CN116363412A - Method and system for constructing classification model of flying bird and unmanned aerial vehicle based on centroid movement characteristics - Google Patents

Abstract

The invention discloses a method and a system for constructing a classification model of a flying bird and an unmanned aerial vehicle based on centroid movement characteristics, and video data of flying of the flying bird and the unmanned aerial vehicle are shot; preprocessing video data, and then cutting the preprocessed video data into a plurality of tracks with fixed duration to form a video data set; marking a centroid movement track by adopting a computer vision algorithm, and performing wavelet denoising pretreatment; calculating the speed and acceleration value motion characteristics of the track fragments in the centroid motion track, and extracting the maximum fluctuation amplitude of the track fragments in different frequency intervals; respectively establishing a speed classifier, an acceleration classifier and a track fluctuation classifier aiming at the speed, acceleration and track fluctuation characteristics of the flying bird and the unmanned aerial vehicle by adopting machine learning; weighting and integrating the prediction results of the three classifiers, obtaining the optimal weight in a five-fold cross validation mode, and establishing a classification model of the flying bird and the unmanned aerial vehicle; the classification model of the flying bird and the unmanned aerial vehicle provided by the invention has the advantages of high speed and high accuracy, does not need manual judgment, and liberates labor cost.

Description

A method for constructing a classification model of flying birds and UAVs based on the motion characteristics of the center of mass and its system

technical field

The invention relates to the technical field of object classification, in particular to a method and system for constructing a classification model of flying birds and drones based on the motion characteristics of the center of mass, and establishes a real-time object classifier.

Background technique

Flying birds and UAVs are typical "low, small and slow" aircraft, that is, aerial vehicles with a flying altitude of less than 1,000 meters, a flying speed of less than 200 km/h, and a radar reflection area of less than 2 square meters. They are the main factors threatening the safety of low-altitude airspace. Flying birds will damage the structure of the aircraft when they collide with the aircraft, and it will seriously cause the aircraft to crash and kill people. Bird strike accidents are a major incident factor in aircraft operations. In addition, with the development of drone technology, drones are gradually becoming a new type of threat. Due to its small size, easy to carry and easy to operate, it is easy to be used in illegal activities such as illegal ground surveying and interference with normal civil aviation flights. Different threats to birds and UAVs require different measures, and the prerequisite for taking appropriate measures is to correctly identify and classify flying objects. When the distance is long, birds and UAVs are small in size and have similar appearance characteristics, so it is difficult to distinguish them accurately.

Since 2017, the SafeShore project funded by the European Union's "Horizon 2020" program has launched a bird and drone detection challenge. This competition is held every two years and has been held three times so far. In this competition, most of the algorithms are mainly based on static image information, that is, based on the differences in the appearance characteristics of flying birds and drones, to establish a convolutional neural network model. However, these algorithms are less effective in identifying long-distance small targets. When the target image size is less than 32 pixels, the false positive rate and false negative rate are relatively high.

Some researchers propose to classify objects based on their motion information. Compared with image information, the advantage of motion information is that it is highly robust to distance and is not affected by object deformation, ambient brightness, and background distractors. Srigrarom et al. calculated five motion features of each trajectory, including average velocity, average acceleration, turning angle, periodicity, and radius of curvature, and used principal component analysis to reduce dimensionality. A support vector machine was established based on the first two main features, and the accuracy rate was greater than 80%, but the disadvantage of this algorithm is that it is established based on the statistical indicators of motion information, and the statistics can only reflect part of the information, which is easily affected by the length of the trajectory segment and cannot be used for real-time classification tasks.

Contents of the invention

In order to solve the above-mentioned existing technical problems, the present invention provides a method and system for constructing a classification model of flying birds and UAVs based on the motion characteristics of the center of mass. The strategy adopted when classifying moving targets, establishes a real-time flying target classifier.

