An evaluation method, system, terminal and medium for a single-target tracking model test

Abstract

The present application belongs to the technical field of tracking model test, and specifically discloses a kind of single target tracking model test evaluation method, system, terminal and medium, comprising: deploying single target tracking model to be evaluated in field test environment;Obtain the tracking box information of model on continuous image frame, including target pixel center coordinates, size parameters and time stamp;Tracking box information is structured and processed, and continuous tracking result sequence is formed;Based on the sequence, a multi-dimensional performance evaluation system is constructed;Each index is normalized, and weighted fusion is obtained according to the experience weight, and the comprehensive performance score is obtained.The evaluation method does not need to rely on test set label, can truly reflect the tracking performance of model in dynamic field environment, realize the comprehensive quantitative evaluation of model stability, smoothness, robustness and real-time response ability, improve the objectivity, engineering adaptability and automation level of model test.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tracking model testing, and particularly relates to a single-target tracking model test evaluation method, system, terminal and medium. BACKGROUND

[0002] Target tracking is one of the important research directions in the field of computer vision, and is widely used in video monitoring, intelligent transportation, automatic driving, security inspection and other fields. In the traditional algorithm development process, a single-target tracking model is usually trained and verified on a standard test dataset. Researchers usually rely on public test set labels to evaluate the performance of the model through common indicators such as MOTA, VOT, etc. These evaluation indicators are mostly based on offline data, and the model results are compared and analyzed with the artificial annotation results in a fixed scene, so as to judge the detection accuracy and positioning error of the model.

[0003] However, as tracking technology is gradually applied to edge computing scenarios in dynamic environments, such as real-time running on a turntable device with automatic servo capability, the traditional label-based offline evaluation method gradually exposes limitations: first, the test set label usually needs to be obtained in advance, and cannot reflect the generalization ability of the model in unknown environments; second, static evaluation cannot describe the response speed and stability of the model to dynamic targets in real scenes; in addition, there are complex situations such as target mutation, servo delay, size change, processing load fluctuation in the actual deployment process, which makes it difficult to evaluate the real performance through static labels.

[0004] There are also some evaluation methods based on artificial testing and experience judgment in the prior art, such as subjective scoring by test personnel observing tracking videos or recording the number of model failures, etc. However, such methods have the problems of strong subjectivity, low efficiency and poor reproducibility, and cannot meet the requirements of testing efficiency, accuracy and standardization in actual engineering deployment. SUMMARY

[0005] The application provides a single-target tracking model test evaluation method, system, terminal and medium, which solves the problems of strong subjectivity, low efficiency and poor reproducibility of the method caused by the test personnel observing the tracking video and performing subjective scoring or recording the number of model failures, etc.

[0006] The technical scheme adopted by the application is as follows:

[0007] In a first aspect, the application provides a single-target tracking model test evaluation method, which comprises the following steps:

[0008] Step S1, deploying a single-target tracking model to be evaluated in a real test environment;

[0009] Step S2, obtaining tracking box information output by the single-target tracking model on the continuous image frames, wherein the tracking box information includes pixel center coordinates of the target in each frame image and image size parameters, and timestamp information is recorded for each frame;

[0010] Step S3, structurally processing the tracking box information obtained in step S2 to obtain a tracking result sequence containing frame index, center coordinate sequence and width-height data;

[0011] Step S4, constructing a multi-dimensional performance evaluation system based on the tracking result sequence, wherein the multi-dimensional performance evaluation system is provided with a plurality of indexes;

[0012] Step S5, quantitatively calculating and normalizing each type of index, and weighting and fusing each result according to the index weight set by experience to obtain a comprehensive score value representing the real tracking performance of the model.

[0013] Further, in step S1, the real test environment includes an electro-optical tracking device provided with an electrically controlled turntable and a camera, and the single-target tracking model is deployed in an image processing device corresponding to the camera, and the electrically controlled turntable is used to drive the camera to rotate with the target.

[0014] Further, in step S2, when obtaining the tracking box information output by the single-target tracking model, the pixel center coordinates of the tracking box in each frame image and its width and height are recorded, and the time interval between frames is calculated based on the timestamp information corresponding to each frame.

[0015] Further, in step S3, the structural processing includes recording the tracking box information according to the timestamp information and the time interval between frames in the order of frames, and establishing the corresponding relationship of frame number, center point coordinates and width-height information to form a tracking result sequence of continuous frames.

