Small moving target detection method under complex background based on phase trajectory prediction

Abstract

The present application discloses a method, apparatus, device and medium for detecting weak moving targets in a complex background based on phase trajectory prediction. The method obtains real-time image data to be processed and determines a total matrix to be processed based on the real-time image data; performs frame-by-frame windowing processing on the total matrix to be processed to obtain each matrix to be processed, and uses a preset matrix segmentation algorithm to segment each matrix to obtain each sparse foreground component matrix, and obtains each frame in the sparse foreground component matrix; uses a preset motion target detection algorithm based on phase trajectory prediction to detect the foreground data of each frame, and obtains the weak moving target detection result and motion trajectory based on the foreground data of each frame; the present invention effectively utilizes the effective information contained in the sparse matrix, does not require background modeling, greatly simplifies the efficiency of target detection, and uses the motion target detection algorithm based on phase trajectory prediction to effectively filter out noise and interference in a complex and changing background.

Description

Technical Field

[0001] The present application relates to the field of target detection technology, and in particular to a method, device, equipment and medium for detecting weak moving targets in a complex background based on phase trajectory prediction. Background Art

[0002] Motion target detection is one of the core research contents in the security field, and the complex and changing background will cause serious interference to the detection work. In particular, when the target itself is relatively small and the target's own shape, movement speed and other aspects are not much different from the changing background, the interference of background noise will become particularly serious.

[0003] Hauberg et al. proposed a scalable robust principal component analysis method based on Gaussian mean (Grassmann Averages for RPCA, GA). The original method performed well in addressing data outliers and scalability, and had high operational efficiency. It was applied to moving target detection based on robust principal component analysis technology. Because this method can effectively detect data outliers, it is more effective in removing common noise in target detection tasks. However, when the background noise is more complex, the performance of this method becomes less ideal. In the same paper, the team also proposed another more efficient robust principal component analysis solution method, namely the scalable robust principal component analysis method based on Gaussian median (Grassmann Median for RPCA, GM). In general, this method has higher operational efficiency than the Gaussian mean (GA) method, but the performance is worse. However, comparative experiments have shown that this method is also prone to the problem of target and noise confusion in moving target detection tasks. Summary of the Invention

[0004] The main purpose of this application is to provide a method, device, equipment and medium for detecting weak moving targets in a complex background based on phase trajectory prediction, aiming to solve the problem of moving target detection when the background is complex and changing, the target is relatively weak, or the target and background are easily confused.

[0005] To achieve the above-mentioned purpose, the present application provides a method for detecting weak and small moving targets in a complex background based on phase trajectory prediction, including: obtaining real-time image data to be processed, and determining a total matrix to be processed based on the real-time image data; performing frame and window processing on the total matrix to be processed to obtain each matrix to be processed, and using a preset matrix segmentation algorithm to segment each matrix to be processed to obtain each sparse foreground component matrix, and obtaining each frame in the sparse foreground component matrix; using a preset motion target detection algorithm based on phase trajectory prediction to detect the foreground data of each frame, and obtaining a weak and small moving target detection result and motion trajectory based on the foreground data of each frame; and outputting the weak and small moving target detection result and motion trajectory.

[0006] Optionally, the use of a preset motion target detection algorithm based on phase trajectory prediction to detect foreground data of each frame includes: searching each sparse foreground component matrix based on a preset extremely sparse criterion to obtain the center points of the current frame and two adjacent frames in each frame; filtering out some center points in the current frame and two adjacent frames according to preset conditions to obtain the remaining multiple center points as targets, and performing neighborhood clustering on each target to obtain predefined phase, phase size, phase and phase weight, and determining the current target based on the phase, the phase size and the phase weight. A plurality of groups of one-to-one corresponding phase center points in a frame and two adjacent frames, wherein each phase center point is the center point, and the phase center point and the phase correspond one-to-one; processing each group of phase center points based on a preset first formula to obtain each group of phase differences, and processing each group of phase differences based on a preset second formula to obtain each phase angle; retaining the phase corresponding to each phase difference angle less than a first preset threshold in the current frame, and filtering out each phase corresponding to each phase difference angle greater than a second preset threshold, to obtain foreground data in the current frame.

