Motion detection method and device based on feature point analysis

Abstract

The invention relates to the field of motion detection, and particularly discloses a motion detection method and device based on feature point analysis, and the method comprises the following steps: a, extracting feature points from an image through employing an SIFT algorithm; b, extracting pixel information in a neighborhood of each feature point, including gradient, color, local contrast and texture features, so as to enrich description of the feature points; c, fusing feature point information from different sensors or image sources, calculating a fusion weight of each feature point, and carrying out weighted averaging on the feature points based on the weights so as to realize effective fusion of multi-source information; according to the method, feature point information from different sensors or image sources is fused, the fusion weight is calculated and weighted averaging is carried out, multi-source information is effectively integrated, the reliability and comprehensiveness of motion detection are improved, and the fusion strategy can make full use of the advantages of different sensors or image sources and make up for the deficiency of single-source information.

Description

Technical Field

[0001] The invention belongs to the field of motion detection, in particular to a method and a device for motion detection based on feature point analysis. Background Art

[0002] Motion detection based on feature point analysis is an important research direction in the field of computer vision. It detects, tracks and analyzes feature points in images to identify, locate and track moving targets. With the continuous development of computer technology, this method has been widely used in video surveillance, human-computer interaction, autonomous driving and other fields.

[0003] In practical applications, motion detection often involves information from multiple sensors or image sources. However, in the prior art, it is often difficult to integrate multi-source information, resulting in insufficient information utilization. Summary of the invention

[0004] In order to solve the above technical problems, the present invention provides a method for motion detection based on feature point analysis to solve the problem in the prior art that integration of multi-source information is often difficult, resulting in insufficient information utilization.

[0005] A method for motion detection based on feature point analysis comprises the following steps:

[0006] a. Use SIFT algorithm to extract feature points in the image;

[0007] b. For each feature point, extract the pixel information in its neighborhood, including gradient, color, local contrast and texture features, to enrich the description of the feature point;

[0008] c. Fuse feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information;

[0009] d. Based on the fused feature points and their neighborhood information, a matching algorithm is used to determine the correspondence between feature points in consecutive frames;

[0010] e. Based on the feature point matching, combined with the block matching algorithm, motion estimation is performed using the pixel information in the neighborhood of the feature point;

[0011] f. Use the Kalman filter algorithm to track moving targets in real time and update the target state based on the observed and predicted values ​​to improve the tracking stability and accuracy;

[0012] g. According to the motion vector, the moving target is tracked precisely, including target identification, trajectory drawing and speed measurement;

[0013] h. Objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking;

[0014] i. Post-process the motion estimation results to improve the accuracy and completeness of motion detection.

[0015] Preferably, in step b, the extraction of neighborhood information of the feature point further includes calculating the local contrast of pixels in the neighborhood to further enhance the stability and distinguishability of the feature point.

[0016] Preferably, in step c, the fusion weight calculation formula of the feature points is as follows:

[0017]

[0018] Among them, ω i is the fusion weight of the i-th sensor;

[0019] d i is the distance between the target detected by the i-th sensor and the sensor;

[0020] μ is the average of the target distances detected by all sensors;

[0021] σ is the standard deviation of the target distances detected by all sensors;

[0022] SNR i is the signal-to-noise ratio of the i-th sensor;

[0023] n is the total number of sensors;

[0024] The target position calculation formula after multi-source information fusion is as follows:

[0025]

[0026] Ρ is the estimated target position after fusion;

[0027] ω i is the fusion weight of the i-th sensor;

[0028] P i is the target position detected by the i-th sensor.

[0029] Preferably, in step d, the feature point matching algorithm adopts the RANSAC algorithm, which estimates the inner point that best fits the model from the matching point pairs in an iterative manner to improve the accuracy and robustness of the matching and reduce the possibility of mismatching.

[0030] Preferably, in step e, the selection and parameter setting of the block matching algorithm are adaptively adjusted according to the application scenario and image characteristics to achieve more precise tracking of the moving target and more accurate motion estimation.

[0031] Preferably, in step f, the update formula of the Kalman filter algorithm is as follows:

[0032] x k|k =X k|k-1 +K k (Z k -H k X k|k-1 )

[0033]

[0034] mathbf k|k =(Ι-K k Η k ) k|k-1 ;

[0035] Among them, X k|k is the state estimate at time k;

[0036] X k|k-1 Predict the state at time k for time k-1;

[0037] K k is the Kalman gain matrix;

[0038] Z k is the observed value at time k;

[0039] Η k is the matrix that maps the state vector to the observation vector;

[0040] R k is the covariance matrix of the observation noise;

[0041] P k|k is the covariance matrix of the estimation error at time k;

[0042] P k|k-1 The error covariance matrix for predicting moment k at moment k-1;

[0043] Ι is the identity matrix.