The present invention adopts following technical scheme:

On the one hand, the present invention provides a kind of method for building a classification model of flying birds and unmanned aerial vehicles based on the motion characteristics of the center of mass, which is characterized in that the building method includes the following steps:

Step 1, shoot the flying birds, and at the same time control the UAV to simulate the flight trajectory of the flying birds, and obtain the video data of the flying birds and the UAV flight respectively;

Step 2, preprocessing the video data and cutting it into a number of tracks with a fixed duration to form a video data set;

Step 3, use computer vision algorithm to mark the position of the center of mass of the flying target in each frame of the video data set, form the trajectory of the center of mass, and perform wavelet denoising preprocessing;

Step 4. Divide each trajectory segment into several fixed-duration trajectory segments without overlap, calculate the velocity, acceleration and trajectory fluctuation motion characteristics of the trajectory segment, obtain all velocity values and acceleration values, and extract the trajectory segment in different frequency intervals. Maximum volatility;

Step 5, using machine learning to establish speed classifiers, acceleration classifiers and trajectory fluctuation classifiers for the speed, acceleration and trajectory fluctuation characteristics of birds and drones;

Step 6, use the weighted integration method to weight the prediction results of the three classifiers obtained in step 5, use the cross-validation method to obtain the optimal weight, establish a motion trajectory classification model, and test the birds and/or UAV videos in the test set The data set is identified.

Further, the method also includes step 7, the motion trajectory classification model calculates the trajectory fragments in the trajectory to obtain the category probability of the trajectory fragments; adopts the stability evaluation module based on the working memory mechanism and decision-making mechanism of the human brain to analyze the trajectory fragments The classification results are output after the stability evaluation of the category probability.

Further, in the step 7, the motion features in the trajectory segment are input into the motion trajectory classification model built in step 6 for calculation to obtain the category probability of the trajectory segment; the motion trajectory classification model continues to input the category probability to the stability assessment module Information; when there are at least 7 class probabilities greater than 0.8 among 10 consecutive input probabilities of the same class, and the class judgments of the speed classifier, acceleration classifier and trajectory fluctuation classifier are consistent, the classification result is output.

Preferably, the constructed motion trajectory classification model performs category probability calculation on trajectory segments containing 24 frames with a duration of 200 ms.

Preferably, in step 2, when the original video data is preprocessed, the motion features of flying birds and drones in the original video data during take-off, landing and hovering are removed to form trajectories of at least 2 seconds each.

Preferably, in the step 3, the ECO target tracking algorithm is used to mark the centroid position of each frame of the flying target in the collected pigeon video data set, and the DIMP target tracking algorithm is used to collect the topaz bird, pearl bird and The position of the center of mass of each frame of the flying target in the UAV video dataset is marked.

Preferably, in the step 4, each segment of the track is divided into several track segments with a duration of 200 ms without overlapping, and the method for calculating the trajectory fluctuation motion characteristics in the track segment is: do a moving average on the centroid motion track in step 3 After smoothing, the smooth trajectory is obtained; the horizontal and vertical coordinates of the smooth trajectory are subtracted from the horizontal and vertical coordinates of the centroid motion trajectory, respectively, to obtain the trajectory fluctuations in the horizontal and vertical directions; the spectrum of the trajectory fluctuations is obtained by fast Fourier transform Analysis, respectively, in the frequency range 0 ~ 10Hz, 10 ~ 20Hz and 20 ~ 30Hz in the maximum fluctuation range.

Preferably, in the step 5, a velocity classifier and an acceleration classifier are respectively established using a naive Bayesian algorithm, and a trajectory fluctuation classifier is established using a support vector machine algorithm.

