Inter-frame difference algorithm selection method for target detection and tracking

Abstract

The invention discloses an inter-frame difference algorithm selection method for target detection and tracking. The inter-frame difference algorithm selection method comprises the specific steps that S1, parameters of optical equipment are input into a calculation system; s2, setting time for restarting algorithm selection, and when the timer times t = time, executing the step S3; s3, calculating a relative velocity measurement factor alpha of the target; s4, judging whether alpha is greater than 1 or not; s5, judging whether alpha is located at (alphamin, 1) or not; and S6, according to the int value, executing a dual-frame difference detection algorithm of an interval int frame. The invention provides an efficient inter-frame difference algorithm selection mechanism based on quantitative analysis, the optimal detection algorithm can be selected according to the current condition, the target detection performance is improved, the method is not only suitable for long-distance small target infrared detection, but also can be migrated to other complex target detection fields needing high precision and high timeliness, and the method has a wide application prospect. Wide application prospects are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and particularly relates to a method for selecting an inter-frame difference algorithm for target detection and tracking. Background Art

[0002] As a simple and effective means for detecting moving targets, the inter-frame difference method is widely used; this method calculates the pixel-level differences between two or more consecutive frames of images, and distinguishes the stationary background from the moving foreground by setting a threshold, so as to realize the recognition of moving objects; in military applications, especially in complex battlefield environments, moving targets generally approach our area gradually from a relatively long distance, and detecting the target as early as possible at a long distance determines the early warning time of our side, which has important military significance; however, for the detection of moving targets at a long distance, due to the different characteristics of the parameter configuration of the detection equipment, the distance and speed of the moving target, etc., and they are changing in real time; therefore, in different application scenarios, the performance of each inter-frame difference algorithm will also show a huge gap; how to select the most suitable algorithm at present is a crucial problem.

[0003] Traditional inter-frame difference methods each have certain limitations. When dealing with fast-moving targets, the double-frame difference method may cause the "double shadow" phenomenon; the triple-frame difference method reduces the occurrence of the "hole" phenomenon by introducing a third frame as a reference, and is particularly suitable for the detection of high-speed moving targets. For low-speed targets, the detection performance of the triple-frame difference method will be greatly reduced; the multi-frame difference method further utilizes more historical information and enhances the ability to resist complex motion patterns and transient interferences, but at the same time increases the computational complexity and resource consumption; according to the relative speed of the target motion, the triple-frame difference algorithm is suitable for high-speed moving targets. For low-speed targets, the triple-frame difference method cannot detect them. At this time, the double-frame difference algorithm is more effective; if the target speed continues to decrease, it is necessary to perform double-frame difference operations on images with an interval of multiple frames.

[0004] The prior art lacks a systematic framework that can automatically select the most suitable inter-frame difference algorithm according to the parameter configuration of the detection device, the distance, speed and other characteristics of the moving target. Most current algorithm selections rely on experience, qualitative analysis or fixed rules, and it is difficult to meet the requirements of diverse and constantly changing actual application scenarios. The existing methods for selecting inter-frame difference algorithms mainly face the following problems: (1) Lack of adaptability: Many applications use fixed inter-frame difference algorithms. When the application scenario changes, the set difference algorithm is still used, which cannot meet the optimal performance requirements in all cases. (2) Lack of quantification and scientific nature: Although there is current experience in the application of different algorithms, most of them are the results of qualitative analysis, which is difficult to support scientific decision-making in actual applications. Moreover, the understanding of the target speed is not scientific enough. The high or low speed of the target movement is relative. It is not simply the actual speed, but is directly related to the distance between the target and the optical device, the projection of the speed relative to the optical device, the shooting frequency of the optical device, the resolution of the optical device, and the viewing angle of the optical device.

[0005] Currently, there is a lack of an effective technical solution that can automatically match the best inter-frame difference algorithm for the detection device according to the relative motion speed of the target to improve the overall performance of moving target detection. Summary of the Invention

[0006] The object of the present invention is to overcome the deficiencies of the prior art and provide a method for selecting an inter-frame difference algorithm for target detection and tracking. According to the speed characteristics of different moving targets, a relative speed measurement factor is calculated to assist decision-makers in selecting the most suitable inter-frame difference algorithm to improve the accuracy and efficiency of moving target detection.