[0044] Preferably, in step h, the accuracy evaluation formula for moving target tracking is as follows:

[0045]

[0046] Among them, Accuracy k is the tracking accuracy at time k;

[0047] X k|k is the state estimate at time k;

[0048] X trueis the true state vector of the target.

[0049] Preferably, the method further includes performing trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in step g, so as to extract useful behavior information.

[0050] A motion detection device based on feature point analysis, comprising:

[0051] A feature point extraction module is used to extract feature points in an image using a SIFT algorithm;

[0052] The feature description module is used to extract pixel information in the neighborhood of each feature point, including gradient, color, local contrast and texture features, to enrich the description of the feature point;

[0053] The information fusion module is used to fuse the feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information;

[0054] A feature point matching module is used to determine the correspondence between feature points of consecutive frames using a matching algorithm based on the fused feature points and their neighborhood information;

[0055] The motion estimation module is used to perform motion estimation based on feature point matching and in combination with a block matching algorithm, using pixel information within the neighborhood of a feature point;

[0056] The real-time tracking module is used to track the moving target in real time using the Kalman filter algorithm and update the target state according to the observed value and predicted value to improve the tracking stability and accuracy;

[0057] A fine tracking module is used to perform fine tracking of moving targets based on motion vectors, including target recognition, trajectory drawing and speed measurement;

[0058] The accuracy evaluation module is used to objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking;

[0059] A post-processing module, used for post-processing the motion estimation result to improve the accuracy and completeness of motion detection;

[0060] The behavior analysis module is used to perform trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in the fine tracking module to extract useful behavior information.

[0061] Compared with the prior art, the present invention has the following beneficial effects:

[0062] By fusing feature point information from different sensors or image sources, calculating fusion weights and performing weighted averaging, multi-source information is effectively integrated, improving the reliability and comprehensiveness of motion detection. This fusion strategy can fully utilize the advantages of different sensors or image sources and make up for the shortcomings of single-source information.

[0063] Based on feature point matching, the block matching algorithm is combined for motion estimation, and the Kalman filter algorithm is used to track the moving target in real time. This combination strategy not only improves the tracking stability, but also significantly improves the tracking accuracy, making the trajectory of the moving target more accurate and continuous. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 It is a schematic diagram of the method flow of the present invention;

[0065] Figure 2 It is a diagram of the algorithm execution steps of the present invention. DETAILED DESCRIPTION

[0066] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0067] like Figure 1 to Figure 2 As shown:

[0068] Embodiment 1: The present invention provides a method for motion detection based on feature point analysis, comprising the following steps:

[0069] a. Use SIFT algorithm to extract feature points in the image;

[0070] b. For each feature point, extract the pixel information in its neighborhood, including gradient, color, local contrast and texture features, to enrich the description of the feature point;

[0071] c. Fuse feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information;

[0072] d. Based on the fused feature points and their neighborhood information, a matching algorithm is used to determine the correspondence between feature points in consecutive frames;

[0073] e. Based on the feature point matching, combined with the block matching algorithm, motion estimation is performed using the pixel information in the neighborhood of the feature point;

[0074] f. Use the Kalman filter algorithm to track moving targets in real time and update the target state based on the observed and predicted values ​​to improve the tracking stability and accuracy;

[0075] g. According to the motion vector, the moving target is tracked precisely, including target identification, trajectory drawing and speed measurement;

[0076] h. Objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking;

[0077] i. Post-processing the motion estimation results to improve the accuracy and completeness of motion detection;

[0078] Among them, this method first uses the SIFT algorithm to extract feature points from the image, and extracts the neighborhood pixel information of each feature point in detail, such as gradient, color, local contrast and texture features, to enhance the descriptiveness of the feature points; then, the feature point information from different sensors or image sources is cleverly fused, and the effective integration of multi-source information is achieved by calculating the fusion weight and performing weighted averaging; on the basis of feature point matching, the block matching algorithm is combined for motion estimation, and the Kalman filter algorithm is used to track the moving target in real time, thereby improving the tracking stability and accuracy; finally, the moving target is finely tracked according to the motion vector, including target recognition, trajectory drawing and speed measurement, and the tracking results are objectively evaluated through the accuracy evaluation formula; in addition, it also includes post-processing of the motion estimation results to further improve the accuracy and completeness of motion detection.