[0016] Further, in step S4, the multi-dimensional performance evaluation system includes at least one or more of the following indexes:

[0017] a) global stability index based on full-sequence center coordinate dispersion;

[0018] b) local stability index based on center coordinate change within a servo period;

[0019] c) abnormal jitter index based on continuous frame center coordinate mutation detection;

[0020] d) smoothness index based on the rate of change of target size between consecutive frames;

[0021] e) trajectory retention capability index based on the proportion of effective frames in continuous tracking;

[0022] f) speed score index based on the processing time of each frame.

[0023] Further, in step S5, when normalizing each performance evaluation index, a maximum-minimum normalization method is adopted to map each index value to the interval [0, 1];

[0024] In the weighted fusion process, the normalized indexes are linearly weighted according to preset empirical weights, and the empirical weights are set according to the importance of the dimensions of stability, speed and robustness in different test scenarios.

[0025] The comprehensive score value obtained after the weighted calculation is limited to 0 to 1.

[0026] In a second aspect, the application provides an evaluation system for single-target tracking model testing, which comprises:

[0027] An image acquisition device configured on a controllable turntable, used to acquire continuous image frames of a target in a real test environment;

[0028] A single-target tracking model processing unit connected with the image acquisition device, used to perform target detection on the image frames and output tracking box information of the target in each image frame, the tracking box information including pixel center coordinates, width, height and corresponding time stamp;

[0029] A data processing unit used to receive and record the tracking box information and perform structured processing on the information to generate a tracking result sequence with frame sequence characteristics;

[0030] An evaluation calculation unit used to construct a performance evaluation system containing multiple evaluation indexes based on the tracking result sequence, and perform normalization and weighting processing on each evaluation index to output a score value for quantifying the real tracking performance of the model.

[0031] Further, the evaluation calculation unit comprises:

[0032] A normalization subunit used to process each evaluation index by a maximum-minimum normalization method to limit its value to the interval [0, 1];

[0033] A weighted fusion subunit used to linearly weight the normalized index results according to preset empirical weights, and the weights are statically configured and set according to the test scenario.

[0034] In a third aspect, the application provides a terminal comprising:

[0035] A memory used to store an evaluation program for single-target tracking model testing;

[0036] A processor used to perform the evaluation system for single-target tracking model testing to realize the steps of the evaluation method for single-target tracking model testing according to the first aspect.

[0037] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer instructions, and when a computer reads the computer instructions in the storage medium, the computer executes the evaluation method of the single target tracking model test according to the first aspect.

[0038] Through the above technical solutions, the advantages of the present application are as follows:

[0039] (1) The method provides an evaluation method of a single target tracking model test suitable for a field deployment stage, which can realize comprehensive quantification of the stability, smoothness, robustness, continuous tracking capability and processing speed of the model and improve the objectivity, engineering adaptability and automation degree of the model test stage by collecting the actual output results of the model and constructing a multi-dimensional evaluation system without relying on the test set labels.

[0040] (2) By limiting the field test environment to an optoelectronic tracking device composed of an electric control turntable and a camera, the test process is ensured to have a dynamic tracking scene, which can truly reflect the time sequence coupling relationship between target motion, servo response and model identification, and enhance the authenticity of the evaluation system and the adaptability of engineering deployment.

[0041] (3) While collecting the tracking box information, the center coordinates, size parameters and time stamps are recorded, and the inter-frame time interval is calculated, which can provide basic data support for the real-time response capability and processing consistency of the subsequent evaluation model, and help to find problems such as frame rate anomaly, frame skipping and delay.

[0042] (4) By structuring the tracking box information, the corresponding relationship between the frame order, center point coordinates and width-height parameters is established, which can form a complete continuous frame tracking result sequence, provide a unified and time sequence consistent data basis for the construction and statistical calculation of various performance indicators, and improve the accuracy and consistency of subsequent analysis.

[0043] (5) An evaluation system including multiple dimensions is constructed, which covers key dimensions such as stability, anomaly detection, size change, tracking continuity and processing speed, so that the comprehensive performance of the model in the actual scene can be fully evaluated, which helps to locate the performance bottleneck, evaluate the deployment adaptability and guide the model optimization.

[0044] (6) By using maximum and minimum normalization to process each index value, and setting weights based on the importance of different test scenes for weighted fusion, the dimensions and ranges of different dimensional indicators can be effectively unified, so that the comprehensive score has comparability and interpretability, which is convenient for model performance sorting and selection. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions of the present application, the drawings required to be used in the description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and all other drawings obtained by those skilled in the art without creative labor based on these drawings also belong to the protection scope of the present application.

