CN-122017859-A - Multi-target ranging method and system based on AI calibration

CN122017859ACN 122017859 ACN122017859 ACN 122017859ACN-122017859-A

Abstract

The invention provides a multi-target ranging method and system based on AI calibration, comprising the following steps of collecting a current scene image, carrying out 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, carrying out laser ranging on one target in the current scene image based on a laser ranging unit and outputting a laser ranging result, and correcting the ranging results of the laser ranging target and targets in the same type as the laser ranging target if a ranging result difference delta d between the laser ranging result of the laser ranging target and the AI ranging result is less than a distance threshold value. According to the invention, only the targets in the same image are required to be subjected to laser ranging, and the ranging AI ranging results of a plurality of similar targets in the same image can be corrected, so that the influence of AI ranging errors on multi-target ranging is reduced, and the accuracy of the multi-target ranging results can be remarkably improved.

Inventors

HUANG CHENG
LI MENG
XIONG XIAOYI

Assignees

武汉高德智感科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251219

Claims (10)

1. The multi-target ranging method based on AI calibration is characterized by comprising the following steps: Collecting a current scene image; performing target recognition 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 one target in the current scene image based on the laser ranging unit, and outputting a laser ranging result of the laser ranging target; and if the difference delta d of the ranging results between the laser ranging results and the AI ranging results of the laser ranging targets is smaller than the distance threshold value, correcting the ranging results of the laser ranging targets and targets in the same category as the laser ranging targets based on the distance correction model.
2. The multi-target ranging method of claim 1, wherein if a ranging result difference Δd between a laser ranging result and an AI ranging result of the laser ranging target is greater than or equal to a distance threshold, outputting the laser ranging result of the laser ranging target and the AI ranging result of the target of the same class as the laser ranging target.
3. The multi-target ranging method of claim 1, wherein the distance correction model is as follows: ; Dr i _c is the correction distance of the ith image target, w1, w2, w3 and w4 are all weight parameters, b is a bias term, dai i is the AI ranging result of the ith image target, de i is the difference between the AI ranging result and the ranging result of the target where the laser beam spot is located in the target image where the ith image target is located, and SU i 、SV i is the transverse offset and the longitudinal offset of the target where the laser beam spot is located in the target image where the ith image target is located relative to the ith image target respectively.
4. A multi-target ranging method as claimed in claim 3 wherein the distance correction model is obtained by: Setting the targets of the same category on sites with different distances from the laser ranging unit, and setting at least 2 targets on each site; Controlling the laser ranging unit to output laser beams, and enabling light spots of the laser beams to fall on one target of each site in sequence; When the light spot of the laser beam falls on a target each time, acquiring a target image of a site where the target is located, wherein the target image comprises the target where the light spot is located and at least one target where the non-light spot is located, which are arranged at the same site; The distance correction model is trained based on a distance correction dataset p= { P 1 , p 2 , ..., p k } and distance correction data P i = (De i , Dai i , SU i , SV i ).
5. The multi-target ranging method of claim 4, wherein the distance correction model is trained with a mean square error MSE as a loss function, and there are: ; Wherein Dr i is the true distance between the i-th image target and the laser ranging unit, and Dr i _c is the corrected distance of the i-th image target output through the distance correction model.
6. The multi-target ranging method of claim 4, wherein the target detection result includes a target detection frame, a target class, and a target segmentation mask for each target; and the laser ranging unit outputs that the light spot of the laser beam falls in the area of one of the object segmentation masks, and outputs the laser ranging result of the object so as to finish the laser ranging of the object.
7. The multi-target ranging method according to claim 2, wherein the targets are tracked based on the laser ranging targets and the corrected distances of targets of the same class as the laser ranging targets obtained when Δd < distance threshold, or the laser ranging results of the laser ranging targets and the AI ranging results of targets of the same class as the laser ranging targets output when Δd is greater than or equal to distance threshold.
8. The multi-target ranging method of claim 1, wherein correcting ranging results of the laser ranging target and the target of the same class as the laser ranging target based on the distance correction model comprises: Inputting distance correction data p Pl of the laser ranging target and distance correction data p Pe of the target in the same category as the laser ranging target into the distance correction model in the current scene image so as to correspondingly output correction distance Dr Pl _c of the laser ranging target and correction distance Dr Pe _c of target Pe in the same category as the laser ranging target; Distance correction data p Pl = (De Pl , Dai Pl , SU Pl , SV Pl of the laser ranging target), distance correction data p Pe = (De Pe , Dai Pe , SU Pe , SV Pe of the target Pe of the same type as the laser ranging target Pl; Dai Pl is the AI ranging result of the laser ranging target, de Pl is the AI ranging result of the laser ranging target and the difference value of the ranging results between the AI ranging results of the laser ranging target and the laser ranging results, SU Pl 、SV Pl is the lateral offset and the longitudinal offset of the target where the laser beam spot is located in the current scene image of the laser ranging target corresponding to the laser ranging target, and SU Pl = SV Pl =0; Dai Pe is the AI ranging result of the same class of targets as the laser ranging targets, de Pe is the difference of the ranging results of the same class of targets as the laser ranging targets, and De Pe =De Pl ;SU Pe 、SV Pe corresponds to the lateral offset and the longitudinal offset of the same class of targets as the laser ranging targets relative to the laser ranging targets respectively.
9. The multi-target ranging method of claim 1, wherein the current scene image is subject to target recognition based on a semantic segmentation model, and a target detection result is output.
10. A multi-target ranging system, comprising: the imaging equipment is used for acquiring a plurality of frames of current scene images in real time; A target recognition unit, which is used for carrying out target recognition on the current scene image; An AI ranging result acquisition unit that outputs an AI ranging result for each target in the current scene image based on the target detection result; The laser ranging unit is used for outputting a laser beam, carrying out laser ranging on one target in the current scene image through the laser beam, and outputting a laser ranging result of the laser ranging target; The distance measurement result correction unit is used for correcting the distance measurement results of the laser distance measurement targets and targets in the same category as the laser distance measurement targets according to the distance correction model when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distance measurement targets is smaller than the distance threshold; The distance measurement result output unit is used for outputting the correction distance of each target in the targets in the same category when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distance measurement targets is smaller than the distance threshold value, and outputting the laser distance measurement results of the laser distance measurement targets and the AI distance measurement results of the targets in the same category as the laser distance measurement targets when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distance measurement targets Pl is larger than or equal to the distance threshold value.