[0079] Specifically, in step b, the extraction of neighborhood information of the feature point also includes calculating the local contrast of pixels in the neighborhood to further enhance the stability and distinguishability of the feature point.

[0080] Specifically, in step c, the fusion weight calculation formula of the feature points is as follows:

[0081]

[0082] Among them, ω i is the fusion weight of the i-th sensor;

[0083] d i is the distance between the target detected by the i-th sensor and the sensor;

[0084] μ is the average of the target distances detected by all sensors;

[0085] σ is the standard deviation of the target distances detected by all sensors;

[0086] SNR i is the signal-to-noise ratio of the i-th sensor;

[0087] n is the total number of sensors;

[0088] The target position calculation formula after multi-source information fusion is as follows:

[0089]

[0090] Ρ is the estimated target position after fusion;

[0091] ω i is the fusion weight of the i-th sensor;

[0092] P i is the target position detected by the i-th sensor.

[0093] As can be seen from the above, this method not only enriches the information of the feature points by calculating the local contrast of pixels in the neighborhood of the feature points, but also improves the stability and discrimination of the feature points, which is crucial for subsequent feature point matching and motion estimation; in step c, a scientific feature point fusion weight calculation formula is adopted, which comprehensively considers factors such as the distance between the sensor and the target, the average and standard deviation of the target distance detected by all sensors, and the signal-to-noise ratio of the sensor, so that the fusion weight of each feature point can be calculated more accurately; based on these weights, the feature points are weighted averaged to achieve effective fusion of multi-source information and improve the estimation accuracy of the target position; therefore, the motion detection method of the present invention is optimized and innovated in feature point description and fusion, providing a more accurate and stable basis for subsequent matching, estimation and tracking steps.

[0094] Embodiment 2: This embodiment is basically the same as the previous embodiment, except that, in step d, the feature point matching algorithm adopts the RANSAC algorithm, which estimates the inner point that best fits the model from the matching point pair in an iterative manner to improve the matching accuracy and robustness and reduce the possibility of mismatching;

[0095] Among them, the specific application process of the RANSAC algorithm is as follows:

[0096] Randomly select subset: Randomly select a subset of points from all matching point pairs as a subset for fitting a model (such as a homography matrix or an essential matrix).

[0097] Model fitting: Use the selected subset to fit a model that describes the relationship between the matching point pairs.

[0098] Inlier point judgment: For the remaining points, their distance from the fitted model is calculated. Points with a distance less than a certain threshold are considered inliers (i.e., points that conform to the model), otherwise they are considered outliers (i.e., points that do not conform to the model).

[0099] Iteration and selection: Repeat the above process multiple times (i.e. iteration), selecting a different subset each time for model fitting and inlier determination. Finally, the model with the largest number of inliers is selected as the final fitted model.

[0100] Through the RANSAC algorithm, the inliers that best fit the model can be estimated from matching point pairs with a large number of mismatches, thereby improving the accuracy and robustness of the matching and reducing the possibility of mismatches.

[0101] Specifically, in step e, the selection and parameter setting of the block matching algorithm are adaptively adjusted according to the application scenario and image characteristics to achieve more precise tracking of the moving target and more accurate motion estimation.

[0102] Specifically, in step f, the update formula of the Kalman filter algorithm is as follows:

[0103] x kk =X kk-1 +K k (Z k -H k X kk-1 )

[0104]

[0105] mathbf kk =(Ι-K k Η k ) kk-1 ;

[0106] Among them, X kk is the state estimate at time k;

[0107] X kk-1 Predict the state at time k for time k-1;

[0108] K k is the Kalman gain matrix;

[0109] Z k is the observed value at time k;

[0110] Η k is the matrix that maps the state vector to the observation vector;

[0111] R k is the covariance matrix of the observation noise;

[0112] P kk is the covariance matrix of the estimation error at time k;

[0113] P kk-1 The error covariance matrix for predicting moment k at moment k-1;

[0114] Ι is the identity matrix.

[0115] Specifically, in step h, the accuracy evaluation formula for moving target tracking is as follows:

[0116]

[0117] Among them, Accuracy k is the tracking accuracy at time k;

[0118] X kk is the state estimate at time k;

[0119] X true is the true state vector of the target.

[0120] Specifically, the method also includes performing trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in step g, so as to extract useful behavior information.