On the other hand, the present invention also provides a kind of bird and unmanned aerial vehicle classification model system based on the motion characteristic of center of mass, and described system comprises: camera, is used for taking the video data of flying bird and unmanned aerial vehicle;

The preprocessing module is used to preprocess the video data and cut it into several fixed-time tracks to form a video data set;

The visual algorithm labeling module is used to label the position of the center of mass of the flight target in each frame of the preprocessed bird and UAV trajectory to form the center of mass movement track of the flight target;

The wavelet denoising module is used to perform wavelet denoising preprocessing on the centroid trajectory to remove high-frequency noise and labeling errors in the centroid trajectory;

A data processing module, which divides each trajectory into several fixed duration trajectory segments without overlap, calculates the velocity, acceleration and trajectory fluctuation motion characteristics of the trajectory segments in each trajectory, obtains all velocity values and acceleration values, and extracts the trajectory The maximum fluctuation amplitude of the segment in different frequency intervals;

The classifier construction module extracts the velocity value, the acceleration value and the trajectory fluctuation amplitude in the data processing module, and adopts machine learning to construct a velocity classifier, an acceleration classifier and a trajectory fluctuation classifier respectively;

The classification model construction module uses the weighted integration method to integrate the obtained prediction results of the speed classifier, acceleration classifier and trajectory fluctuation classifier, and uses the cross-validation method to obtain the optimal weight to obtain the motion trajectory classification model.

Further, the system also includes a stability evaluation module, the motion trajectory classification model calculates the category probability of the trajectory segments in each trajectory, and outputs the classification result after the probability evaluation is performed by the stability evaluation module.

The technical solution of the present invention has the following advantages:

A. The flying bird and unmanned aerial vehicle classification model provided by the present invention, its speed is fast, and accuracy rate is high. Based on the vision system, the present invention establishes a motion trajectory classification model based on short-term trajectory segments, which can realize automatic and real-time judgment of the category of moving objects, without manual judgment, and liberates labor costs; The classification accuracy of motion trajectory is high, reaching more than 90%, and the framework is simple, which can be transferred to other tasks of classifying objects based on motion features.

B. The amount of samples required for the training of the present invention is relatively small. The image features of existing flying targets are affected by factors such as target category, viewing angle, deformation and occlusion, and there are many types. Establishing a classifier based on image features requires a large number of samples for training. However, the motion characteristics are limited by flight dynamics. There are essential differences in the motion characteristics of flying birds and drones, and this difference is not affected by the type of flying birds and the model of the drone. Therefore, the present invention can obtain Stable results, applicable to the identification of different types of birds and drones.

Description of drawings

In order to illustrate the specific implementation of the present invention more clearly, the accompanying drawings used in the specific implementation will be briefly introduced below. Obviously, the accompanying drawings in the following description are some implementations of the present invention, which are common to those skilled in the art. As far as the skilled person is concerned, other drawings can also be obtained based on these drawings on the premise of not paying creative work.

Fig. 1 is the flow chart I of the construction method of motion trajectory classification model provided by the present invention;

Fig. 2 is flow chart II of the construction method of motion trajectory classification model provided by the present invention;

Fig. 3 is the application flow diagram of motion trajectory classification model provided by the present invention;

Fig. 4 is an example diagram of the flight trajectory provided by the present invention;

Fig. 5 is the flow chart of bird and unmanned aerial vehicle classifier provided by the present invention;

Fig. 6 is a composition diagram of the construction system of the motion track classification model provided by the present invention;

Fig. 7 is the flying bird test track classification example-pearl bird provided by the present invention;

Fig. 8 is an example of UAV test track classification provided by the present invention - Tello.

Detailed ways

The technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Apparently, the described embodiments are part of the embodiments of the present invention, but not all of them. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without making creative efforts fall within the protection scope of the present invention.

As shown in Fig. 1, Fig. 3 and Fig. 5, the present invention provides a kind of flying bird and UAV classification model construction method based on centroid motion feature, and the method adopted is as follows:

【S01】The camera is used to shoot the flying birds, and at the same time, the UAV is controlled to simulate the flight trajectory of the flying birds, and the video data of the flying birds and the UAV flight are respectively obtained.

In order to improve the scope of application of the model, the present invention selects 3 types of flying birds and unmanned aerial vehicles respectively. The flying birds are pearl birds, topaz birds and pigeons from small to large; the unmanned aerial vehicles are Tello (Tello) from small to large. DJI Royal Pro (MavicPro) and DJI Phantom 4 (Phantom4). Collect flight video data of 3 types of flying birds indoors. The shortest video duration is 2 seconds. Then choose a calm and sunny day. In an open parking lot outdoors, please experienced pilots control 3 types of drones to simulate the movement of flying birds. Trajectory flight, get the flight video data of the UAV.