[0007] Optionally, filtering out some center points in the current frame and two adjacent frames according to preset conditions includes: determining whether the current frame is the first frame in each of the sparse foreground component matrices; if the current frame is the first frame, retaining each of the center points of the current frame; if the current frame is not the first frame, performing a neighborhood search on each of the center points, and determining whether the current center point belongs to the neighborhood search range of the previous center point during the neighborhood search; if the current center point belongs to the neighborhood search range of the previous center point, retaining the current center point; if the current center point does not belong to the neighborhood search range of the previous center point, filtering out the current center point.

[0008] Optionally, taking the center point as the phase center point, the target points are subjected to neighborhood clustering to obtain predefined phase, phase size, phase and phase weight, including: performing neighborhood clustering on each target point, taking the clustering result as the phase, taking the points contained in the clustering result as the phase size, taking the phase center point coordinates as the phase, and taking the average value of the values ​​of the points contained in the clustering result as the phase weight.

[0009] Optionally, the determining of multiple groups of one-to-one corresponding phase center points in the current frame and two adjacent frames based on the phase, the phase size and the phase weight includes: judging whether the absolute value of the phase difference between each corresponding phase center point in the two adjacent frames is less than a preset value; if the absolute value is less than a first preset value, determining that the two phase center points in the two adjacent frames match; if the absolute value is greater than or equal to the first preset value and less than or equal to a second preset value, performing neighborhood clustering on all phase center points in the subsequent frame with the phase of the phase center point in the previous frame of the two adjacent frames as the center, obtaining the phase size and phase weight of all phase center points in the subsequent frame, and searching the subsequent frame for the phase center point that is closest in phase size and phase weight to the phase center point of the previous frame, obtaining a phase center point that matches the phase center point of the previous frame; if the absolute value is greater than the second preset value, determining that there is no matching object between the phase center point in the previous frame and the phase center point in the subsequent frame.

[0010] Optionally, each group of phase center points is processed based on a preset first formula to obtain each group of phase differences, and each group of phase differences is processed based on a preset second formula to obtain each phase angle, including: obtaining each first phase difference and each second phase difference based on the corresponding matching phase center points in the current frame and the two adjacent frames, and obtaining each phase angle of the current frame based on each first phase difference and each second phase difference.

[0011] Optionally, the first phase difference and the second phase difference are obtained based on the phase center points that match each other in the current frame and the two adjacent frames, and the phase angles of the current frame are obtained based on the first phase difference and the second phase difference, including: calculating the vector difference between the phase center points that match each other in the current frame and the previous frame, and obtaining the first phase differences; calculating the vector difference between the phase center points that match each other in the current frame and the next frame, and obtaining the second phase differences, wherein the second phase differences correspond to the first phase differences one-to-one; and obtaining the phase angles based on the one-to-one correspondence between the first phase differences and the second phase differences.

[0012] To achieve the above-mentioned purpose, the present application also provides a device for detecting weak and small moving targets in a complex background based on phase trajectory prediction, including: an acquisition module, used to acquire real-time image data to be processed, and determine the total matrix to be processed based on the real-time image data; a matrix segmentation module, used to perform frame and window processing on the total matrix to be processed to obtain each matrix to be processed, and use a preset matrix segmentation algorithm to segment each matrix to be processed to obtain each sparse foreground component matrix, and obtain each frame in the sparse foreground component matrix; a processing module, used to detect the foreground data of each frame using a preset motion target detection algorithm based on phase trajectory prediction, and obtain weak and small moving target detection results and motion trajectories based on the foreground data of each frame; an output module, used to output the weak and small moving target detection results and motion trajectories.

[0013] To achieve the above-mentioned purpose, the present application also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the method for detecting weak moving targets in a complex background based on phase trajectory prediction provided by any of the aforementioned embodiments.

[0014] To achieve the above-mentioned purpose, the present application also provides an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for detecting weak moving targets in a complex background based on phase trajectory prediction provided by any of the aforementioned embodiments.