[0046] Figure 1 The flowchart of the evaluation method for the single-target tracking model test provided by the embodiment of the present application. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present application will be described clearly and completely in the following description with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor also belong to the protection scope of the present application.

[0048] Please refer to Figure 1 The present application provides an evaluation method for a single-target tracking model test, which comprises the following steps:

[0049] Step S1, deploying a single-target tracking model to be evaluated in a real test environment;

[0050] In an embodiment of the present application, a single-target tracking model to be evaluated is deployed in a real test environment. The test environment is constructed by using an optoelectronic tracking device, which comprises a camera installed on a controllable electric control turntable. The camera can rotate around the horizontal or vertical direction under the control of the turntable, and is used to collect target images in a moving state in real time.

[0051] The turntable and the camera are linked through a synchronous control system. The camera adjusts its posture according to the spatial displacement of the target, so as to realize continuous following shooting of the target. The electric control turntable is driven by a servo motor, which can update the angle at a fixed servo period in response to the control instruction, so as to ensure that the target is continuously in the center area of the field of view.

[0052] The single-target tracking model is deployed in an image processing device connected to the camera, which can be an edge computing unit or an embedded image processing platform. The image processing device receives the video stream from the camera and runs the tracking algorithm frame by frame. The model locates and identifies the target and outputs the tracking box information, which serves as the basic data for subsequent performance evaluation. After the model is deployed, it can be continuously run without relying on manual intervention or external label annotation, and can be effectively tested under real conditions.

[0053] The test environment can be arranged in an outdoor test field, an unmanned system platform or other test area with moving targets according to different experimental scenarios, and has good universality and expansibility.

[0054] In step S2, the tracking box information output by the single-target tracking model is acquired, wherein the tracking box information includes the pixel center coordinates of the target in each frame image and image size parameters, and the timestamp information is recorded for each frame;

[0055] In step S2, when the tracking box information output by the single-target tracking model is acquired, the pixel center coordinates of the tracking box in each frame image and the width and height thereof are recorded, and the inter-frame time interval is calculated based on the timestamp information corresponding to each frame;

[0056] In step S3, the tracking box information acquired in step S2 is structured to obtain a tracking result sequence including frame index, center coordinate sequence and width and height data.

[0057] In step S3, the structured processing includes recording the tracking box information in frame order according to the timestamp information and the inter-frame time interval, and establishing the corresponding relationship of frame number, center point coordinates and width and height information to form a tracking result sequence of consecutive frames.

[0058] In step S4, a multi-dimensional performance evaluation system is constructed based on the tracking result sequence, and the multi-dimensional performance evaluation system is provided with a plurality of indexes.

[0059] In step S4, the multi-dimensional performance evaluation system includes at least one or more of the following indexes:

[0060] a) global stability index based on full sequence center coordinate dispersion;

[0061] The experimental standard deviation is used as an index for evaluating global stability, which quantifies the stability of the tracking result and describes the smoothness and stability of the tracking box from the beginning to the end. The smaller the value, the more stable the tracking result and the better the smoothness.

[0062] Tracking box center point sequence:

[0063]

[0064]

[0065]

[0066] wherein, is the volatility index, is the average value of the tracking box pixel center point coordinates, x and y are the tracking box pixel center point coordinates, n is the total number of all tracking results from the beginning of the evaluation to the end of the evaluation, and i represents the current frame.

[0067] b) local stability index based on the change of the center coordinates within a servo cycle;

[0068] This is a kind of stability evaluation of turntable servo cycle, using Bezier formula as a period of stability evaluation calculation method, quantifying the stability of tracking results within a period, describing the smoothness of its motion, each cycle is calculated independently, wherein and The calculation reference refers to the relevant requirements in repeatability test, and the calculation starts when n is greater than 6 or 10, only the coordinate value within a certain evaluation period is taken, and the determination of the evaluation period is positively related to the servo period. The smaller the value, the more stable the tracking results and the better the smoothness.

[0069]

[0070]

[0071] In the formula, is the volatility index, and is the average value of the tracking frame pixel center coordinates within a period, x and y are the tracking frame pixel center coordinates, n represents the number of frames within a period, the value is greater than 6 or 10, if t is equal to 1, it represents the current stability with the experimental standard deviation of any tracking result within a period, if t is greater than 1, it represents the current stability with the average value of t times, and i represents the current frame.

