System and method for determining three-dimensional target detection evaluation index

Abstract

The invention discloses a system and a method for determining a three-dimensional target detection evaluation index, which are applied to the technical field of precision measurement and aim at solving the problems that continuous tracking cannot be realized and the position of a three-dimensional true value frame of a three-dimensional target is difficult to directly associate in the prior art. According to the method, a plurality of laser trackers are used as a rectangular frame, at least three surfaces in the rectangular frame are selected as dynamic truth value frames of a three-dimensional target object (T), a brand-new target position explaining mode is given through a plurality of truth value frame center points, and high-precision positioning is realized in different scenes through a mode of joint calibration of the plurality of laser trackers or differential positioning of the plurality of RTK mobile stations.

Description

Technical Field

[0001] This invention belongs to the field of precision measurement technology, and particularly relates to the field of three-dimensional target image data processing technology. Specifically, it relates to a method, electronic device, apparatus and system for determining three-dimensional target detection evaluation indicators. Background Technology

[0002] In fields such as autonomous driving, robot navigation, and industrial measurement, the accuracy of 3D target detection technology directly depends on the reliability of evaluation metrics. Current mainstream evaluation methods (such as IoU and mAP) are typically based on simulation data or ground truth boxes calculated under static calibration environments, but they suffer from the following key problems:

[0003] (1) Insufficient precision in obtaining the truth value:

[0004] Traditional methods rely on manual annotation or high-precision 3D scanning equipment (such as LiDAR point cloud registration) to obtain the ground truth box. However, manual annotation is subject to subjective errors, while point cloud registration is limited by environmental noise and occlusion, making it difficult to guarantee the accuracy of the ground truth box position in dynamic scenes.

[0005] (2) Poor adaptability to dynamic scenes:

[0006] Existing evaluation systems (such as camera-based motion capture systems) are susceptible to interference in outdoor environments with large areas or complex lighting conditions, and the multi-sensor calibration process is cumbersome and makes it difficult to synchronize the ground truth data of moving targets in real time.

[0007] (3) The contradiction between static calibration and dynamic scenarios:

[0008] Traditional laser trackers use a fixed target ball for system calibration (such as Leica multi-station measurements), but when the target moves, the motion needs to be repeatedly paused for recalibration, making continuous tracking impossible. While RTK-based localization methods support dynamic targets, a single RTK base station cannot directly correlate the 3D ground truth bounding box position of a 3D target. Summary of the Invention

[0009] To address the aforementioned technical problems, this invention provides a method and system for determining evaluation indicators for three-dimensional target detection that features high measurement accuracy, simple calibration, and a wider range of applications.

[0010] One of the technical solutions adopted in this invention is: a system for determining the evaluation index of three-dimensional target detection, applied to scenarios where the target and the data acquisition system do not move simultaneously, including a three-dimensional target object (T) inscribed in a rectangular frame, a laser tracker group, and a target ball group. The laser tracker group includes three laser trackers, and three faces in the rectangular frame are selected as three dynamic truth boxes of the three-dimensional target object (T). The target ball group includes three target balls, which are respectively fixed at the center points of the three truth boxes of the three-dimensional target object (T) and fixed on the three-dimensional target object (T) according to the center point positions of the truth boxes. The three laser trackers track one target ball respectively.

[0011] The second technical solution adopted in this invention is: a system for determining the evaluation index of three-dimensional target detection, applied to a scenario where the target and the data acquisition system move synchronously, including a three-dimensional target object (T) inscribed in a rectangular frame, one fixed-station RTK and three mobile-station RTKs. The fixed-station RTK is set at a fixed position on the ground as a base station, and three faces in the rectangular frame are selected as three dynamic truth boxes of the three-dimensional target object (T). The antennas of the three mobile-station RTKs are respectively fixed at the center point of the three truth boxes of the three-dimensional target object (T). The base station and the three mobile-station RTKs maintain communication.

[0012] The third technical solution adopted in this invention is: a method for determining three-dimensional target detection evaluation indicators, applied to scenarios where the target and the data acquisition system do not move simultaneously, including:

[0013] A1. The three-dimensional target object (T) is inscribed within a rectangular frame;

[0014] A2. Select at least three faces within the rectangular frame as the dynamic truth box of the 3D target object (T);

[0015] A3. Fix a target ball at the center point of each dynamic truth box of the three-dimensional target object (T), and each laser tracker tracks a target ball.