[0015] The embodiments of the present application propose a method, device, equipment and medium for detecting weak moving targets in a complex background based on phase trajectory prediction, which obtains real-time image data to be processed and determines the total matrix to be processed based on the real-time image data; performs frame and window processing on the total matrix to be processed to obtain each matrix to be processed, and uses a preset matrix segmentation algorithm to segment each matrix to obtain each sparse foreground component matrix, and obtains each frame in the sparse foreground component matrix; uses a preset motion target detection algorithm based on phase trajectory prediction to detect the foreground data of each frame, and obtains weak moving target detection results and motion trajectories based on the foreground data of each frame; outputs the weak moving target detection results and motion trajectories, which can effectively filter out noise and interference in a complex and changing background. The present invention effectively utilizes the effective information contained in the sparse matrix, no longer performs background modeling, and greatly simplifies the efficiency of target detection. Moreover, the present invention uses the motion target detection algorithm based on phase trajectory prediction to detect targets, which can effectively improve the detection accuracy and reduce the impact of noise and interference on detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flowchart of a method for detecting weak moving targets in a complex background based on phase trajectory prediction in this application;

[0017] Figure 2 This is a schematic diagram of the principle of the method for detecting weak moving targets in a complex background based on phase trajectory prediction in this application;

[0018] Figure 3 Schematic diagram of the phase, phase center point, phase, phase difference, and phase difference angle of the method for detecting weak moving targets in a complex background based on phase trajectory prediction in this application;

[0019] Figure 4 is a diagram showing the actual detection results of the method for detecting weak moving targets in a complex background based on phase trajectory prediction of this application, wherein: Figure 4(a) is an unframed original video; Figure 4(b) is a diagram showing the detection results of the robust principal component analysis moving target detection method based on Gaussian mean (GA); Figure 4(c) is a diagram showing the detection results of the robust principal component analysis moving target detection method based on Gaussian median (GM); and Figure 4(d) is a diagram showing the detection results of the present invention.

[0020] Figure 5 This is a structural block diagram of the device for detecting weak moving targets in a complex background based on phase trajectory prediction in this application. DETAILED DESCRIPTION

[0021] It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application.

[0022] In order to effectively solve the problem of moving target detection when the background is complex and changing, the target is relatively weak or the target and the background are easily confused, a weak moving target detection method based on phase trajectory prediction under complex background is invented. Figure 1 ,The detection methods of weak moving targets in complex backgrounds based on phase trajectory prediction include:

[0023] Step S10: acquiring real-time image data to be processed, and determining a total matrix to be processed based on the real-time image data;

[0024] It should be noted that the present application can use the server to obtain input image or video frame sequence data, extract the grayscale value of each pixel in the image or video frame sequence, and construct a spatiotemporal grayscale information matrix as the total matrix to be processed (Xt) based on the spatial position and time sequence of each pixel. The total matrix to be processed can be updated in real time, that is, if there is new input data, it will be filled in the back of the original total matrix to be processed according to the time sequence, and the processed part of the total matrix to be processed can be released on demand.

[0025] In a specific embodiment, a video of a long-range tiltrotor aircraft shot by a person holding a camera device is input into the server as input data, the grayscale value of each pixel in the video is extracted frame by frame, and a spatiotemporal grayscale information matrix is ​​constructed according to its spatial position and time sequence as the total matrix to be processed (Xt).

[0026] Step S20: performing frame segmentation and windowing processing on the total matrix to be processed to obtain individual matrices to be processed, and using a preset matrix segmentation algorithm to segment each matrix to be processed to obtain corresponding sparse foreground component matrices, and obtaining each frame in the sparse foreground component matrix;

[0027] refer to Figure 2 Specifically, the matrix cross-section of the total matrix to be processed (Xt) at each moment is called a frame, and each matrix value in each frame is called a point. The sliding window detection method of this application divides the total matrix to be processed (Xt) into frames and windows according to a fixed time window length to obtain the matrix to be processed (X). In other words, the current data (i.e., the data in the current time window) submitted to the detection method for processing is the matrix to be processed (X).

[0028] In a specific embodiment, when executing the sliding window detection method, the server uses 25 frames as a fixed time window length for frame windowing, and performs frame windowing on the total matrix to be processed (Xt) to obtain the matrix to be processed (X).

[0029] It should be noted that the present invention can be used for real-time detection. However, real-time detection is divided into two stages: target detection and noise filtering. Target detection occurs in the first frame of each time window, while noise filtering occurs in the remaining frames of each time window. Therefore, during real-time detection, it is necessary to use a fixed time window for frame division and windowing, and the time window should not be too long to ensure effective detection of new targets.

[0030] After obtaining the matrix to be processed, the server performs a sparse matrix decomposition on the matrix (X) using the inexact Lagrange multiplier method to obtain a low-rank background component matrix (Lb) and a sparse foreground component matrix (Sf). The matrix cross-section of each moment in the low-rank background component matrix (Lb) and the sparse foreground component matrix (Sf) can be called a frame according to the time sequence, and each matrix value in each frame is called a point.