[S02] Preprocess the captured raw video data and cut it into several fixed-length trajectories, and all trajectories form a video data set.

The preprocessing here includes using video software to cut all the captured video materials into 2 seconds, and during preprocessing, remove the typical movement characteristics of flying birds and drones, such as take-off, landing and hovering, etc., such as every The flight-like object consists of 30 segments of flight video data of 2 seconds each to form a video data set. There is no specific limitation on the duration and number of segments of the flight video data set.

[S03] Use computer vision algorithm to mark the position of the center of mass of the flying target in each frame of the video data set, form the trajectory of the center of mass, and perform wavelet denoising preprocessing.

Computer vision algorithms (including ECO and DIMP) are used to mark the centroid position of each frame of the flying target in the video data set. The specific method is: draw a bounding box around the flying target, and define the center of the bounding box as the centroid, as shown in Figure 4 Show. The position of the centroid object in each frame of the video constitutes the motion trajectory of the centroid. After all annotations are completed, the accuracy of video annotations is manually checked. Among them, the ECO target tracking algorithm has a higher accuracy rate when marking pigeons, and the DIMP target tracking algorithm has a higher accuracy rate when marking other five types of flying targets (Topaz Bird, Pearl Bird, Tello, Mavic Pro and Phantom4).

Then wavelet denoising preprocessing is performed on the center of mass trajectory formed after labeling to remove high-frequency noise and possible labeling errors in the trajectory.

The UAVs used in the present invention are all four-rotor UAVs. The high-frequency rotation of the helical wings will cause high-frequency fluctuations in the trajectory. In addition, jitter (random noise) and labeling errors will also cause abnormal fluctuations in the trajectory. In order to remove these noises and obtain a relatively accurate centroid position, it is necessary to preprocess the original trajectory data. In the present invention, the wavelet denoising method is used to denoise the original signal. The wavelet denoising method can preserve signal features while removing high-frequency noise.

[S04] The present invention divides the video data set into a training set and a test set in a ratio of 4:1. The 2-second trajectories in the training set are segmented without overlap into short-term (200ms) trajectories. Calculate the three motion characteristics of the trajectory segment, including velocity, acceleration and trajectory fluctuation motion characteristics, obtain all velocity values and acceleration values, and extract the maximum fluctuation range of each trajectory in different frequency intervals.

The trajectory fluctuation here refers to the up and down fluctuation of the original centroid motion trajectory compared with the smooth trajectory. The smooth trajectory refers to the trajectory obtained by smoothing the moving average of the original centroid trajectory, and the smoothing window is 15. The calculation method of trajectory fluctuation subtracts the horizontal and vertical coordinates of the smooth trajectory from the horizontal and vertical coordinates of the original centroid trajectory respectively to obtain the fluctuations in the horizontal and vertical directions; Take the maximum fluctuation range in 3 frequency intervals, including 0-10Hz, 10-20Hz and 20-30Hz.

[S05] Use machine learning to establish speed classifiers, acceleration classifiers, and trajectory fluctuation classifiers for the speed, acceleration, and trajectory fluctuation characteristics of birds and UAVs. The present invention preferably adopts the naive Bayesian algorithm in machine learning, and according to all velocities and accelerations obtained in [S04] and their corresponding categories, respectively establishes a speed classifier and an acceleration classifier, and adopts a support vector machine algorithm in machine learning , according to the trajectory fluctuation amplitude extracted in [S04], a trajectory fluctuation classifier is established.

[S06] According to the three classifiers obtained in step [S05], for any 200ms trajectory segment, the velocity classifier calculates the category probability of each velocity value in the trajectory segment, and the mean of these category probabilities is the velocity category probability of the trajectory segment ; The acceleration classifier calculates the category probability of each acceleration value in the trajectory segment, and the average of these category probabilities is the acceleration category probability; the trajectory fluctuation classifier calculates the volatility category probability according to the amplitude characteristics. The prediction results of the three classifiers are weighted and integrated using the weighted integration method, and the optimal weight is obtained by the cross-validation method. Machine video dataset for recognition.

