Multi-target ranging method and system based on AI calibration

Abstract

The invention provides a multi-target distance measurement method and system based on AI calibration. The method comprises the following steps: collecting a current scene image; performing target identification on the current scene image, and outputting a target detection result; outputting an AI ranging result of each target in the current scene image based on the target detection result; performing laser ranging on a target in the current scene image based on a laser ranging unit, and outputting a laser ranging result; and if the distance measurement result difference delta d between the laser distance measurement result of the laser distance measurement target and the AI distance measurement result is less than the distance threshold value, correcting the distance measurement results of the laser distance measurement target and the target of the same type as the laser distance measurement target. According to the method, the AI ranging results of multiple targets of the same type in the same image can be corrected only by performing laser ranging on the targets in the same image, so that the influence of AI ranging errors on multi-target ranging is reduced, and the precision of the multi-target ranging results can be remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of ranging technology, and in particular to a multi-target ranging method and system based on AI calibration. Background Technology

[0002] A laser rangefinder is an instrument that uses laser light to accurately measure the distance to a target. It has been widely used in various mission scenarios. Its main working principle is as follows: the laser rangefinder emits a laser beam towards the target. The laser beam reflected by the target is received by a photoelectric element. A timer measures the time from the emission of the laser beam to the reception of the reflected laser beam, thereby calculating the distance from the observer to the target.

[0003] However, the aforementioned laser rangefinder can only obtain the distance measurement result of one target at a time. If it is necessary to measure the distance of other targets, the laser point of contact needs to be adjusted, resulting in low laser ranging efficiency.

[0004] In addition, existing technologies include AI-based solutions for detecting targets in images and estimating their distances. For example, deep learning image algorithms can be used to detect the location of targets, and then monocular ranging algorithms can be used to obtain the distance to each target. This allows for simultaneous ranging of multiple targets. However, this method is limited by the detection accuracy of the model. If the model's detection accuracy is insufficient, the final ranging result will be inaccurate and unusable. Summary of the Invention

[0005] The purpose of this invention is to provide a multi-target ranging method and system based on AI calibration. It only requires laser ranging of targets in the same image. Furthermore, based on the laser ranging result of the target where the light spot is located, the AI ​​ranging result, and the distance correction model, the ranging AI ranging results of multiple similar targets in the same image can be corrected to reduce the impact of AI ranging error on multi-target ranging and significantly improve the accuracy of multi-target ranging results.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] On the one hand, an AI-calibrated multi-target ranging method is provided, which includes the following steps:

[0008] Acquire images of the current scene;

[0009] Perform target recognition on the current scene image and output the target detection results;

[0010] Based on the target detection results, the AI ​​ranging results for each target in the current scene image are output;

[0011] The laser ranging unit performs laser ranging on a target in the current scene image and outputs the laser ranging result of the target.

[0012] If the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target is less than the distance threshold, then the ranging results of the laser ranging target and targets of the same category as the laser ranging target are corrected based on the distance correction model.

[0013] On the other hand, a multi-target ranging system is also provided, which includes:

[0014] An imaging device used to acquire several frames of images of the current scene in real time;

[0015] A target recognition unit is used to perform target recognition on the current scene image;

[0016] The AI ​​ranging result acquisition unit outputs the AI ​​ranging result for each target in the current scene image based on the target detection result;

[0017] A laser ranging unit is used to output a laser beam, and to perform laser ranging on a target in the current scene image through the laser beam, and output the laser ranging result of the laser ranging target;

[0018] The ranging result correction unit is used to correct the ranging results of the laser ranging target and the AI ​​ranging target when the difference Δd between the ranging results of the laser ranging target and the AI ​​ranging result of the laser ranging target is less than the distance threshold, according to the distance correction model.

[0019] The ranging result output unit is used to output the corrected distance of each target in the same category when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target is less than the distance threshold, and to output the laser ranging result of the laser ranging target and the AI ​​ranging result of the target in the same category when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target Pl is greater than or equal to the distance threshold.

[0020] In summary, the present invention has the following advantages compared with the prior art:

[0021] The present invention only requires laser ranging of targets in the same image. Furthermore, based on the laser ranging result of the target where the spot is located, the AI ​​ranging result, and the distance correction model, the ranging AI ranging result of multiple similar targets in the same image can be corrected. Through the error correction mechanism, the influence of AI ranging error on multi-target ranging can be reduced, and the accuracy of multi-target ranging result can be significantly improved. Attached Figure Description

[0022] Figure 1 This is a flowchart illustrating the steps of the AI-calibrated multi-target ranging method in this invention.