[0031] Among them, the method of using the inexact Lagrange multiplier method to perform matrix sparse decomposition on the processing matrix is ​​as follows:

[0032] Where, X = Lb + Sf, and the decomposition method is as follows:

[0033]

[0034] In the formula, ||·|| *is the kernel function, ||·||1 is the l1 norm, and λ=0.06.

[0035] refer to Figure 2 Step S30: Detect foreground data of each frame using a preset moving target detection algorithm based on phase trajectory prediction, and obtain a weak moving target detection result and a motion trajectory based on the foreground data of each frame;

[0036] In an embodiment of the present application, step S30 may include the following execution process:

[0037] S301, searching each sparse foreground component matrix based on a preset extremely sparse criterion to obtain the center point of the current frame and two adjacent frames in each frame;

[0038] The extremely sparse criterion is to select a number of maximum points in the current frame of the sparse foreground component matrix (Sf), or first divide each frame of data into blocks and then select the regional maximum point in each block. This application uses the method of selecting a number of maximum points in the current frame and two adjacent frames of the sparse foreground component matrix (Sf) as the center point.

[0039] S302: Filter out some center points in the current frame and two adjacent frames according to preset conditions, obtain the remaining multiple center points as target points, perform neighborhood clustering on each target point, obtain predefined phase, phase size, phase, and phase weight, and determine multiple groups of one-to-one corresponding phase center points in the current frame and two adjacent frames based on the phase, phase size, and phase weight, wherein each phase center point is a center point, and the phase center point and the phase have a one-to-one correspondence;

[0040] In an embodiment of the present application, filtering out some center points in the current frame and two adjacent frames according to a preset condition may include the following execution process:

[0041] Determine whether the current frame is the first frame in each sparse foreground component matrix;

[0042] If the current frame is the first frame, the center points of the frame are retained;

[0043] If the current frame is not the first frame, a neighborhood search is performed on each center point of the current frame, and during the neighborhood search, it is determined whether the current center point belongs to the neighborhood search range of the previous center point;

[0044] If the current center point belongs to the neighborhood search range of the previous center point, the current center point is retained;

[0045] If the current center point does not belong to the neighborhood search range of the previous center point, the current center point is filtered out.

[0046] Specifically, the above steps are used to determine the center points of each frame. The server detects whether the current frame is the first frame of each time window. If so, all center points are retained. If it is a frame other than the first frame, a neighborhood search is performed on each center point of the other frame. If the current center point is within the neighborhood search range of the previous center point, the center point is retained. If the current center point is not within the neighborhood search range of the previous center point, the center point is filtered out.

[0047] It should be noted that neighborhood search refers to searching within a certain neighborhood area. The neighborhood area can be manually defined or generated by a server. The neighborhood area is a fixed range. The selection of the range can be determined by the phase size and the phase difference. For example, the range of the neighborhood area can be set to [4*4], [8*8], [16*16], etc. The neighborhood area can also be a superpixel area. In this application, the neighborhood area is set to [25*25].

[0048] In another embodiment, when the background is relatively pure or no filtering is required, this step can be omitted; when the background is relatively complex, this step can be performed multiple times within one cycle using different neighborhood sizes.

[0049] In an embodiment of the present application, step S302 may further include the following execution process:

[0050] S3021. Perform neighborhood clustering on each target point, use the clustering result as the phase, the points included in the clustering result as the phase size, the coordinates of the phase center point as the phase, and the average value of the values ​​of the points included in the clustering result as the phase weight.

[0051] Specifically, after obtaining the center point of each frame, the value, coordinates and other information of the center point that meets the retention requirements are recorded, and neighborhood clustering is performed with this point as the target. The clustering result is called a phase, the number of points contained in the phase is called the phase size, the coordinates of the phase center point are called the phase, and the average value of the values ​​of the points contained in the phase is calculated as the phase weight.

[0052] refer to Figure 3 , in the figure: 1-S 1,1 ,S 1,2 ,S 1,3 They are the phase center points of the same phase (S1) in the first, second and third frames, and their coordinates are S 1,1 (x,y),S 1,2 (x,y),S 1,3 (x,y) is called the phase;

[0053] 2-P 1,1-2 is the phase difference between the first and second frames of the same phase (S1), P 1,2-3 is the phase difference between the second and third frames of the same phase (S1), and the phase difference is a vector;

[0054] 3-φ 1,1-3 The angle between the phase differences of the same phase (S1) in the first and second frames and the phase differences in the second and third frames is called the phase angle or phase difference angle of the S1 phase in the first to third frames.