The present invention preferably adopts a five-fold cross-validation method to calculate the optimal weight. The specific method is: divide the training set into 5 subsets on average, use 4 subsets for training in turn, and use the other subset for verification. This process is repeated 5 times, each time a correct rate is calculated, and the average of the five correct rates is used as an estimate of the correct rate corresponding to the weight. The optimal weight is the weight corresponding to the highest correct rate.

In order to apply the established motion trajectory classification model to real-time classification tasks, as shown in Figure 2, the present invention also simulates the human working mechanism in step [S07], constructs a machine observer, and uses it to classify the objects in the test set track. The machine observer as shown in Figure 5 includes two modules. The first module is the constructed motion trajectory classification model (i.e. bird-drone classifier), which simulates the feature detection network in the brain and continuously extracts the trajectory fragments. motion features and assign them different weights. The motion trajectory classification model in the present invention calculates the category probability of trajectory segments (200 ms, 24 frames in total) with a fixed duration, and is tested by inputting trajectory data in real time when applied in the test set. When the input trajectory length is less than 200 milliseconds, no classification is made; when the input trajectory length reaches 200 milliseconds, the 200 millisecond trajectory segment is judged and the category probability is output; every time a new frame of data is input, the latest 200 millisecond trajectory is obtained fragment, outputting new class probabilities. In actual classification tasks, it is not enough to just calculate the class probability, but also requires the help of working memory and decision-making mechanism. Therefore, the present invention simulates the working principle of the human brain, and also constructs a second module, that is, a stability evaluation module. The trajectory classification model continues to input updated probability information to the stability assessment module. When the input probability of the same category is among 10 consecutive ones, at least 7 of them are greater than 0.8, and the category judgments of the three classifiers are consistent at this time, it is regarded as Stable and reliable judgment, output the final classification result; if the category judgments of the three classifiers are inconsistent, the classification result will not be output. If the trajectory classification model does not make a judgment at the end of the trajectory, that is, it does not reach stability, the correct rate is recorded as 0.5, and the reaction time is recorded as the average response time of other trajectories that make judgments.

Input the trajectory data in the test set into the motion trajectory classification model established in [S06] to obtain the predicted category of each trajectory; compare the consistency between the predicted category and the real category to obtain the classification accuracy of the trajectory classification model.

On the other hand, as shown in Figure 6, the present invention also provides a bird and drone classification model system based on the motion characteristics of the center of mass, including a camera and a preprocessing module built in a computer, a visual algorithm labeling module, a wavelet Noise blocks, data processing blocks, classifier building blocks, and classification model building blocks. The camera is used to shoot the video data of flying birds and drones; the preprocessing module is used to preprocess the video data and cut it into several fixed-length trajectories to form a video data set containing several continuous trajectory segments ;The visual algorithm labeling module is used to mark the position of the center of mass of each frame of the flying target in the preprocessed bird and UAV track fragments to form the center of mass movement track of the flying target; the wavelet denoising module is used to perform wavelet analysis on the center of mass movement track Denoising preprocessing, removing high-frequency noise and labeling errors in the center of mass motion trajectory; the data processing module divides each trajectory into several fixed-length trajectory segments without overlap, and calculates the velocity, acceleration and Trajectory fluctuation motion characteristics, obtain all velocity values and acceleration values, and extract the maximum fluctuation amplitude of trajectory segments in different frequency intervals; the classifier building module extracts the velocity values, acceleration values and trajectory fluctuation amplitudes in the data processing module, and uses machine learning Construct the velocity classifier, acceleration classifier and trajectory fluctuation classifier respectively; the classification model building module uses the weighted integration method to weight the prediction results of the obtained velocity classifier, acceleration classifier and trajectory fluctuation classifier, and uses the cross-validation method to obtain Optimal weights to obtain a motion trajectory classification model.