[0121] As can be seen from the above, this embodiment adopts the RANSAC algorithm, and selects the inner point that best fits the model from a large number of matching point pairs in an iterative manner, which significantly improves the accuracy and robustness of the matching and effectively reduces the mismatch rate; in the selection of the block matching algorithm, adaptive adjustment is implemented, and the algorithm is flexibly selected and the parameters are set according to the actual application scenario and image characteristics, thereby achieving more precise tracking of the moving target and more accurate motion estimation; in the application of the Kalman filter algorithm, the algorithm update formula is given in detail, and by comprehensively considering the observed value and the predicted value, the real-time update of the moving target state is achieved, and the tracking stability and accuracy are improved; in addition, the accuracy evaluation formula for moving target tracking is introduced, and the accuracy of the tracking result is objectively evaluated by comparing the state estimate with the true state vector; finally, steps such as trajectory analysis, speed change analysis, acceleration analysis and target interaction analysis are added to extract useful behavior information, which provides strong support for the in-depth understanding and analysis of the moving target.

[0122] Embodiment 3: A motion detection device based on feature point analysis, comprising:

[0123] A feature point extraction module is used to extract feature points in an image using a SIFT algorithm;

[0124] The feature description module is used to extract pixel information in the neighborhood of each feature point, including gradient, color, local contrast and texture features, to enrich the description of the feature point;

[0125] The information fusion module is used to fuse the feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information;

[0126] A feature point matching module is used to determine the correspondence between feature points of consecutive frames using a matching algorithm based on the fused feature points and their neighborhood information;

[0127] The motion estimation module is used to perform motion estimation based on feature point matching and in combination with a block matching algorithm, using pixel information within the neighborhood of a feature point;

[0128] The real-time tracking module is used to track the moving target in real time using the Kalman filter algorithm and update the target state according to the observed value and predicted value to improve the tracking stability and accuracy;

[0129] A fine tracking module is used to perform fine tracking of moving targets based on motion vectors, including target recognition, trajectory drawing and speed measurement;

[0130] The accuracy evaluation module is used to objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking;

[0131] A post-processing module, used for post-processing the motion estimation result to improve the accuracy and completeness of motion detection;

[0132] The behavior analysis module is used to perform trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in the fine tracking module to extract useful behavior information.

[0133] As can be seen from the above, this embodiment introduces a motion detection device based on feature point analysis, which is composed of multiple modules, each of which has a specific function and works together to achieve efficient and accurate motion detection; the feature point extraction module uses the SIFT algorithm to extract feature points from the image to provide a basis for subsequent processing, and the feature description module describes each feature point in detail, including information such as gradient, color, local contrast and texture features in its neighborhood to enrich the representation of the feature point; the information fusion module can fuse feature point information from different sensors or image sources, and calculate the fusion weight of each feature point and perform weighted averaging to achieve efficient and accurate motion detection. The effective integration of multi-source information is realized. The feature point matching module uses the matching algorithm to determine the correspondence between feature points between consecutive frames based on the fused feature points and their neighborhood information. The motion estimation module uses the block matching algorithm to perform motion estimation on the pixel information in the neighborhood of the feature points on the basis of feature point matching. The real-time tracking module uses the Kalman filter algorithm to track the moving target in real time to improve the stability and accuracy of tracking. The fine tracking module performs fine tracking of the moving target based on the motion vector, including target recognition, trajectory drawing and speed measurement, etc. The accuracy evaluation module objectively evaluates the tracking results and verifies the accuracy of tracking. The post-processing module further optimizes the motion estimation results to improve the accuracy and completeness of motion detection. Finally, the behavior analysis module analyzes the trajectory, speed, acceleration and interaction between moving targets based on the data of the fine tracking module to extract useful behavior information.

[0134] The standard parts used in the present invention can all be purchased from the market, and the special-shaped parts can be customized according to the description and the drawings. The specific connection methods of each part adopt conventional means such as mature bolts, rivets, welding, etc. in the prior art. The machinery, parts and equipment all adopt conventional models in the prior art, and the circuit connection adopts the conventional connection method in the prior art, which will not be described in detail here. The contents not described in detail in this specification belong to the prior art known to professional and technical personnel in this field.

[0135] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. "Multiple" means two or more, unless otherwise clearly and specifically defined.

[0136] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0137] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0138] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0139] In the drawings of the embodiments disclosed in the present invention, only the structures related to the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0140] Although the present invention has been described in detail with reference to the aforementioned embodiments, it is still possible for those skilled in the art to modify the technical solutions described in the aforementioned embodiments, or to make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for motion detection based on feature point analysis, characterized in that: The following steps are involved: a. Use SIFT algorithm to extract feature points in the image; b. For each feature point, extract the pixel information in its neighborhood, including gradient, color, local contrast and texture features, to enrich the description of the feature point; c. Fuse feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information; d. Based on the fused feature points and their neighborhood information, a matching algorithm is used to determine the correspondence between feature points in consecutive frames; e. Based on the feature point matching, combined with the block matching algorithm, motion estimation is performed using the pixel information in the neighborhood of the feature point; f. Use the Kalman filter algorithm to track moving targets in real time and update the target state based on the observed and predicted values ​​to improve the tracking stability and accuracy; g. According to the motion vector, the moving target is tracked precisely, including target identification, trajectory drawing and speed measurement; h. Objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking; i. Post-process the motion estimation results to improve the accuracy and completeness of motion detection.