[0055] For example, the i-th corresponding phase center point in the j-th frame is denoted as S i,j , its coordinate value, i.e. phase, is S i,j (x,y), value is S i,j (x,y)[z], where the phase is S i or S i (j) (When the previous and next frames are not distinguished, it is recorded as S i , called the same phase, specifically the i-th phase of the j-th frame is recorded as S i (j)), the phase point is denoted as S i (m,n), (m,n) is the coordinate value of a point in the phase, and the value of the phase point is recorded as S i (m,n)[z], the phase size, that is, the number of phase points is M i (j) = numS i (m,n), the phase weight is the average value of the phase point is recorded as:

[0056]

[0057] In an embodiment of the present application, step S302 may further include the following execution process:

[0058] S3022, determining whether the absolute value of the phase difference between the corresponding phase center points in two adjacent frames is less than a preset value;

[0059] S3023: If the absolute value is less than the first preset value, determine that the two phase center points in the two adjacent frames match;

[0060] S3024. If the absolute value is greater than or equal to the first preset value and less than or equal to the second preset value, performing neighborhood clustering on all phase center points in the subsequent frame with the phase of the phase center point in the previous frame of the two adjacent frames as the center, obtaining the phase sizes and phase weights of all phase center points in the subsequent frame, and searching in the subsequent frame for the phase center point that is closest in both phase size and phase weight to the phase center point in the previous frame, to obtain the phase center point that matches the phase center point in the previous frame;

[0061] S3025: If the absolute value is greater than the second preset value, determine that there is no matching object between the phase center point in the previous frame and the phase center point in the next frame.

[0062] By executing steps S3022 to S3025, the server can determine the phases and phase centers corresponding to adjacent frames (previous and next frames). Specifically, when the phase distances of the corresponding phase centers of the previous and next frames are close (the absolute value of the two phase differences is less than a first preset value) and the phase distances of different phase centers are far apart (the absolute value of the two phase differences is greater than a third preset value), the matching relationship between the phase centers of the previous and next frames can be found based solely on the phase information, and the phase center point with the farther distance can be determined as a non-matching point. The phases of the corresponding phase centers of the previous and next frames are said to be the same phase, that is, the two phase centers match. When the phase distances between the corresponding phase center points and different phase center points in the current and subsequent frames are not much different (the absolute value of the two phase difference values ​​is greater than the first preset value and less than or equal to the second preset value) and confusion occurs, in order to find the corresponding phase center point of a phase center point in the previous frame in the subsequent frame, first, the phase of a phase center point in the previous frame is taken as the center, and all the phase center points in the subsequent frame are clustered using a clustering method. Then, the phase size, phase weight and other information of the phase where the phase center point that meets the above-mentioned clustering requirements is located in the subsequent frame are compared. It is considered that the phase with the closest phase size and phase weight in the previous and subsequent frames is the same phase, and the phase center point where it is located is the matching phase center point.

[0063] S303, processing each group of phase center points based on a preset first formula to obtain each group of phase differences, and processing each group of phase differences based on a preset second formula to obtain each phase angle;

[0064] In an embodiment of the present application, step S303 may include the following execution process:

[0065] The first formula is used to process the phase center points matched in the current frame and two adjacent frames to obtain the first phase differences and the second phase differences, and the second formula is used to process the first phase differences and the second phase differences to obtain the phase angles of the current frame.

[0066] For example, the phase distance (vector difference) between the phase center points of the previous and next frames is calculated as the phase difference (vector). i , a phase difference P can be obtained for every two adjacent frames (j-1 frame, j frame) i,(j-1)-j =S i,j (x j ,y j )-S i,j-1 (x j-1 ,y j-1 )(First formula), where (x j ,y j ) and (x j-1 ,y j-1 ) represent S i The phase difference P is between the jth frame and the j-1th frame. Then, two phase differences P can be obtained for every three adjacent frames (j-1th frame, jth frame, j+1th frame). i,(j-1)-jand P i,j-(j+1) , according to the time sequence, the previous phase difference is called P i,(j-1)-j is the phase difference of the previous frame, and the next (currently calculated) phase difference is called P i,j-(j+1) The current phase difference P is calculated using the second formula. i,j-(j+1) Phase difference P from the previous frame i,(j-1)-j The second formula is:

[0067]

[0068] S304 , retaining phases corresponding to phase difference angles smaller than a first preset threshold in the current frame, and filtering out phases corresponding to phase difference angles larger than a second preset threshold, to obtain foreground data in the current frame.