[0007] The technical solution adopted by the present invention is as follows:

[0008] A method for selecting an inter-frame difference algorithm for target detection and tracking, the specific steps are as follows:

[0009] S1: Input the horizontal viewing angle θ x and the vertical viewing angle θ y , the horizontal resolution r x and the vertical resolution r y , the frame rate f of the optical detection device and the minimum relative motion speed factor α min into the calculation system; the minimum relative motion speed factor α min has a value range of (0, 1);

[0010] S2: According to the specific application scenario, preset a time time for re - starting the algorithm selection, start the timer. If the timing t of the timer is less than time, the system continues to maintain the selection of the original algorithm and does not take any action; when the timer timing t = time, perform operation S3, and clear the timer.

[0011] S3: Input the distance d from the current target to the optical device, the horizontal velocity projection v x and the vertical velocity projection v y into the system, and calculate the relative velocity measurement factor of the target as α; the specific process is as follows:

[0012] First, calculate the ground sampling distance in the horizontal and vertical directions according to the parameters of S1. The ground sampling distance, GroundSample Distance, i.e., GSD, describes the physical size covered by each pixel in the actual scene:

[0013]

[0014] The horizontal distance D x and vertical distance D y that the target moves during the time interval between consecutive frames of images are:

[0015]

[0016] Then, the relative velocity measurement factors α x and α y in the horizontal and vertical directions can be expressed as:

[0017]

[0018] The overall relative velocity measurement factor α can be expressed as:

[0019] α = max(α x , α y );

[0020] S4: Judge whether α is greater than 1. If so, adopt the three - frame - between detection algorithm and exit the calculation system; if not, perform operation S5.

[0021] S5: Judge whether α is in the range of (α min , 1). If so, adopt the two - frame - between detection algorithm and exit the calculation system. When the value of α is less than α min , perform operation S6.

[0022] S6: Set the value of α best , where α best represents the ideal relative motion speed factor using the two - frame - between difference algorithm, and α bestThe preferred value range is (α min , 0.75). Calculate the multiple of α relative to α best , that is, the number of frame intervals int. The ceil function represents rounding up to an integer; according to the int value, execute the double-frame difference detection algorithm with an interval of int frames and exit the system.

[0023] Preferably, the value range of the minimum relative motion speed factor α min is (0.25, 0.5).

[0024] Due to the above-mentioned technical solution, the present invention has the following advantages:

[0025] The present invention provides a quantitative analysis-based and efficient frame difference algorithm selection mechanism, which can select the best detection algorithm according to the current conditions, improve the target detection performance, and is not only applicable to long-distance small-target infrared detection, but also can be migrated to other complex target detection fields that require high precision and high timeliness, and has a wide application prospect. Brief Description of the Drawings

[0026] Figure 1 is a flowchart of the present invention.

[0027] Figure 2 is a table of the scene conditions of the infrared image weak aircraft target detection and tracking dataset in the ground / air background in the embodiment of the present invention.

[0028] Figure 3 is the basic performance parameter information of the infrared camera in the embodiment of the present invention.

[0029] Figure 4 is the basic parameter information of the unmanned aerial vehicle in the embodiment of the present invention.

[0030] Figure 5 is a comparison chart of the detection probabilities of the selection algorithm and the fixed algorithm in the embodiment of the present invention. Detailed Embodiment

[0031] The following further explains the present invention in conjunction with the drawings and embodiments. The protection scope of the present invention cannot be limited thereby. The purpose of disclosing the present invention is to protect all technical improvements within the scope of the present invention.

[0032] This embodiment is executed relying on a computer equipped with AMD Ryzen 5 5600K CPU, 16GB RAM, RTX 3060 GPU hardware function, and MATLAB 2024a software; the dataset is selected as "infrared image weak aircraft target detection and tracking dataset in the ground / air background", as Figure 2As shown, this data set is provided by the ATR Key Laboratory of the School of Electronic Science, National University of Defense Technology, which provides 21 types of frame sequence infrared image data and has relatively complete information markings; the present invention selects the data21 data set for experiments, and this data set simulates the infrared imaging of small moving targets at a long distance; Figure 3 and Figure 4 are the relevant parameters of the infrared camera and the drone provided by the data set, which are input into step S1 as prior information.