[0072] c) Abnormal jitter index based on the sudden change detection of the center coordinates between consecutive frames;

[0073] Within a servo cycle, whether the tracking frame has a displacement mutation is evaluated, the results of the previous three frames are counted, and if the displacement of a frame is greater than the average displacement, it is considered to have a mutation. Each cycle is calculated independently, which is used to quantify the reliability of the tracking results. This index also reflects the stability to some extent. The smaller the value, the more reliable the tracking within a period.

[0074]

[0075]

[0076]

[0077]

[0078] If then it is determined to be abnormal jitter, is an empirical value which is adapted according to the deployment environment, and is usually between 1 and 3.

[0079]

[0080] l represents the cycle value, which can be reflected by the average value of multiple cycle mutation cases, and further reflect the overall process reliability.

[0081] d) a smoothness index based on the change rate of target size between consecutive frames;

[0082] In the servo control cycle, the size change of the tracking frame is evaluated, and each cycle is calculated independently. A mature and stable model, the size change of the target of the two frames should be uniform and close to linear, and the smaller the value, the more stable the tracking.

[0083]

[0084]

[0085] w, h represent the width and height of the tracking frame, and l represents the cycle value, which can be reflected by the average value of multiple cycle mutation cases, and further reflect the overall process smoothness.

[0086] e) a trajectory retention capability index based on the proportion of valid frames in continuous tracking;

[0087] Evaluate whether the model can track the same target for a long time, reflect the main reliability, and reflect the stability. Each cycle is calculated independently, and the final output result is averaged to reflect the overall trajectory retention capability. The smaller the value, the stronger the continuous tracking ability, and vice versa.

[0088]

[0089] Where ETD represents the total number of frames in the cycle, and D represents the number of frames in the cycle that continuously track the target.

[0090] f) a speed score index based on the processing time of each frame;

[0091] In percentage, the processing speed of each frame is used as a variable to build a segmented function to evaluate the processing speed of the model. Assuming that the model is input with a video stream each time, from the beginning of input to the completion of all result output, 0 is written in place when the tracking is lost, so as to reflect the processing ability of the model in the deployment segment. The larger the value, the faster the processing speed, and vice versa.

[0092]

[0093]

[0094] Step S5, quantitative calculation and normalization processing are performed on each type of index, and a comprehensive score value representing the field tracking performance of the model is obtained by weighting and fusing each result according to the index weight set by experience;

[0095] In step S5, when normalizing each performance evaluation index, a maximum-minimum normalization method is used to map each index value to the interval [0, 1];

[0096] In the weighting fusion process, the normalized indexes are linearly weighted according to the preset empirical weight. The empirical weight is set according to the importance of the stability, speed and robustness dimensions in different test scenarios.

[0097] The comprehensive score value obtained by weighting calculation is limited between 0 and 1.

[0098] To make the index value more understandable, the final index is normalized. The value tends to 1, indicating that the evaluation result is better, and vice versa. Among them represents the proportion weight of each index, which can be taken as an empirical value.

[0099]

[0100] Implementation case: assuming that a photoelectric tracking device with a coaxial turntable and camera continuously tracks a target in a field environment, the evaluation indexes of the tracking result are = 1.2, = 0.8, = 2, = 0.3, = 0.9, = 0.8. The weight is set as: = 0.2, = 0.3, = 0.1, = 0.2, = 0.1, = 0.1, and the calculation result is as follows.

[0101] = 0.44.

[0102] In some embodiments, the present application provides an evaluation system for a single-target tracking model test, which comprises:

[0103] An image acquisition device is arranged on a controllable turntable, and the image acquisition device comprises a high-definition camera and a turntable controller. The turntable drives the camera to continuously follow the target according to a preset motion track to obtain continuous image frames covering the whole motion of the target. The camera and the controller are connected through an industrial Ethernet or an RS485 interface to realize data and control signal interaction, so as to synchronize the time sequence of image acquisition and the posture information of the turntable.

[0104] A single-target tracking model processing unit is connected with the image acquisition device, and is used to receive the continuous image frames and perform target detection and tracking on the image frames. A trained single-target tracking model is arranged in the processing unit, and the model can output tracking box information for each image frame, wherein the tracking box information comprises a pixel center coordinate, a width and a height of the tracking box, and a timestamp corresponding to the current frame. The timestamp information is aligned with a camera trigger signal and a turntable control system clock, and is used as a time sequence reference for subsequent performance evaluation.