[0069] Among them, the angle When the angle is less than a certain value, the phase is retained as the detection foreground and enters the next frame detection cycle; when the angle When it is greater than a certain value, the phase is filtered out. The present invention calls the above process the determination of target motion direction not changing suddenly, based on which the present invention proposes a criterion of target motion direction not changing suddenly. The criterion of target motion direction not changing suddenly believes that the turning of a maneuvering target in the motion process has limitations, and there is a process of first slowing down and then turning, or first slowing down and then turning around. In particular, for targets detected by computer vision detection technologies including the present invention, they generally do not have the ability to suddenly and drastically change their motion direction within a very short period of time between adjacent frames (previous and next frames), while the movement of noise is arbitrary. Based on this, effective filtering of complex background noise can be achieved.

[0070] Step S40: Output the weak moving target detection result and motion trajectory.

[0071] Referring to Figure 4, Figure 4(a) is the unframed image of the original video; Figure 4(b) is the detection result diagram of the robust principal component analysis motion target detection method based on Gaussian mean (GA); Figure 4(c) is the detection result diagram of the robust principal component analysis motion target detection method based on Gaussian median (GM); Figure 4(d) is the detection result diagram of the present invention.

[0072] During the specific execution process, the detection result of the current frame is saved and output. When there is data to be detected, the next frame or the next sliding time window is entered for subsequent detection; when there is no data to be detected, the detection is paused or ended.

[0073] The beneficial effects of the method of the present invention mainly include:

[0074] (1) Real-time operation is crucial for the practical application of moving target detection technology. A moving target detection method that cannot achieve real-time operation is actually of no practical value. The present invention has a clear process and relatively low computational complexity. As verified by the experiment of the specific embodiment 1, it can achieve rapid target detection and can run in real time on a computer with average performance.

[0075] (2) The present invention solves the problem that sparse decomposition-based motion target detection methods generally require the use of background low-rank matrix information and the adoption of background modeling methods to achieve target detection. It creatively directly uses the foreground sparse matrix (noise matrix) to extract target information for detection, and at the same time effectively filters out a large amount of irrelevant information such as noise and interference contained in the sparse matrix, thereby better detecting moving targets, especially weak moving targets.

[0076] (3) The size, dimensions, distance, and posture of the same object in different frames of a video vary, making it objectively difficult to establish a connection between them. This invention draws on relevant concepts in the field of traditional signal processing and creatively defines concepts such as phase, phase center, phase size, phase weight, phase, phase difference, and phase difference angle of video content. By calculating this information, the same object (moving target, etc.) in adjacent frames of a video can be effectively matched and connected, thus providing important value for detecting and tracking targets.

[0077] (4) The present invention proposes a criterion for no sudden change in target motion direction and a corresponding filtering algorithm, which can effectively deal with the impact of complex and changing backgrounds on motion target detection, thereby better adapting to situations where the camera may move or shake, as well as situations where the background content may change dramatically, thereby effectively reducing the impact of noise and interference on detection, making the detection results more stable and accurate.

[0078] refer to Figure 5 On the basis of the above embodiments, the present application further provides a device for detecting weak moving targets in a complex background based on phase trajectory prediction. The target detection device 100 includes an acquisition module 1001, a matrix segmentation module 1002, a processing module 1003 and an output module 1004: the acquisition module 1001 is used to acquire real-time image data to be processed, and determine the total matrix to be processed based on the real-time image data; the matrix segmentation module 1002 is used to perform frame segmentation and windowing processing on the total matrix to be processed to obtain each matrix to be processed, and use a preset matrix segmentation algorithm to segment each matrix to be processed to obtain each sparse foreground component matrix, and obtain each frame in the sparse foreground component matrix; the processing module 1003 is used to detect the foreground data of each frame using a preset motion target detection algorithm based on phase trajectory prediction, and obtain a weak moving target detection result and motion trajectory based on the foreground data of each frame; the output module 1004 is used to output the weak moving target detection result and motion trajectory.