[0023] Figure 2 This is a schematic diagram showing the layout of the laser ranging unit, imaging device, and point targets in this invention;

[0024] Figure 3 This refers to the current scene image and the target segmentation mask in this invention;

[0025] Figure 4 This is a schematic diagram of the multi-target ranging system in this invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] Example 1

[0028] Figure 1 As shown, this embodiment provides a multi-target ranging method based on AI calibration, which includes the following steps:

[0029] S1. Construct a semantic segmentation model and a distance correction model, and train the semantic segmentation model and the distance correction model;

[0030] In this embodiment, the semantic segmentation model can assign each pixel in the image to a corresponding category in order to detect the target from the image. For example, the semantic segmentation model includes the Mask R-CNN algorithm, the YOLO series of algorithms, etc.

[0031] Furthermore, training the semantic segmentation model includes the following steps: acquiring images of a specified target scene through an imaging device 100 (such as a camera), performing semantic segmentation annotation on the images, training the model based on a loss function, and outputting the model after satisfying the iterative convergence condition, so as to use it for subsequent detection and recognition of targets in the images.

[0032] The distance correction model is obtained through the following steps:

[0033] S11. Construct the training dataset, which includes the following steps:

[0034] A laser ranging unit 200 is set at a predetermined location, and targets of the same type are set at locations at different distances from the laser ranging unit 200, with at least two targets set at each location; in this embodiment, there are a total of n targets, namely target 1, target 2, ..., target n, which are respectively set at locations p1, p2, ..., pm at different distances from the laser ranging unit 200, and at least two targets of the same type are set at each location, for example, target 1, target 2, and target 3 are set at location p1, and target 4, target 5, etc. are set at location p2;

[0035] The laser ranging unit 200 is controlled to output a laser beam. While keeping the position of the laser ranging unit 200 unchanged, the laser beam output angle is changed so that the laser beam spot falls on one target at each point in sequence. That is, the laser beam spot falls on one target at the same point in sequence, such as target 1 at point p1, target 4 at point p2, etc.

[0036] Each time the laser beam spot falls on a target (that is, the target where the spot is located), the imaging device 100 acquires a target image of the target location. The target image contains the target where the spot is located and at least one non-spot target located at the same location. Targets appearing in the target image are recorded as image targets (i.e., targets appearing in the target image, including the target where the spot is located and non-spot targets). The non-spot target refers to other targets at the same location in the current target image besides the target where the spot is located. The imaging device 100 and the laser ranging unit 200 are fixed in their installation positions and their optical axes are parallel to each other.

[0037] For example, if the laser beam spot currently falls on target 1 at point p1, and the target image at point p1 contains target 1 and target 2 but not target 3, then target 1 is the target where the spot is located, and target 2 is the target not where the spot is located. Since both target 1 and target 2 appear in the target image, they are both recorded as image targets. Target 3 does not appear in the target image and is not recorded as an image target. Similarly, if the target image at point p2 contains target 4 and target 5, and the spot falls on target 5, then both target 4 and target 5 are recorded as image targets. Target 5 is the target where the spot is located, and target 4 is the target not where the spot is located. Thus, in this embodiment, at least one target image can be obtained by repeatedly illuminating different targets at each point with the laser beam, and in each target image, the center of the spot coincides with the center of the image.

[0038] The laser ranging unit 200 outputs the laser ranging result of the target where the light spot is located in each target image, for example, outputs the laser ranging results of target 1 and target 5;

[0039] For each target image, obtain the AI ​​ranging result, laser ranging result, and target detection box of the target where the light spot is located, as well as the AI ​​ranging result and target detection box of all targets that are not the light spot in each target image; among them, obtaining the AI ​​ranging result of the target where the light spot is located based on the target image is existing technology and will not be described in detail.