[0105] A data processing unit is used to receive and record the tracking box information, and the data processing unit comprises a data cache module and a structured processing module. The data cache module sequentially caches the model output results, and the structured processing module analyzes the tracking box information and the timestamp of each frame to generate a tracking result sequence comprising a frame number, a center coordinate, a width and height, and a frame interval. The sequence is stored in a database or a local file as input data for performance evaluation.

[0106] An evaluation calculation unit is used to construct a multi-dimensional performance evaluation system based on the tracking result sequence. The evaluation calculation unit internally comprises an index calculation module, a normalization module and a weighted fusion module. The index calculation module analyzes the center coordinate change, the size change and the frame interval in the tracking result sequence to generate multiple evaluation indexes including global stability, periodic stability, abnormal jitter, size smoothness, trajectory retention capability and tracking speed. The normalization module limits the index values in the interval [0, 1] by using a maximum and minimum normalization method. The weighted fusion module linearly weights and calculates the index results according to preset weight parameters to output a comprehensive score value, and the comprehensive score value is used to quantify the overall tracking performance of the single-target tracking model in the field test process.

[0107] In specific applications, the system can display a real-time tracking picture, a tracking index curve and a comprehensive score value through a human-computer interface, so as to facilitate users to intuitively obtain the model performance results.

[0108] In some embodiments, the evaluation calculation unit further comprises a normalization subunit and a weighted fusion subunit.

[0109] The normalization subunit is used for numerical standardization of various performance evaluation indexes. Specifically, for each index, the normalization subunit first sets the minimum and maximum value range of the index in historical samples or empirical statistics, and then converts the actual calculated index value into a standardized numerical value according to the interval proportion, so that the processed result is limited between zero and one. After normalization, the index values can be directly compared and weighted under the same dimension, avoiding the result deviation caused by the difference in index value range.

[0110] The weighted fusion subunit is used for linear weighting of the normalized evaluation indexes according to the pre-set weight. Each index corresponds to a statically configured weight coefficient, which can be set according to the key requirements of different test scenarios, for example, increasing the weight of the processing speed index in the scenario focusing on model response speed, and increasing the weight of the stability-related index in the scenario focusing on stability. The weighted fusion subunit multiplies the normalization results of each index with the corresponding weight item by item, and sums all the calculation results to obtain the final comprehensive score value. The score is used to represent the overall performance level of the single target tracking model under real deployment conditions.

[0111] The normalization subunit and the weighted fusion subunit can be realized by software configuration, or integrated as part of the same calculation module, with good expansibility, and can be flexibly adjusted according to different model types or evaluation requirements.

[0112] In some embodiments, the present application provides a terminal, comprising:

[0113] a memory for storing an evaluation program for single target tracking model test;

[0114] a processor for executing the evaluation system for single target tracking model test to implement the steps of the evaluation method for single target tracking model test.

[0115] In some embodiments, the present application provides a computer readable storage medium, which stores computer instructions. When the computer reads the computer instructions in the storage medium, the computer executes the evaluation method for single target tracking model test.

[0116] It can be understood that the systems, apparatuses, modules or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer, and the specific form of the computer can be a personal computer, a laptop computer, a personal digital assistant, a tablet computer, a wearable device, or a combination of any of these devices.

[0117] In a typical configuration, a computer includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0118] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory, etc. The memory is an example of computer readable media.

[0119] Computer readable media includes permanent and non-permanent, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic disk storage, quantum memory, graphene-based storage media, or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to computing devices. According to the definition in this embodiment, computer readable media does not include transitory media, such as modulated data signals and carriers.

[0120] It should also be noted that the terms "comprising," "including," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements recited, but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the statement "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0121] It should be understood that although the terms first, second, third, etc. can be used herein to describe various information, these terms are not intended to denote a particular order or hierarchy. These terms are used only to distinguish one category of information from another category of information. For example, a first information can be termed a second information, and similarly, a second information can be termed a first information without departing from the scope of one or more embodiments described herein. Depending on the context, the word "if' as used herein can be interpreted as meaning "when" or "upon" or "in response to determining."

[0122] The above description is only the preferred embodiment of one or more embodiments of the specification, and is not used to limit one or more embodiments of the specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of the specification should be included in the protection range of one or more embodiments of the specification.