[0079] It is not difficult to find that this embodiment is a system embodiment corresponding to the above-mentioned method embodiment, and this embodiment can be implemented in conjunction with the above-mentioned method embodiment. The relevant technical details and technical effects mentioned in the above-mentioned embodiments are still valid in this embodiment, and to reduce repetition, they are not repeated here. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above-mentioned embodiments.

[0080] It is worth mentioning that all modules involved in this embodiment are logical modules. In actual applications, a logical unit can be a physical unit, a part of a physical unit, or a combination of multiple physical units. In addition, to highlight the innovation of this application, this embodiment does not introduce units that are not closely related to solving the technical problems proposed by this application. However, this does not mean that other units do not exist in this embodiment.

[0081] Another embodiment of the present application proposes an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for detecting weak moving targets in a complex background based on phase trajectory prediction in the above-mentioned method embodiments.

[0082] The memory and processor are connected using a bus, which can include any number of interconnected buses and bridges. The bus connects various circuits of one or more processors and memories. The bus can also connect various other circuits such as peripheral devices, voltage regulators, and power management circuits. These are all well known in the art and therefore will not be described further in this article. The bus interface provides an interface between the bus and the transceiver. The transceiver can be a single component or multiple components, such as multiple receivers and transmitters, providing a unit for communicating with various other devices on a transmission medium. Data processed by the processor is transmitted on a wireless medium via an antenna. Furthermore, the antenna also receives data and transmits it to the processor.

[0083] The processor is responsible for managing the bus and general processing, and may also provide various functions, including timing, peripheral interfacing, voltage regulation, power management, and other control functions. Magnetic memory can be used to store data used by the processor while performing operations.

[0084] Another embodiment of the present application relates to a computer-readable storage medium storing a computer program, which implements the above method embodiment when executed by a processor.

[0085] That is, those skilled in the art will understand that all or part of the steps in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a program, which is stored in a storage medium and includes a number of instructions for causing a device (which may be a single-chip microcomputer, chip, etc.) or a processor to execute all or part of the steps in the various embodiments of the present application. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), a magnetic disk, or an optical disk, etc., various media that can store program code.

[0086] Those skilled in the art will appreciate that the above embodiments are specific embodiments for implementing the present application, and that in actual applications, various changes may be made thereto in form and detail without departing from the spirit and scope of the present application.

[0087] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for detecting small moving targets in complex backgrounds based on phase trajectory prediction, characterized in that: include: Acquiring real-time image data to be processed, and determining a total matrix to be processed based on the real-time image data; Performing frame-by-frame windowing on the total matrix to be processed to obtain individual matrices to be processed, and segmenting each matrix to be processed using a preset matrix segmentation algorithm to obtain corresponding sparse foreground component matrices, and obtaining each frame in the sparse foreground component matrices; wherein a matrix cross-section of the total matrix to be processed at each moment is called a frame, and the sparse foreground component matrices are obtained by performing matrix sparse decomposition on the matrix to be processed using an inexact Lagrange multiplier method; Detecting foreground data of each frame using a preset moving target detection algorithm based on phase trajectory prediction, and obtaining a weak moving target detection result and a motion trajectory based on the foreground data of each frame; Outputting the weak moving target detection result and motion trajectory; The detecting of the foreground data of each frame by using a preset moving target detection algorithm based on phase trajectory prediction includes: Searching each of the sparse foreground component matrices based on a preset extremely sparse criterion to obtain the center points of the current frame and the two adjacent frames in each of the frames; Among them, the extremely sparse criterion is to screen out several maximum value points in the current frame of the sparse foreground component matrix; Filtering out some center points in the current frame and two adjacent frames according to preset conditions to obtain the remaining multiple center points as target points, performing neighborhood clustering on each of the target points to obtain predefined phases, phase sizes, phases, and phase weights, and determining multiple groups of one-to-one corresponding phase center points in the current frame and two adjacent frames based on the phases, the phase sizes, and the phase weights, wherein each phase center point is the center point, and the phase center points and phases have a one-to-one correspondence; The sparse foreground component matrix refers to the matrix cross section at each moment in time sequence as a frame, and each matrix value in each frame is called a point; Processing each group of phase center points based on a preset first formula to obtain each group of phase differences, and processing each group of phase differences based on a preset second formula to obtain each phase angle; retaining the phases corresponding to the phase difference angles smaller than a first preset threshold in the current frame, and filtering out the phases corresponding to the phase difference angles larger than a second preset threshold, to obtain foreground data in the current frame; Taking the center point as the phase center point, performing neighborhood clustering on each of the target points to obtain predefined phases, phase sizes, phases, and phase weights include: Neighborhood clustering is performed on each of the target points, and the clustering result is used as the phase, the points included in the clustering result are used as the phase size, the coordinates of the phase center point are used as the phase, and the average value of the values ​​of the points included in the clustering result is used as the phase weight.