Description

Multi-target ranging method and system based on AI calibration Technical Field The invention relates to the technical field of ranging, in particular to a multi-target ranging method and system based on AI calibration. Background The laser range finder is an instrument for accurately measuring the distance of a target by utilizing laser, and has been widely applied to various task scenes, and the main working principle is that the laser range finder emits a laser beam to the target, the laser beam reflected by the target is received by a photoelectric element, and 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 an observer to the target. However, the laser range finder can only obtain one target ranging result at a time, and if other target distances need to be measured, the laser landing point needs to be adjusted, so that the laser ranging efficiency is low. In addition, in the prior art, there is a technical scheme of detecting image targets based on an AI algorithm and estimating target distances, for example, detecting positions of the targets by a deep learning image algorithm, and further obtaining distances of each target by a monocular ranging algorithm, so that a plurality of targets can be simultaneously measured, but the method is limited by the detection precision of a model, if the detection precision of the model is insufficient, the final ranging result precision is low, and the method cannot be used. Disclosure of Invention The invention aims to provide a multi-target ranging method and system based on AI calibration, which only need to perform laser ranging on targets in the same image, further correct ranging AI ranging results of a plurality of similar targets in the same image according to the laser ranging results, the AI ranging results and a distance correction model of the targets where light spots are located, so as to reduce the influence of AI ranging errors on multi-target ranging, and remarkably improve the precision of the multi-target ranging results. In order to achieve the above purpose, the present invention provides the following technical solutions: In one aspect, there is provided a multi-target ranging method based on AI calibration, including the steps of: Collecting a current scene image; performing target recognition 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 one target in the current scene image based on the laser ranging unit, and outputting a laser ranging result of the laser ranging target; and if the difference delta d of the ranging results between the laser ranging results and the AI ranging results of the laser ranging targets is smaller than the distance threshold value, correcting the ranging results of the laser ranging targets and targets in the same category as the laser ranging targets based on the distance correction model. In another aspect, there is also provided a multi-target ranging system, comprising: the imaging equipment is used for acquiring a plurality of frames of current scene images in real time; A target recognition unit, which is used for carrying out target recognition on the current scene image; An AI ranging result acquisition unit that outputs an AI ranging result for each target in the current scene image based on the target detection result; The laser ranging unit is used for outputting a laser beam, carrying out laser ranging on one target in the current scene image through the laser beam, and outputting a laser ranging result of the laser ranging target; The distance measurement result correction unit is used for correcting the distance measurement results of the laser distance measurement targets and targets in the same category as the laser distance measurement targets according to the distance correction model when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distance measurement targets is smaller than the distance threshold; The distance measurement result output unit is used for outputting the correction distance of each target in the targets in the same category when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distance measurement targets is smaller than the distance threshold value, and outputting the laser distance measurement results of the laser distance measurement targets and the AI distance measurement results of the targets in the same category as the laser distance measurement targets when the distance measurement result difference delta d between the laser distance measurement results and the AI distance measurement results of the laser distanc