[0040] Construct a distance-corrected dataset P = {p1, p2, ..., p} k}, to be used as the training dataset; where the distance correction data p i = (De i Dai i SU i , SV i ), where i represents the i-th image target, and i∈[1,k], k is the total number of all image targets, and k≤n, where n is the total number of all targets; Dai i The AI ​​ranging result for the i-th image target; De i SU represents the difference between the AI ​​ranging result and the laser ranging result of the target containing the light spot in the target image where the i-th image target is located; i SV i These correspond to the horizontal and vertical offsets of the target where the light spot is located in the target image relative to the i-th image target. The horizontal and vertical offsets can be determined based on the position of the target detection box.

[0041] For example, if the target image at point p1 contains target 1 and target 2, and target 1 is the target where the light spot is located, and target 2 is the target where the light spot is not located, then the corresponding outputs are the AI ​​ranging result Dai1, the laser ranging result DL1 and the target detection box Box1 for target 1, and the AI ​​ranging result Dai2 and the target detection box Box2 for target 2.

[0042] Furthermore, based on the AI ​​ranging result Dai1 and the laser ranging result DL1 of target 1, the ranging result difference De1 is obtained. Since the target detection box Box1 does not shift laterally or longitudinally relative to the target where the light spot is located, i.e., target 1 itself, the lateral offset SU1 and the longitudinal offset SV1 of target 1 are both 0. Assuming Dai1 and DL1 are 99.8m and 98m respectively, then De1 is 1.8m. Therefore, the distance correction data of target 1 is recorded as p1 = (De1, Dai1, SU1, SV1) = (1.8, 99.8, 0, 0).

[0043] Based on this, the distance difference De2 of target 2 is directly taken from the distance difference De1 of target 1, that is, De2 = De1. Then, based on the position of the target detection box Box2 of target 2 and the target detection box Box1 of target 1, the horizontal offset SU2 and vertical offset SV2 of target detection box Box2 relative to target detection box Box1 are determined. For example, the coordinates of the midpoint of the bottom edge of target detection box Box2 and the midpoint of the bottom edge of target detection box Box1 can be determined first. The horizontal difference and vertical difference of the two bottom edge center coordinates can be used as the horizontal offset SU2 and vertical offset SV2 respectively. Assuming Dai2 is 101 meters, the horizontal offset SU2 and vertical offset SV2 are 60px and 3px respectively. Then, the distance correction data of target 2 is recorded as p2 = (De2, Dai2, SU2, SV2) = (1.8, 101, 60, 3).

[0044] This allows us to obtain the distance correction data p for each image target;

[0045] S12. Construct a distance correction model and train the distance correction model based on the training dataset. The distance correction model is as follows:

[0046] ;

[0047] Among them, Dr i _c is the corrected distance of the i-th image target; w1, w2, w3, and w4 are all weight parameters, and b is the bias term;

[0048] Furthermore, in this embodiment, the weight parameters w1, w2, w3, and w4 are optimized using the backpropagation algorithm, and the distance correction model is trained using a loss function to improve the predicted value (i.e., the corrected distance Dr). i _c) Minimize the error between the actual distance and the true distance to improve the model's prediction accuracy. In this embodiment, the mean squared error (MSE) is used as the loss function, and:

[0049] ;

[0050] Among them, Dr i The true distance between the i-th image target and the laser ranging unit 200 can be determined by means of measuring with a tape measure, etc.

[0051] S2, Acquire several frames of the current scene image in real time using imaging device 100 (i.e., Figure 3 As shown in part (a) of the image, the current scene image is preprocessed.

[0052] In this embodiment, the imaging device 100 can be a fixed installation structure, such as a fixed camera, or it can be mounted on unmanned equipment such as drones or unmanned vehicles. The image preprocessing includes one or more of the following operations: image denoising (which can be implemented based on median filtering, mean filtering, Gaussian filtering, wavelet denoising, etc.), image enhancement, and normalization, in order to reduce image noise and improve the target recognition accuracy of the subsequent semantic segmentation model.

[0053] S3. Based on the trained semantic segmentation model, perform target recognition on the current scene image and output the target detection result. Based on the target detection result, output the AI ​​ranging result of each target in the current scene image.