2. The method for detecting small moving targets in complex backgrounds based on phase trajectory prediction according to claim 1, characterized in that: The filtering out some center points in the current frame and two adjacent frames according to a preset condition includes: Determining whether the current frame is the first frame in each of the sparse foreground component matrices; If the current frame is the first frame, retaining the center points of the current frame; If the current frame is not the first frame, performing a neighborhood search on each center point of the current frame, and determining whether the current center point belongs to a neighborhood search range of a previous center point during the neighborhood search; If the current center point belongs to the neighborhood search range of the previous center point, the current center point is retained; If the current center point does not belong to the neighborhood search range of the previous center point, the current center point is filtered out.

3. The method for detecting small moving targets in complex backgrounds based on phase trajectory prediction according to claim 1, characterized in that: The determining of a plurality of groups of one-to-one corresponding phase center points in a current frame and two adjacent frames based on the phase, the phase size, and the phase weight includes: Determine whether the absolute value of the phase difference between the corresponding phase center points in two adjacent frames is less than a preset value; If the absolute value is less than a first preset value, determining that the two phase center points in the two adjacent frames match; If the absolute value is greater than or equal to a first preset value and less than or equal to a second preset value, then taking the phase of the phase center point in the previous frame of two adjacent frames as the center, perform neighborhood clustering on all phase center points in the next frame to obtain the phase sizes and phase weights of all phase center points in the next frame, and search in the next frame for the phase center point that is closest in both phase size and phase weight to the phase center point in the previous frame to obtain the phase center point that matches the phase center point in the previous frame; If the absolute value is greater than the second preset value, it is determined that there is no matching object between the phase center point in the previous frame and the phase center point in the next frame.

4. The method for detecting small moving targets in complex backgrounds based on phase trajectory prediction according to claim 1, wherein: The processing of each group of phase center points based on a preset first formula to obtain each group of phase differences, and the processing of each group of phase differences based on a preset second formula to obtain each phase angle, includes: The first formula is used to process the phase center points that match the current frame and the two adjacent frames to obtain the first phase differences and the second phase differences, and the second formula is used to process the first phase differences and the second phase differences to obtain the phase angles of the current frame.

5. The method for detecting small moving targets in complex backgrounds based on phase trajectory prediction according to claim 4, characterized in that: Obtaining a first phase difference and a second phase difference based on the respective matching phase center points in the current frame and two adjacent frames, and obtaining the phase angles of the current frame based on the first phase difference and the second phase difference, including: Calculating vector differences between the phase center points that match the current frame and the previous frame, and correspondingly obtaining the first phase differences; Calculating vector differences between the phase center points that match the current frame and the next frame to obtain corresponding second phase differences, wherein each second phase difference corresponds to each first phase difference one by one; Based on the one-to-one correspondence between each of the first phase differences and each of the second phase differences, corresponding phase angles are obtained.

6. A device for detecting small moving targets in a complex background based on phase trajectory prediction, using the method for detecting small moving targets in a complex background based on phase trajectory prediction as claimed in any one of claims 1 to 5, characterized in that: include: an acquisition module, configured to acquire real-time image data to be processed, and determine a total matrix to be processed based on the real-time image data; a matrix segmentation module, configured to perform frame segmentation and windowing processing on the total matrix to be processed to obtain individual matrices to be processed, and to segment each of the matrices to be processed using a preset matrix segmentation algorithm to obtain corresponding sparse foreground component matrices, and to obtain each frame in the sparse foreground component matrices; a processing module, configured to detect foreground data of each frame using a preset moving target detection algorithm based on phase trajectory prediction, and obtain a weak moving target detection result and a motion trajectory based on the foreground data of each frame; The output module is used to output the weak moving target detection result and motion trajectory.

7. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for detecting weak moving targets in a complex background based on phase trajectory prediction as described in any one of claims 1 to 5 can be implemented.

8. An electronic device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; In which, the memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the method for detecting weak moving targets in a complex background based on phase trajectory prediction as described in any one of claims 1 to 5.

Images