[0033] A method for selecting an inter-frame difference algorithm for moving target detection, the specific steps are as follows:

[0034] Execute S1, input the visual angle of the optical device in the computing system: the horizontal visual angle θ x is 3, the vertical visual angle θ y is 3, the device resolution: the horizontal resolution r x is 256, the vertical resolution r y is 256, the frame rate f of the optical detection device is 100, and the minimum relative motion speed factor α min is 0.25.

[0035] Execute S2, according to the specific application scenario, preset a time time for restarting the algorithm selection, set it to 120 seconds, start the timer, and when the timer counts to 120 seconds, continue to execute the operation of S3.

[0036] Execute S3, input the distance d from the current target to the optical device in the computing system as 5000 meters, the velocity projection v of the target in the horizontal direction x is 8.3, the velocity projections of the target in the vertical direction are respectively v y is 8.3, and calculate that the relative velocity measurement factor α of the target is 0.0815.

[0037] Execute S4, judge that 0.0815 is less than 1, and execute the operation of S5.

[0038] Execute S5, judge that 0.0815 is less than α min , and execute the operation of S6.

[0039] Execute S6, set the value of α best to 0.3, calculate to get int as 4, and output: execute the double inter-frame difference detection algorithm with an interval of 4 frames, and exit the system.

[0040] To illustrate the effectiveness of the selection algorithm provided by the present invention, the three-frame difference algorithm, the two-frame difference algorithm, and the two-frame difference algorithm with an interval of 4 frames provided by the present invention are respectively used to perform target detection operations on the data21 dataset, and the detection threshold is set to 20; the evaluation index refers to the description provided in "Dataset for Detecting and Tracking Small Aircraft Targets in Infrared Images under Ground / Air Backgrounds". A correct detection means that "there is exactly 1 detection result within the 3×3 annotation box (inclusive)". After completing the detection of all image sequences as required, the detection probability is the number of frames with correct detections divided by the total number of image frames; as Figure 5 shown: the detection probability of the three-frame difference algorithm is the lowest, at 16.06%, the two-frame difference algorithm is 37.88%, and the algorithm provided by the present invention has the best performance, with a detection probability of 83.06%, proving that the algorithm of the present invention is effective; this shows that when the relative motion of the target is too slow, the motion characteristics shown on the difference image are weak. At this time, the detection probability of the two-frame difference is relatively low, and the three-frame difference will be even lower due to the "AND operation"; the improved method is to perform the two-frame difference operation with an interval of multiple frames, so as to increase the relative speed and thus increase the detection probability of the target.

[0041] The settings of α, α min , time, α best in the present invention are not fixed and need to be flexibly changed according to the scene at that time; the principle of the algorithm selection method of the present invention is:

[0042] When the target moves slowly, it is inclined to use the two-frame difference detection algorithm; when the target speed is relatively fast, it is best to use the three-frame difference detection algorithm; however, by combining the basic principles of the difference algorithm and optical image generation, it can be found that the speed of the target is relative and is directly related to the distance between the target and the optical device, the angle between the target flight direction and the line connecting to the device, the resolution of the optical device, the shooting frequency of the optical device, and the viewing angle of the optical device, etc.; therefore, the present invention proposes to use the relative speed measurement factor to quantitatively analyze the speed of the moving target.

[0043] According to the two-frame difference principle, if the position of the moving target in the latter frame differs from that in the previous frame by more than 1 pixel, assuming that the target is a small target at this time, the "small" here is a relative concept. Quantitatively analyzed, it means that the target size is smaller than the corresponding unit GSD, and double imaging is likely to occur in target detection; therefore, for targets with relatively fast moving speeds, it is best to use the three-frame difference algorithm for stable detection and tracking. At this time, the relative speed measurement factor is greater than 1.