[0016] A4. Move the three-dimensional target object (T) and record the measurement data and measurement time of each laser tracker respectively;

[0017] A5. Based on the three-dimensional position measured by each laser tracker and the relative position calibrated by the laser tracker, calculate the three-dimensional position information of each target ball at each measurement time in the same coordinate system;

[0018] A6. Based on the real-time three-dimensional position information of the target ball in the same coordinate system, obtain the position information of the center point of the dynamic truth frame of the three-dimensional target object when the target and the data acquisition system move at different times.

[0019] The fourth technical solution adopted in this invention is: a method for determining three-dimensional target detection evaluation indicators, applied to scenarios where the target and the data acquisition system move synchronously, including:

[0020] B1. The three-dimensional target object (T) is inscribed within a rectangular frame;

[0021] B2. Select at least three faces within the rectangular frame as the dynamic truth box of the 3D target object (T);

[0022] B3. Fix an antenna at the center of each of the three truth boxes of the three-dimensional target (T), and connect one antenna to each mobile station RTK.

[0023] B4. Move the three-dimensional target object (T) and record the measurement data and measurement time of each rover station's RTK.

[0024] B5. Obtain the real-time three-dimensional position of the antenna measured by RTK;

[0025] B6. Based on the real-time three-dimensional position information of the antenna in the same coordinate system, obtain the position information of the center point of the three-dimensional target truth frame when the target and the data acquisition system move synchronously.

[0026] The beneficial effects of this invention are as follows: This invention inscribes a three-dimensional target within an inscribed rectangular frame, representing the three-dimensional target using at least three points on the three surfaces of the frame, thus proposing a novel way to interpret target position. When the target moves but the data acquisition system remains stationary, a laser tracker system is used; when the target is stationary but the data acquisition system moves, a laser tracker system is also used; when both the target and the data acquisition system move, an RTK system is used. The laser tracker system, combined with multi-laser tracker joint calibration, can achieve millimeter-level measurement accuracy. The RTK system, through multi-station differential positioning, can achieve centimeter-level real-time positioning in open environments, and can serve as an evaluation index for three-dimensional target detection. The scene-specific dynamic truth box calibration method proposed in this invention has the following significant advantages:

[0027] 1. This invention provides a high-precision ground truth acquisition method. The laser tracker system achieves millimeter-level three-dimensional position measurement accuracy through rigid binding of the target sphere to the center point of the ground truth box, combined with joint calibration using multiple laser trackers. This significantly outperforms manual annotation or lidar point cloud registration methods. The RTK system, through multi-rover differential positioning, can achieve centimeter-level real-time positioning in open environments, meeting the high-frequency ground truth box update requirements for dynamic targets.

[0028] 2. This invention has stronger adaptability to dynamic scenes. The laser tracker mode is suitable for scenarios where the target or data acquisition system is stationary. It can continuously output truth box data through a single calibration, avoiding the repeated calibration problems caused by target movement in traditional methods. The RTK mode is suitable for scenarios where the target and acquisition system move synchronously. It directly associates the center point of the truth box with the coordinates of the RTK rover station, eliminating the need for additional calculations and reducing system complexity.

[0029] 3. This invention features multimodal data uniformity. Both modes are based on the center point of the truth box as a benchmark, ensuring the consistency of evaluation index calculations (such as IoU, mAP, etc.) and avoiding data alignment errors caused by sensor differences.

[0030] 4. This invention offers system deployment flexibility. The laser tracker array can be distributed to adapt to the measurement needs of targets of different sizes; the RTK system does not require a fixed base station location and supports large-scale outdoor dynamic testing. It can automatically switch operating modes based on the target's motion state (e.g., switching from laser tracker to RTK), improving the efficiency of true value acquisition.