In order to improve the classification accuracy, a stability evaluation module is also set up in the system. The motion trajectory classification model calculates the category probability of the trajectory fragments in each trajectory. After the probability evaluation is performed by the stability evaluation module, the classification result is output.

As shown in Fig. 7 and Fig. 8, the classification model constructed by the present invention is a schematic diagram. According to the difference in speed and acceleration probability distribution of flying bird and UAV, as well as the difference in trajectory fluctuation, the two are distinguished. It can be seen from the figure that compared with drones, flying birds have a larger distribution range of speed and acceleration, and larger trajectory fluctuations.

The left side of the first line in Figure 7 and Figure 8 is the original centroid trajectory, and the right side of the first line is the real-time output result of the classifier probability judgment, where the white vertical line indicates the output moment of the result (the average response time is 380.79ms) , the classification results are output in the response box. The second row is the extracted trajectory features, which are velocity, acceleration, and trajectory fluctuations in the horizontal and vertical directions. Below the speed and acceleration is the probability distribution template of the speed and acceleration of the bird and drone obtained based on the trajectory data in the training set. The speed and acceleration of flying birds are distributed in the probability template of flying birds, and the trajectory of flying birds fluctuates greatly, as shown in Figure 7; the speed and acceleration of UAVs are distributed in the probability template of UAVs, and the trajectory of UAVs fluctuates relatively small, as shown in Figure 8.

After testing 30 segments of trajectories in the test set, the correct rate reaches 100%.

The amount of samples required for the training of the present invention is small, only part of the flight trajectory needs to be collected, and there is no need to calculate all the flight trajectories. The present invention can obtain stable output results by using small sample training, and can be applied to different types of birds and accurate identification of drones.

Apparently, the above-mentioned embodiments are only examples for clear description, rather than limiting the implementation. For those of ordinary skill in the art, other changes or changes in different forms can be made on the basis of the above description. It is not necessary and impossible to exhaustively list all the implementation manners here. However, the obvious changes or changes derived therefrom still fall within the scope of protection of the present invention.

Claims (10)