[0044] When the relative speed measurement factor is equal to 1, the detectable characteristics exhibited by the target in the double-frame difference image are the most obvious and easiest to detect. Therefore, the transition of its value around 1 determines the conversion of the system's optimal algorithm from the three-frame difference algorithm to the double-frame difference algorithm. When the relative speed measurement factor is less than 1, as it gradually decreases, the relative motion speed of the target becomes slower and slower, and the detectable characteristics exhibited by the target in the double-frame difference image also become weaker and weaker. For example, in infrared target detection, the gray value of a single pixel is related to the radiation energy within the corresponding GSD. As the target speed gradually slows down, the contribution of the target motion to the difference value of a single pixel becomes smaller and smaller. At this time, the probability of detecting the target using the double-frame difference decreases continuously, and the three-frame difference algorithm performs an "AND operation" on this basis, further reducing the target detection probability. To improve the detectability of the target, it is necessary to perform the double-frame difference algorithm at intervals of multiple frames.

[0045] When the change of the relative speed measurement factor α crosses α min or 1, the selection strategy of the algorithm will change. It should be noted that when α is close to the boundary between slow and fast, this is a problem of fuzzy decision-making. The decision-making logic of the system not only depends on the relative speed measurement factor itself but also needs to comprehensively consider the specific application scenario.

[0046] When α is less than 1, it indicates that the relative speed of the moving target is relatively slow. We consider choosing the double-frame difference algorithm. As α decreases, the relative speed of the moving target continuously decreases until it reaches a critical state. At this time, it is very difficult for the double-frame difference algorithm to detect it. Increasing the frame interval can effectively solve this problem. The speed at this critical state is represented by α min and α min is not a fixed value. The setting of α min needs to be based on the specific application scenario. For targets with relatively weak features or targets emphasizing timeliness, α min can be set relatively large. The present invention gives an empirical estimate and suggests taking a value between 0.25 and 0.5.

[0047] time represents the time value for reselecting the algorithm, and the setting also needs to be based on the specific application scenario. For tasks that emphasize timeliness, it needs to be updated in a timely manner, and the set value of time should be relatively small to quickly select the most suitable strategy for the current situation.

[0048] The parts not detailed in the present invention are prior art.

[0049] The embodiments selected in this article to disclose the invention purpose of the present invention are currently considered appropriate. However, it should be understood that the present invention is intended to include all changes and improvements of all embodiments belonging to the concept and scope of the invention.

Claims

1. A method for selecting an inter-frame difference algorithm for target detection and tracking, characterized in that: The specific steps are: S1: Set the horizontal viewing angle of the optical device to θ x and vertical viewing angle θ y , horizontal resolution r x and vertical resolution r y , the frame rate f of the optical detection device and the minimum relative motion speed factor α min Input calculation system; minimum relative motion speed factor α min The value range is (0,1); S2: According to the specific application scenario, a time for re-enabling algorithm selection is pre-set, and the timer is started. If the timer timing t is less than time, the system continues to select the original algorithm and does not take any action; when the timer timing t = time, the S3 operation is executed and the timer is reset; S3: Project the distance d from the current target to the optical device and the target's horizontal velocity v x and the vertical velocity projection v y Input the system and calculate the relative speed measurement factor of the target as α; the specific process is: First, the ground sampling distance in the horizontal and vertical directions is calculated according to the parameters of S1. The ground sampling distance GroundSample Distance, or GSD, describes the physical size covered by each pixel in the actual scene: The horizontal distance D that the target moves in the time interval between the previous and next frame images x , vertical distance D y for: Then, the relative speed measurement factor α of the target in the horizontal and vertical directions is x , α y It can be expressed as: The overall relative speed measurement factor α can be expressed as: α=max(α x ,a y ); S4: Determine whether α is greater than 1. If yes, use the three-frame detection algorithm and exit the calculation system. If no, execute S5; S5: Determine whether α is located at (α min ,1), if yes, then use the double-frame detection algorithm and exit the calculation system. When the α value is less than α min When , execute S6 operation; S6: Setting α best Value, α best represents the ideal relative motion speed factor using the two-frame difference algorithm, α best The preferred value range of is (α min , 0.75), calculate α relative to α best The multiple of is the number of inter-frame intervals int, The ceil function means rounding up to an integer; according to the int value, the double-frame difference detection algorithm with an interval of int frames is executed and the system is exited.

2. The method for selecting an inter-frame difference algorithm for target detection and tracking according to claim 1, characterized in that: The minimum relative motion speed factor α min The value range is (0.25,0.5).

Images