[0031] In summary, this invention has good accuracy, efficiency and adaptability, and provides a standardized truth acquisition scheme with high accuracy, high dynamics and high flexibility for three-dimensional target detection and evaluation. It can be widely used in scenarios such as UAV target detection, autonomous driving simulation verification and industrial robot accuracy evaluation. Attached Figure Description

[0032] Figure 1 A calibration system based on a laser tracker;

[0033] Figure 2 This is an RTK-based calibration system. Detailed Implementation

[0034] The present invention will be further described below with reference to the accompanying drawings and specific embodiments:

[0035] Example 1

[0036] To address the problems existing in the prior art, this invention proposes a scene-specific dynamic truth box calibration method: When the target and the data acquisition system do not move simultaneously, a laser tracker system is used, and the three-dimensional target object (T) is inscribed in a rectangular frame. The rectangular frame includes six faces: top, bottom, left, right, front, and back. At least one face from each of the top, bottom, left, right, and front / back groups is selected as the dynamic truth box. The target ball is fixed at the center point of the truth box, and the real-time three-dimensional target truth box position is obtained by calculating the calibration results and the output three-dimensional position of the target. When the target and the data acquisition system move synchronously, an RTK system is used, and the target's three-dimensional truth box position is directly output through differential positioning of multiple mobile stations and a base station fixed at the center point of the truth box.

[0037] A three-dimensional object (T) is not necessarily a regular shape, making it difficult to determine its centroid and centroid. In this invention, when the three-dimensional object is inscribed within a rectangular frame, the intersection point of the object with the rectangular frame on one of its faces is not necessarily the center point of that rectangular frame. This invention selects the geometric center of the rectangular frame as the center point of the truth box.

[0038] Those skilled in the art will understand that the data acquisition system referred to here refers to various sensors used for three-dimensional target recognition, detection, positioning, and tracking. The calibration results of the laser tracker and RTK of this invention are used as true values ​​to evaluate the accuracy, precision, and error of the measurement results from these sensors.

[0039] The evaluation index provided by this invention adopts a scenario-specific dynamic truth box calibration method. In this embodiment, the scenario using a laser tracker system is mainly introduced. When the target moves synchronously with the data acquisition system, an RTK system is used.

[0040] Here, we'll first explain the meaning of "ground truth": In machine learning, "ground truth" generally refers to the real information we obtain through observation and measurement during the dataset collection phase. It's used to evaluate model performance or guide model training; it represents the true labels or real values ​​in the dataset. In point cloud processing, "ground truth" typically refers to a reference dataset or standard used to compare and evaluate algorithm performance. In 3D computer vision and point cloud processing, "ground truth" can be an accurate 3D model or a dataset containing the precise location, category, or other attributes of each point in the point cloud.

[0041] Specifically:

[0042] (1) Definition and uses:

[0043] Ground truth can be a standard used to evaluate the performance of an algorithm. For example, in a semantic segmentation task, ground truth might be a dataset that labels the category to which each point belongs.

[0044] The ground truth can be compared with the algorithm's output to calculate performance metrics such as intersection-union ratio, precision, and recall, thereby evaluating the algorithm's quality.

[0045] (2) Acquisition method:

[0046] Manual annotation: Creating ground truth by manually annotating point cloud data. Manual annotation is a time-consuming and expensive process, but it yields very accurate results.

[0047] Raw sensor observations: Accurate ground truth is generated directly using raw or fused data from various sensors (such as LiDAR, cameras, etc.).

[0048] Simulated data: In some applications, simulated data can be used to generate ground truth. This method can quickly generate large amounts of data, but it may not be as accurate as real data.

[0049] (3) Application in point cloud processing:

[0050] 3D Reconstruction: In 3D reconstruction tasks, the ground truth can be an accurate 3D model used to evaluate the accuracy of the reconstruction algorithm.

[0051] Target detection and tracking: In target detection and tracking tasks, ground truth can include the target's precise location, size, and trajectory information.

[0052] (3) Challenges and limitations:

[0053] Creating high-quality ground truth is a challenge, especially when dealing with large-scale or complex point cloud data.

[0054] Due to factors such as labeling errors or sensor noise, the ground truth itself may also have some inaccuracies.

[0055] In some applications, obtaining complete ground truth is difficult, which limits the evaluation and improvement of algorithms.

[0056] In summary, ground truth is an important tool for evaluating algorithm performance, improving algorithms, and driving the development of the field. This invention provides a standardized ground truth acquisition scheme for 3D object detection that offers better accuracy, efficiency, and adaptability.

[0057] This embodiment uses the selection of three dynamic truth boxes as an example for illustration:

[0058] This laser tracker system includes a laser tracker assembly, a three-dimensional target object (T), and a target sphere assembly. For example... Figure 1 As shown, the laser tracker group includes laser trackers L1, L2, and L3, with three laser trackers in the group, corresponding to three dynamic truth boxes of the 3D target. The target sphere group includes target spheres B1, B2, and B3, with three target spheres in the group. Target spheres B1, B2, and B3 are fixed on the target object (T) according to the center point of the truth box. First, one target sphere is selected as the target, and the position of the target sphere on each of the three laser trackers is obtained. The relative positions between the laser trackers are calculated to complete the system calibration. The position coordinate system of the laser trackers can be based on the coordinates of any one of the laser trackers.