[0054] Furthermore, the laser ranging unit 200 performs laser ranging on a target in the current scene image, which is denoted as laser ranging target P1, and outputs the laser ranging result of laser ranging target P1;

[0055] The target detection results include the target detection bounding box, target category, and target segmentation mask (i.e., the target detection box, target category, and target segmentation mask) for each target. Figure 3 As shown in part (b) of the document, the laser ranging unit and the image acquisition device can be installed on the same platform, such as on the same UAV. Therefore, this embodiment can accurately detect targets in the image through a semantic segmentation model to identify the category and location of all targets in the image and output a pixel-level target mask as a correlation factor for subsequent laser ranging. Furthermore, the laser ranging unit 200 outputs a laser beam, and the spot of the laser beam falls within the area of ​​one of the target segmentation masks (i.e., the target segmentation mask of the laser ranging target P1), and outputs the laser ranging result of the target to complete the laser ranging of the target.

[0056] S4. Obtain the distance difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target P1. If the distance difference Δd < the distance threshold, proceed to step S5. If the distance difference Δd ≥ the distance threshold, proceed to step S6. The distance threshold can be determined based on factors such as the distance accuracy requirements.

[0057] S5. Correct the distance data p of the laser ranging target Pl in the current scene image. Pl And distance correction data p for targets Pe of the same category as the laser ranging target Pl. Pe All are input into the distance correction model to correspond to the corrected distance Dr of the laser ranging target Pl. Pl _c and the corrected distance Dr of target Pe, which is of the same category as laser ranging target Pl. Pe _c, to be used as the ranging result for each target in the same category;

[0058] Among them, the distance correction data p of the laser ranging target Pl Pl = (De Pl Dai Pl SU Pl , SV Pl The distance correction data p of a target Pe of the same category as the laser ranging target Pl. Pe = (De Pe Dai Pe SU Pe , SV Pe );

[0059] Dai Pl The AI ​​ranging result for laser ranging target Pl; De Pl De represents the difference between the AI ​​ranging result and the laser ranging result for the laser ranging target Pl. Pl =Δd;SU Pl SV Pl These correspond to the lateral and longitudinal offsets of the laser ranging target Pl relative to the target containing the light spot in the current scene image, respectively. Since the target containing the light spot is the laser ranging target Pl, both the lateral and longitudinal offsets of the laser ranging target Pl relative to itself are 0, i.e., SU. Pl =SV Pl =0;

[0060] Dai Pe AI ranging results for target Pe, which is of the same category as laser ranging target Pl; De Pe Let De be the difference in ranging results between a target Pe and a target Pe of the same category as the laser ranging target Pl, and De Pe =De Pl ;SU Pe SV Pe These correspond to the lateral and longitudinal offsets of a target Pe, which is of the same category as the laser ranging target Pl, relative to the laser ranging target Pl. These offsets can also be determined based on the position of the target detection box.

[0061] S6, directly output the laser ranging result of laser ranging target Pl and the AI ​​ranging result of target Pe of the same category as laser ranging target Pl, as the ranging result of each target in the same category;

[0062] For example, if targets A and B of the same category can be detected in the current scene image, and the laser spot of the laser ranging unit 200 falls on target B, then target B is the laser ranging target. The distance difference Δd (e.g., 1.8m) between the laser ranging result (e.g., 99.8m) and the AI ​​ranging result (e.g., 98m) of target B is calculated, along with the lateral offset SU of target A relative to the laser ranging target B. A Vertical offset SV A , such as SU A =60px, SV A =3px;

[0063] If Δd < distance threshold, such as 2.0m, it indicates that the errors in laser ranging and AI ranging results are small, and the AI ​​ranging result is relatively accurate. It is necessary to further correct the ranging results using a distance correction model. Therefore, the distance correction data p of target B of the same category should be used. B = (1.8, 98, 0, 0), Distance correction data p for target A A Inputting (1.8, 101, 60, 3) into the distance correction model above yields the corresponding corrected distances Dr for targets A and B. A _c、Dr B _c, to be used as the ranging results for targets A and B respectively;

[0064] If the distance difference Δd between the laser ranging result of 99.8m and the AI ​​ranging result of 97.5m for target B is 2.3m, then Δd > distance threshold of 2.0m. This indicates that the laser ranging result and the AI ​​ranging result have large errors, and the AI ​​ranging result has a large deviation. Therefore, there is no need to further correct the ranging result through the distance correction model. Instead, the laser ranging result of 99.8m for target B and the AI ​​ranging result of 101m for target A, which is of the same category as target B, are directly output as the ranging results for targets A and B, respectively.