2. A method for motion detection based on feature point analysis as claimed in claim 1, characterized in that: In step b, the extraction of neighborhood information of the feature point also includes calculating the local contrast of pixels in the neighborhood to further enhance the stability and distinguishability of the feature point.

3. A method for motion detection based on feature point analysis as claimed in claim 2, characterized in that: In step c, the fusion weight calculation formula of the feature points is as follows: Among them, ω i is the fusion weight of the i-th sensor; d i is the distance between the target detected by the i-th sensor and the sensor; μ is the average of the target distances detected by all sensors; σ is the standard deviation of the target distances detected by all sensors; SNR i is the signal-to-noise ratio of the i-th sensor; n is the total number of sensors; The target position calculation formula after multi-source information fusion is as follows: Ρ is the estimated target position after fusion; ω i is the fusion weight of the i-th sensor; P i is the target position detected by the i-th sensor.

4. A method for motion detection based on feature point analysis as claimed in claim 3, characterized in that: In step d, the feature point matching algorithm adopts the RANSAC algorithm, which estimates the inner point that best fits the model from the matching point pairs in an iterative manner to improve the accuracy and robustness of the matching and reduce the possibility of mismatching.

5. A method for motion detection based on feature point analysis as claimed in claim 4, characterized in that: In step e, the selection and parameter setting of the block matching algorithm are adaptively adjusted according to the application scenario and image characteristics to achieve more precise tracking of the moving target and more accurate motion estimation.

6. A method for motion detection based on feature point analysis as claimed in claim 5, characterized in that: In step f, the update formula of the Kalman filter algorithm is as follows: mathbfX k|k =X k|k-1 +K k (Z k -Η k X k|k-1 ) mathbfP k|k =(I-K) k OR k )P k|k-1 ; Among them, X k|k is the state estimate at time k; X k|k-1 Predict the state at time k for time k-1; K k is the Kalman gain matrix; Z k is the observed value at time k; Η k is the matrix that maps the state vector to the observation vector; R k is the covariance matrix of the observation noise; P k|k is the covariance matrix of the estimation error at time k; P k|k-1 The error covariance matrix for predicting moment k at moment k-1; Ι is the identity matrix.

7. A method for motion detection based on feature point analysis as claimed in claim 6, characterized in that: In step h, the accuracy evaluation formula for moving target tracking is as follows: Among them, Accuracy k is the tracking accuracy at time k; X k|k is the state estimate at time k; X true is the true state vector of the target.

8. The method for motion detection based on feature point analysis as claimed in claim 1, characterized in that: It also includes performing trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in step g, so as to extract useful behavior information.

9. A motion detection device based on feature point analysis, characterized in that: include: A feature point extraction module is used to extract feature points in an image using a SIFT algorithm; The feature description module is used to extract pixel information in the neighborhood of each feature point, including gradient, color, local contrast and texture features, to enrich the description of the feature point; The information fusion module is used to fuse the feature point information from different sensors or image sources, calculate the fusion weight of each feature point, and perform weighted averaging of the feature points based on these weights to achieve effective fusion of multi-source information; A feature point matching module is used to determine the correspondence between feature points of consecutive frames using a matching algorithm based on the fused feature points and their neighborhood information; The motion estimation module is used to perform motion estimation based on feature point matching and in combination with a block matching algorithm, using pixel information within the neighborhood of a feature point; The real-time tracking module is used to track the moving target in real time using the Kalman filter algorithm and update the target state according to the observed value and predicted value to improve the tracking stability and accuracy; A fine tracking module is used to perform fine tracking of moving targets based on motion vectors, including target recognition, trajectory drawing and speed measurement; The accuracy evaluation module is used to objectively evaluate the tracking results through the accuracy evaluation formula to verify the accuracy of tracking; A post-processing module, used for post-processing the motion estimation result to improve the accuracy and completeness of motion detection; The behavior analysis module is used to perform trajectory analysis, speed change analysis, acceleration analysis and interaction analysis between targets on the moving target based on the trajectory and speed measurement data obtained in the fine tracking module to extract useful behavior information.

Images