[0059] The system calibration method refers to the description in patent application No. 2025105526011, entitled "A Six-DOF Target Positioning and Tracking Accuracy Calibration System and Method Based on Laser Tracker". The process is as follows: First, select a target ball as the target, ensuring that all three laser trackers can collect the laser signal of the target ball. Then, obtain the position L1(x) of the target ball on each of the three laser trackers. 10 y 10 , z 10 L2(x) 20 y 20 , z 20 ) and L3 (x 30 y 30 , z 30 The relative positions between the laser trackers were calculated. , , Complete system calibration. The position coordinate system of the laser tracker can be based on the coordinates of any laser tracker.

[0060] In this embodiment, the method for determining the evaluation index of three-dimensional target detection includes the following steps:

[0061] S11. Set the target ball to be fixed at the center point of the three-dimensional target object truth box, and the laser tracker tracks the corresponding target ball respectively;

[0062] S12. Move the three-dimensional target object (T) and record the measurement data and measurement time of the laser tracker. Alternatively, move the data acquisition system and record the measurement data and measurement time of the laser tracker.

[0063] S13, Based on the three-dimensional position L1 (x) measured by the laser tracker 1i y 1i , z 1i L2(x) 2i y 2i , z 2i L3(x) 3i y 3i , z 3i The relative position calibrated by the laser tracker , , Calculate the three-dimensional position information B1(x) of the three target balls at each measurement time in the same coordinate system. 1i y 1i , z 1i ), B2(x 2i y 2i , z 2i ), B3 (x 3i y 3i , z3i ); where (x, y, z) represent the measured location coordinates, and i represents the different times of measurement.

[0064] The calculation of the target's three-dimensional position information based on the three-dimensional position measured by the laser tracker and the relative position calibrated by the laser is a known existing technology and will not be described in detail here.

[0065] S14. Based on the real-time position information of the target ball in the same coordinate system, obtain the position information of the center point of the three-dimensional target object's true value frame, and represent it as B1(x 1i y 1i , z 1i ), B2(x 2i y 2i , z 2i ), B3 (x 3i y 3i , z 3i ).

[0066] This scenario applies to both situations where the target moves but the data acquisition system remains stationary, and situations where the target is stationary but the data acquisition system moves. Combined with multi-laser tracker joint calibration, this laser tracker system can achieve millimeter-level measurement accuracy and can serve as an evaluation metric for 3D target detection. Because the method of this invention uses the center point of the truth box as a reference, it ensures the consistency of the calculation of evaluation metrics (such as IoU, mAP, etc.).

[0067] There are generally seven common evaluation metrics for object detection: (1) positive and negative samples; (2) true positive (TP), false positive (FP), true negative (TN), and false negative (FN); (3) intersection-over-union ratio (IoU); (4) precision; (5) recall; (6) geometric mean score (F Score); and (7) average precision for each class.

[0068] 1. Positive and negative samples

[0069] In computer vision evaluation, samples are a crucial concept. Positive samples are objects to be detected, while negative samples are targets to be excluded. In actual algorithms, a portion of the detected candidate regions is treated as positive samples, and a portion as negative samples. For example, when detecting faces, faces are positive samples, while non-faces are negative samples.

[0070] 2. True Positive (TP), False Positive (FP), True Negative (TN), False Negative (FN)

[0071] 21. Correct Positive Prediction (TP): The number of positive samples that are correctly detected. Three conditions must be met: (a) The confidence score is greater than the threshold. In this example, the threshold is set to 85%. In fact, all predicted boxes must meet this condition; (b) The prediction type matches the label type; (c) The intersection over union (IoU) ratio between the predicted bounding box and the ground truth box is greater than the threshold (the threshold ranges from 0 to 1). When there are multiple pre-selected boxes that meet the conditions, the one with the highest confidence score is selected as TP, and the rest are selected as FP.

[0072] 22. False Positive (FP): The number of negative samples that are detected as positive samples. Also known as false alarms, these are detection boxes whose IoU with the ground truth is less than a threshold (localization error) or whose predicted type does not match the label type (classification error).