[0065] Therefore, by combining the advantages of laser ranging and AI ranging, this embodiment only needs to perform laser ranging on targets in the same image to correct the ranging and AI ranging results of multiple similar targets in the same image. Through the error correction mechanism, the influence of AI ranging error on multi-target ranging can be reduced, and the accuracy of multi-target ranging results can be significantly improved.

[0066] Example 2:

[0067] The only difference between this embodiment and Embodiment 1 is that the multi-target ranging method further includes step S7: tracking the target based on the ranging result of each target of the same category output in step S5 or S6, which includes the following steps:

[0068] Input the ranging results of the same target in the same category in the previous frame image and the current frame image into the target tracking model, as well as the target detection box of the target. The ranging result of the target is the ranging result output in step S5 or S6, that is, the corrected distance of the laser ranging target P1 and the target Pe of the same category as the laser ranging target P1 in step S5, or the laser ranging result of the laser ranging target P1 and the AI ​​ranging result of the target Pe of the same category as the laser ranging target P1 in step S6.

[0069] The target tracking model compares the ranging results of the same target in the previous frame image and the current frame image with the target detection box of the target, and performs data fitting processing based on smoothing filtering algorithms to output the processed target ranging result as the final target ranging result, making the target ranging result more stable and accurate, and reducing the fluctuation range of the ranging result.

[0070] Example 3:

[0071] This embodiment provides a multi-target ranging system that can implement the AI-calibrated multi-target ranging method described in Embodiment 1 or 2, such as... Figure 4 As shown, the multi-target ranging system includes:

[0072] Imaging device 100 is used to acquire several frames of current scene images in real time and perform image preprocessing on the current scene images. The preprocessing process is the same as step S2.

[0073] The target recognition unit 300 performs target recognition on the current scene image based on a trained semantic segmentation model and outputs the target detection result;

[0074] AI ranging result acquisition unit 400 outputs AI ranging results for each target in the current scene image based on the target detection results;

[0075] The laser ranging unit 200 is used to output a laser beam and perform laser ranging on a target in the current scene image through the laser beam. The target is denoted as laser ranging target P1, and the laser ranging result of the laser ranging target P1 is output.

[0076] The ranging result correction unit 500 is used to correct the ranging results of the laser ranging target P1 and the AI ​​ranging result when the difference Δd between the ranging results of the laser ranging target P1 and the AI ​​ranging result is less than the distance threshold, according to the distance correction model, so as to obtain the corrected distance of each target in the same category. The process is the same as step S5.

[0077] The ranging result output unit 600 is used to output the corrected distance of each target in the same category when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target P1 is less than the distance threshold, and to output the laser ranging result of the laser ranging target P1 and the AI ​​ranging result of the target Pe in the same category as the laser ranging target P1 when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target P1 is greater than or equal to the distance threshold.

[0078] In summary, this invention combines the advantages of laser ranging and AI ranging. It only requires laser ranging of targets in the same image. Furthermore, based on the laser ranging results, AI ranging results, and distance correction model of the target where the light spot is located, the ranging and AI ranging results of multiple similar targets in the same image can be corrected. Through the error correction mechanism, the impact of AI ranging error on multi-target ranging can be reduced, and the accuracy of multi-target ranging results can be significantly improved.

[0079] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-target ranging method based on AI calibration, characterized in that, Includes the following steps: Acquire images of the current scene; Perform target recognition on the current scene image and output the target detection results; Based on the target detection results, the AI ​​ranging results for each target in the current scene image are output; The laser ranging unit performs laser ranging on a target in the current scene image and outputs the laser ranging result of the target. If the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target is less than the distance threshold, then the ranging results of the laser ranging target and targets of the same category as the laser ranging target are corrected based on the distance correction model.

2. The multi-target ranging method as described in claim 1, characterized in that, If the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target is greater than or equal to the distance threshold, then the laser ranging result of the laser ranging target and the AI ​​ranging result of the target of the same category as the laser ranging target will be output.

3. The multi-target ranging method as described in claim 1, characterized in that, The distance correction model is as follows: ； Among them, Dr i _c represents the corrected distance of the i-th image target; w1, w2, w3, and w4 are weight parameters, and b is the bias term; Dai i The AI ​​ranging result for the i-th image target; De i The difference between the AI ​​ranging result and the laser ranging result for the target containing the laser beam spot in the target image where the i-th image target is located; SU i SV i These correspond to the lateral and longitudinal offsets of the laser beam spot relative to the target image in which the i-th image target is located.