1. a kind of flying bird and unmanned aerial vehicle classification model building method based on centroid motion feature, it is characterized in that, its building method comprises the steps: Step 1, shoot the flying birds, and at the same time control the UAV to simulate the flight trajectory of the flying birds, and obtain the video data of the flying birds and the UAV flight respectively; Step 2, preprocessing the video data and cutting it into a number of tracks with a fixed duration to form a video data set; Step 3, use computer vision algorithm to mark the position of the center of mass of the flying target in each frame of the video data set, form the trajectory of the center of mass, and perform wavelet denoising preprocessing; Step 4. Divide each trajectory segment into several fixed-duration trajectory segments without overlap, calculate the velocity, acceleration and trajectory fluctuation motion characteristics of the trajectory segment, obtain all velocity values and acceleration values, and extract the trajectory segment in different frequency intervals. Maximum volatility; Step 5, using machine learning to establish speed classifiers, acceleration classifiers and trajectory fluctuation classifiers for the speed, acceleration and trajectory fluctuation characteristics of birds and drones; Step 6, use the weighted integration method to weight the prediction results of the three classifiers obtained in step 5, use the cross-validation method to obtain the optimal weight, establish a motion trajectory classification model, and test the birds and/or UAV videos in the test set The data set is identified. 2. the flying bird and the unmanned aerial vehicle classification model construction method based on the center of mass motion feature according to claim 1, it is characterized in that, described method also comprises step 7, and motion track classification model calculates the track segment in track, obtains The category probability of the trajectory segment; the stability assessment module based on the working memory mechanism and decision-making mechanism of the human brain is used to evaluate the stability of the category probability of the trajectory segment and output the classification result. 3. the flying bird and the unmanned aerial vehicle classification model building method based on the center of mass motion feature according to claim 2, it is characterized in that, in described step 7, the motion track classification that the motion feature input step 6 builds in the trajectory segment Calculate in the model to obtain the category probability of the trajectory segment; the motion trajectory classification model continues to input the category probability information to the stability assessment module; when the same category probability is input, there are at least 7 category probabilities greater than 0.8 in 10 consecutive ones, and When the classification judgments of the speed classifier, acceleration classifier and trajectory fluctuation classifier are consistent, the classification result is output. 4. The method for constructing a bird and unmanned aerial vehicle classification model based on the motion characteristics of the center of mass according to claim 3, wherein the constructed motion trajectory classification model carries out category probability calculations for trajectory segments comprising 24 frames with a duration of 200ms. 5. the method for building a flying bird and unmanned aerial vehicle classification model based on the motion characteristics of the center of mass according to claim 1, characterized in that, in the step 2, when the original video data is preprocessed, remove the flying bird in the original video data And the movement characteristics of the drone when it takes off, lands and hovers, forming a trajectory of at least 2 seconds each. 6. the flying bird and the unmanned aerial vehicle classification model building method based on the center of mass motion feature according to claim 1, it is characterized in that, in described step 3, adopt ECO target tracking algorithm to collect each pigeon video data set The position of the center of mass of the frame flying target is marked, and the DIMP target tracking algorithm is used to mark the position of the center of mass of each frame of the flying target in the collected topaz bird, pearl bird and UAV video data sets. 7. the method for building a flying bird and unmanned aerial vehicle classification model based on the motion characteristics of the center of mass according to claim 1, wherein, in the step 4, each section of the trajectory is divided into several sections of trajectory segments that are 200ms long without overlapping , the method of calculating the trajectory fluctuation motion characteristics in the trajectory segment is: to obtain the smooth trajectory after performing moving mean smoothing on the centroid trajectory in step 3; subtract the horizontal coordinate of the smooth trajectory from the horizontal and vertical coordinates of the centroid trajectory and vertical coordinates to obtain the trajectory fluctuations in the horizontal and vertical directions; use the fast Fourier transform to analyze the frequency spectrum of the trajectory fluctuations, and obtain the maximum fluctuation amplitudes in the frequency intervals of 0-10Hz, 10-20Hz and 20-30Hz respectively. 8. the flying bird and the unmanned aerial vehicle classification model construction method based on the center of mass motion feature according to claim 1, it is characterized in that, in the described step 5, adopt naive Bayesian algorithm to set up speed classifier and acceleration classifier respectively, adopt Support vector machine algorithm is used to establish trajectory fluctuation classifier. 9. A bird and unmanned aerial vehicle classification model system based on the motion characteristics of the center of mass, is characterized in that, said system comprises: video camera, is used to take the video data of flying bird and unmanned aerial vehicle; The preprocessing module is used to preprocess the video data and cut it into several fixed-time tracks to form a video data set; The visual algorithm labeling module is used to label the position of the center of mass of the flight target in each frame of the preprocessed bird and UAV trajectory to form the center of mass movement track of the flight target; The wavelet denoising module is used to perform wavelet denoising preprocessing on the centroid trajectory to remove high-frequency noise and labeling errors in the centroid trajectory; A data processing module, which divides each trajectory into several fixed duration trajectory segments without overlap, calculates the velocity, acceleration and trajectory fluctuation motion characteristics of the trajectory segments in each trajectory, obtains all velocity values and acceleration values, and extracts the trajectory The maximum fluctuation amplitude of the segment in different frequency intervals; The classifier construction module extracts the velocity value, the acceleration value and the trajectory fluctuation amplitude in the data processing module, and adopts machine learning to construct a velocity classifier, an acceleration classifier and a trajectory fluctuation classifier respectively; The classification model construction module uses the weighted integration method to integrate the obtained prediction results of the speed classifier, acceleration classifier and trajectory fluctuation classifier, and uses the cross-validation method to obtain the optimal weight to obtain the motion trajectory classification model. 10. the flying bird and unmanned aerial vehicle classification model system based on center of mass motion feature according to claim 9, it is characterized in that, described system also comprises stability evaluation module, and described trajectory classification model is to the trajectory in every section trajectory The category probability is calculated for the segment, and the classification result is output after the probability assessment is performed by the stability assessment module.

Images