[0073] 23. False Negative (FN): The number of positive samples that are not detected as positive samples, also known as missed detections, refers to undetected Ground Truth regions.

[0074] 24. True Negative (TN): This refers to the number of negative samples that are detected. It cannot be calculated, and TN is usually not considered in object detection.

[0075] 3. Intersection over Union (IoU)

[0076] Intersection over Union (IoU) calculates the overlap ratio between two regions by dividing the intersection of the two regions by their union. The formula is as follows, where DR represents the detection result and GT represents Ground Truth.

[0077]

[0078] 4. Accuracy

[0079] Precision, also known as accuracy, is the percentage of total accuracy (TP) among the identified objects.

[0080]

[0081] This represents the total number of objects identified.

[0082] 5. Recall

[0083] Recall rate is the ratio of correctly identified objects to the total number of objects.

[0084]

[0085] This indicates the total number of objects that need to be detected.

[0086] 6. Geometric Mean (F Score)

[0087] Precision and Recall may seem similar, but in most cases they are mutually exclusive.

[0088] Therefore, using Precision and Recall as evaluation criteria alone is too one-sided. A higher threshold can be set to obtain high Precision, or a lower threshold can be set to obtain a higher Recall value. Combining Precision and Recall yields a more comprehensive result. The calculation formula is as follows:

[0089]

[0090] It's the harmonic mean of Precision and Recall. B acts as a weight, adjusting the proportions of Precision and Recall. A characteristic of the harmonic mean is that if either Precision or Recall is small, the overall mean will be small; that is, the penalty for having a particularly small value between the two is relatively large. It's also common to see the F1 score used as one of the evaluation criteria, with B set to 1. .

[0091] 7. Average Precision (Average Precision for Each Class)

[0092] Average accuracy is a common metric for object detectors (object detection algorithm models). While AP literally stands for average precision, simply focusing on the precision value alone doesn't accurately measure a model's performance. Therefore, AP isn't calculated by simply averaging. Generally, AP is defined as the area under the precision curve (PR curve), as shown in the following formula:

[0093]

[0094] The values ​​of Precision and Recall are between 0 and 1, so the value range of AP is also [0, 1]. mAP (mean Average Precision) is the average of AP calculated after AP is calculated for each category.

[0095] Example 2

[0096] The evaluation index provided by this invention adopts a scenario-based dynamic truth box calibration method. In this embodiment, the scenario using an RTK system is mainly introduced.

[0097] This RTK system includes an RTK group and a three-dimensional target object (T). For example... Figure 2 As shown, an RTK is set up at a fixed position on the ground as a base station, and the antennas R1, antenna R2 and antenna R3 of the three mobile station RTKs are fixed on the target object (T) according to the center point of the truth box to ensure communication between the base station and the mobile station.

[0098] RTK surveying typically uses the WGS84 system. When RTK surveying requires a different coordinate system, coordinate transformation should be performed. Because a unified coordinate system is used, no additional relative position calibration or calculation between RTK groups is needed.

[0099] In this embodiment, the method for determining the evaluation index of three-dimensional target detection includes the following steps:

[0100] S11. Set the antenna to be fixed at the center point of the 3D target object's truth box, and connect the antenna to the RTK board;

[0101] S12. Move the three-dimensional target object (T) and record the measurement data and measurement time of the rover station RTK respectively;

[0102] S13, RTK real-time tracking and recording of 3D position R1 (x) 1i y 1i , z 1i ), R2(x 2i y 2i , z 2i ), R3(x 3i y 3i , z 3i (x, y, z) and time data, where (x, y, z) represent the measured location coordinates and i represents the different times of measurement;

[0103] S14. Based on the real-time position information of the antenna in the same coordinate system, the position information of the center point of the three-dimensional target object's true value frame is obtained and represented as R1(x1i y 1i , z 1i ), R2(x 2i y 2i , z 2i ), R3(x 3i y 3i (z3). Differential positioning using multiple rover stations in RTK systems is a known existing technology, and will not be described in detail here.

[0104] This scenario applies to situations where both the target and the data acquisition system are moving. The RTK system, through multi-station differential positioning, can achieve centimeter-level real-time positioning in open environments, serving as an evaluation metric for 3D target detection. Because this method uses the center point of the ground truth box as a reference, it ensures the consistency of evaluation metric calculations (such as IoU, mAP, etc.).