Claims

1. A method of evaluating a single-target tracking model test, characterized by, The method comprises the following steps: Step S1, deploying a single-target tracking model to be evaluated in a field test environment; The field test environment comprises an optoelectronic tracking device provided with an electrically controlled turntable and a camera, the single-target tracking model is deployed in an image processing device corresponding to the camera, and the electrically controlled turntable is used to drive the camera to rotate with the target; Step S2, obtaining tracking box information output by the single-target tracking model on continuous image frames, wherein the tracking box information comprises pixel center coordinates and image size parameters of the target in each frame image, and timestamp information of each frame image is recorded; Step S3, performing structured processing on the tracking box information obtained in step S2 to obtain a tracking result sequence comprising a frame index, a center coordinate sequence and width-height data; Step S4, constructing a multi-dimensional performance evaluation system based on the tracking result sequence, and a plurality of indexes are provided in the multi-dimensional performance evaluation system; The multi-dimensional performance evaluation system comprises the following indexes: a) a global stability index based on the discrete degree of the whole sequence of center coordinates; b) a local stability index based on the change of the center coordinates within a servo cycle; c) an abnormal jitter index based on the detection of center coordinate mutations between continuous frames; d) a smoothness index based on the change rate of the target size between continuous frames; e) a trajectory retention capability index based on the proportion of effective frames in continuous tracking; f) a speed score index based on the processing time of each frame; Step S5, performing quantitative calculation and normalization processing on each type of index, and performing weighted fusion on each result according to the index weight set by experience to obtain a comprehensive score value representing the field tracking performance of the model.

2. The evaluation method of a single-target tracking model test according to claim 1, characterized in that, In step S2, when obtaining the tracking box information output by the single-target tracking model, the pixel center coordinates, width and height of the tracking box in each frame image are recorded, and the time interval between frames is calculated based on the timestamp information corresponding to each frame.

3. The method of claim 2, wherein, In step S3, the structured processing comprises recording the tracking box information according to the timestamp information and the time interval between frames in the order of frames, and establishing a corresponding relationship among the frame number, the center point coordinates and the width-height information to form a tracking result sequence of continuous frames.

4. The method of claim 1, wherein, In step S5, when performing normalization processing on each type of performance evaluation index, a maximum-minimum normalization method is adopted to map each index value to the interval [0, 1]; In the weighted fusion process, the normalized indexes are linearly weighted according to the preset empirical weight, and the empirical weight is set according to the importance of the stability, speed and robustness dimensions in different test scenarios; The comprehensive score value obtained after weighted calculation is limited to 0 to 1.

5. An evaluation system for a single-target tracking model test, characterized by, The system comprises: An image acquisition device configured on a controllable turntable, used to obtain continuous image frames of a target in a field test environment; The field test environment comprises an optoelectronic tracking device provided with an electrically controlled turntable and a camera; A single-target tracking model processing unit connected with the image acquisition device, used to perform target detection on the image frames and output tracking box information of the target in each frame image, the tracking box information comprising pixel center coordinates, width, height and corresponding timestamp; The single-target tracking model is deployed in an image processing device corresponding to the camera, and the electrically controlled turntable is used to drive the camera to rotate with the target; A data processing unit is configured to receive and record the tracking frame information, and to process the information in a structured manner to generate a tracking result sequence with frame sequence characteristics. An evaluation calculation unit is configured to construct a performance evaluation system including multiple evaluation indexes based on the tracking result sequence, and to perform normalization and weighting processing on each evaluation index to output a score value for quantifying the field tracking performance of the model. The multi-dimensional performance evaluation system includes the following indexes: a) a global stability index based on the discrete degree of the center coordinates of the whole sequence; b) a local stability index based on the change of the center coordinates within a servo cycle; c) an abnormal jitter index based on the detection of center coordinate mutations between consecutive frames; d) a smoothness index based on the change rate of the target size between consecutive frames; e) a trajectory retention capability index based on the proportion of effective frames in continuous tracking; f) a speed score index based on the processing time of each frame.

6. The evaluation system of single-target tracking model test according to claim 5, wherein, The evaluation calculation unit includes: a normalization subunit configured to process each evaluation index using a maximum-minimum normalization method to limit the value within the interval [0, 1]; a weighted fusion subunit configured to perform linear weighting on the normalized index results according to a preset empirical weight, and the weight is statically configured according to the test scene.

7. A terminal, characterized by comprising: It includes: a memory configured to store an evaluation program for single-target tracking model testing; a processor configured to implement the steps of the single-target tracking model testing evaluation method according to claim 1 when executing the single-target tracking model testing evaluation system.

8. A computer-readable storage medium, characterized in that, The storage medium stores computer instructions, and when the computer reads the computer instructions in the storage medium, the computer executes the single-target tracking model testing evaluation method according to claim 1.

Images