4. The multi-target ranging method as described in claim 3, characterized in that, The distance correction model is obtained through the following steps: Targets of the same type are placed at locations at different distances from the laser ranging unit, with at least two targets placed at each location; Control the laser ranging unit to output a laser beam, and make the laser beam spot fall on one of the targets at each point in sequence; Each time the laser beam spot falls on a target, a target image of the location of that target is acquired, and the target image contains the target where the spot is located and at least one target that is not the spot located at the same location; Based on the distance correction dataset P = {p1, p2, ..., p k The distance correction model is trained, and the distance correction data p i = (De i Dai i SU i , SV i ).

5. The multi-target ranging method as described in claim 4, characterized in that, The distance correction model is trained using the mean squared error (MSE) as the loss function, and the following holds: ； Among them, Dr i Dr represents the true distance between the i-th image target and the laser ranging unit. i _c is the corrected distance of the i-th image target output by the distance correction model.

6. The multi-target ranging method as described in claim 4, characterized in that, The target detection results include the target detection bounding box, target category, and target segmentation mask for each target; Furthermore, the laser beam output by the laser ranging unit falls within the area of ​​one of the target segmentation masks and outputs the laser ranging result of that target to complete the laser ranging of that target.

7. The multi-target ranging method as described in claim 2, characterized in that, The target is tracked based on the laser ranging target obtained when Δd < distance threshold and the corrected distance of the target and the target of the same category as the laser ranging target; or, when Δd ≥ distance threshold, the laser ranging result of the laser ranging target and the AI ​​ranging result of the target of the same category as the laser ranging target are output.

8. The multi-target ranging method as described in claim 1, characterized in that, The ranging results for laser ranging targets and targets of the same category as laser ranging targets are corrected based on a distance correction model, including: The distance correction data p of the laser ranging target in the current scene image. Pl And distance correction data p for targets of the same category as laser ranging targets. Pe All are input into the distance correction model to output the corrected distance Dr of the laser ranging target. Pl _c and the corrected distance Dr of Pe, which is of the same category as the laser ranging target. Pe _c; Among them, the distance correction data p of the laser ranging target Pl = (De Pl Dai Pl SU Pl , SV Pl The distance correction data p of a target Pe of the same category as the laser ranging target Pl. Pe = (De Pe Dai Pe SU Pe , SV Pe ); Dai Pl AI ranging results for laser ranging targets; De Pl AI ranging results for laser ranging targets, and the difference between laser ranging results; SU Pl SV Pl These correspond to the lateral and longitudinal offsets of the laser ranging target relative to the target within the current scene image, where the laser beam spot is located. Pl =SV Pl =0; Dai Pe AI ranging results for targets of the same category as laser ranging targets; De Pe De represents the difference in ranging results between targets of the same category as the laser ranging target, and De Pe =De Pl ;SU Pe SV Pe These correspond to the lateral and longitudinal offsets of a target of the same category as the laser ranging target relative to the laser ranging target, respectively.

9. The multi-target ranging method as described in claim 1, characterized in that, The target is identified in the current scene image based on the semantic segmentation model, and the target detection result is output.

10. A multi-target ranging system, characterized in that, include: An imaging device used to acquire several frames of images of the current scene in real time; A target recognition unit, used to perform target recognition on the current scene image; The AI ​​ranging result acquisition unit outputs the AI ​​ranging result for each target in the current scene image based on the target detection result; A laser ranging unit is used to output a laser beam, and to perform laser ranging on a target in the current scene image through the laser beam, and output the laser ranging result of the laser ranging target; The ranging result correction unit is used to correct the ranging results of the laser ranging target and the AI ​​ranging target when the difference Δd between the ranging results of the laser ranging target and the AI ​​ranging result of the laser ranging target is less than the distance threshold, according to the distance correction model. The ranging result output unit is used to output the corrected distance of each target in the same category when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target is less than the distance threshold, and to output the laser ranging result of the laser ranging target and the AI ​​ranging result of the target in the same category when the difference Δd between the laser ranging result and the AI ​​ranging result of the laser ranging target Pl is greater than or equal to the distance threshold.

Images