[0105] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A system for determining evaluation indicators for three-dimensional target detection, characterized in that, This system is applied to scenarios where the target and the data acquisition system do not move simultaneously. It includes a three-dimensional target object (T) enclosed in a rectangular frame, a laser tracker group, and a target ball group. At least three faces within the rectangular frame are selected as the dynamic truth frames of the three-dimensional target object (T). The number of laser trackers in the laser tracker group is the same as the number of dynamic truth frames, and the number of target balls in the target ball group is the same as the number of dynamic truth frames. A target ball is fixed at the center point of each dynamic truth frame of the three-dimensional target object (T), and one laser tracker tracks one target ball.

2. The system for determining evaluation indicators for three-dimensional target detection according to claim 1, characterized in that, The rectangle consists of six faces: top, bottom, left, right, front, and back. At least one face from each of the top, bottom, left, right, and front groups is selected as one of the three dynamic truth boxes.

3. A system for determining evaluation indicators for three-dimensional target detection, characterized in that, This system is applied to scenarios where the target and data acquisition system move synchronously. It includes a three-dimensional target object (T) enclosed in a rectangular frame, one fixed-station RTK, and multiple mobile RTKs. The fixed-station RTK is set at a fixed position on the ground as a base station, and at least three faces within the rectangular frame are selected as the dynamic truth frames of the three-dimensional target object (T). The number of mobile RTKs is the same as the number of dynamic truth frames. An antenna of a mobile RTK is fixed at the center point of each dynamic truth frame of the three-dimensional target object (T). The base station and each mobile RTK maintain communication.

4. The system for determining evaluation indicators for three-dimensional target detection according to claim 3, characterized in that, The rectangle consists of six faces: top, bottom, left, right, front, and back. At least one face from each of the top, bottom, left, right, and front groups is selected as one of the three dynamic truth boxes.

5. A method for determining evaluation indicators for three-dimensional target detection, characterized in that, Applications include scenarios where the target and the data acquisition system do not move simultaneously, including: A1. The three-dimensional target object (T) is inscribed within a rectangular frame; A2. Select at least three faces within the rectangular frame as the dynamic truth box of the 3D target object (T); A3. Fix a target ball at the center point of each dynamic truth box of the three-dimensional target object (T), and each laser tracker tracks a target ball. A4. Move the three-dimensional target object (T) and record the measurement data and measurement time of each laser tracker respectively; A5. Based on the three-dimensional position measured by each laser tracker and the relative position calibrated by the laser tracker, calculate the three-dimensional position information of each target ball at each measurement time in the same coordinate system; A6. Based on the real-time three-dimensional position information of the target ball in the same coordinate system, obtain the position information of the center point of the dynamic truth frame of the three-dimensional target object when the target and the data acquisition system move at different times.

6. The method for determining evaluation indicators for three-dimensional target detection according to claim 5, characterized in that, The process of calibrating the relative position using a laser tracker is as follows: Select a target sphere and ensure that each laser tracker can collect the laser signal of the target sphere. Obtain the position of the target sphere on each laser tracker, and then calculate the relative position between the laser trackers to complete the system calibration.

7. The method for determining evaluation indicators for three-dimensional target detection according to claim 6, characterized in that, The rectangle consists of six faces: top, bottom, left, right, front, and back. At least one face from each of the top, bottom, left, right, and front groups is selected as the dynamic truth box.

8. A method for determining evaluation indicators for three-dimensional target detection, characterized in that, Applications include scenarios where the target and data acquisition system move synchronously, including: B1. The three-dimensional target object (T) is inscribed within a rectangular frame; B2. Select at least three faces within the rectangular frame as the dynamic truth box of the 3D target object (T); B3. Fix an antenna at the center of each of the three truth boxes of the three-dimensional target (T), and connect one antenna to each mobile station RTK. B4. Move the three-dimensional target object (T) and record the measurement data and measurement time of each rover station's RTK. B5. Obtain the real-time three-dimensional position of the antenna measured by RTK; B6. Based on the real-time three-dimensional position information of the antenna in the same coordinate system, obtain the position information of the center point of the three-dimensional target truth frame when the target and the data acquisition system move synchronously.

9. The method for determining evaluation indicators for three-dimensional target detection according to claim 8, characterized in that, The rectangle consists of six faces: top, bottom, left, right, front, and back. At least one face from each of the top, bottom, left, right, and front groups is selected as one of the three dynamic truth boxes.

Images