the multi-target motion is determined to be linear, the multi-target motion is determined to be linear. when Less than At that time, the multiple targets are determined to be turning.

6. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: The process of fitting a straight line to the target trajectory in a linear motion state uses the least squares method, and the fitted straight line equation is denoted as follows: ,in, The intercept is... The slope is used; the target trajectory in the turning state is fitted with an arc using the least squares method, and the fitted arc equation is denoted as... ,in, The intercept is... The coefficient of the parabola, This is the radial coefficient.

7. The multi-target feature recognition method based on photoelectric images according to claim 1, characterized in that: For different overall multi-target motion states, corresponding individual motion trajectory statistical analysis methods are used to statistically analyze the motion trajectories of individual targets, specifically including: For multiple targets whose overall movement is in a straight line, the straight line where the target is located is used as the criterion, and the individual movement trajectories of the target are statistically analyzed based on the movement trajectories of the multiple targets. For multiple targets whose overall motion is in an arc, the arc trajectory of the target is used as the criterion, and the individual motion trajectories of the targets are statistically analyzed based on the motion trajectories of the multiple targets. For multiple targets whose overall movement is in a turning state, the angle corresponding to the target's position relative to its initial position on the movement trajectory is used as the judgment criterion. Based on the movement trajectory of the multiple targets, the individual movement trajectories of the targets are statistically analyzed.

8. The multi-target feature recognition method based on photoelectric images according to claim 7, characterized in that: For multiple targets whose overall motion is linear, the straight line containing the target is used as the criterion. Based on the motion trajectory of the multiple targets, the individual motion trajectories of the targets are statistically analyzed, specifically including: Let the trajectory of the i-th target in the multi-target group be... The trajectory of the j-th target is ; The parameters of the i target trajectory lines are set as follows: , , , No. The parameters of the target trajectory straight line are: , , Among them, let ; Calculate the intersection point of the two straight lines; if an intersection point exists, then the two motion trajectories intersect. The trajectory that intersects with all the target trajectories is determined as the main trajectory of the multi-target linear motion.

9. The multi-target feature recognition method based on photoelectric images according to claim 7, characterized in that: For multiple targets whose overall motion is in an arc, the arc trajectory of the target is used as the criterion. Based on the motion trajectory of the multiple targets, the individual motion trajectories of the targets are statistically analyzed, specifically including: The vertical coordinate of the trajectory of the object in the multi-target object is set as... ; Let the equation of the straight line tangent to the trajectory of the target be: ; When the straight line is tangent to the target trajectory, there exists a unique solution. , making ; The arc trajectories that are tangent to each target trajectory are statistically analyzed and determined as the main trajectory of the multi-target arc motion.

10. The multi-target feature recognition method based on photoelectric images according to claim 7, characterized in that: For multiple targets whose overall movement is in a turning state, the angle corresponding to the target's position relative to its initial position on the trajectory is used as the judgment criterion. Based on the multi-target movement trajectory, the individual movement trajectories of the targets are statistically analyzed, specifically including: The initial and final positions of each target are calculated relative to the main trajectory of the multiple moving targets, and denoted as θ. If the final state is in the same position as the initial state, then θ=0, and the target is in a stationary state. If there are no other pixels on the trajectory between the target positions in two frames except for the start and end points, and θ≠0, then the target is in a turning state; If the pixels on the trajectory of the initial state and the final state do not belong to the same straight line as the pixels on the trajectory of the linear motion state, and θ≠0, then the target is in a turning state.

Images