A dynamic object suppression visual odometry method, system, and medium

Abstract

The application relates to a dynamic object suppression visual odometry method, system and medium, comprising the following steps: acquiring two continuous input image frames, inputting a dynamic object suppression visual odometry model, and obtaining camera pose changes; wherein the dynamic object suppression visual odometry model comprises a matching network, a dynamic detection network and a pose estimation network connected in sequence, the matching network is used for estimating forward optical flow and backward optical flow, performing optical flow consistency detection according to the forward optical flow and the backward optical flow, and obtaining forward optical flow consistency and backward optical flow consistency; the dynamic detection network generates a probability graph based on the forward optical flow consistency and the backward optical flow consistency, converts the probability graph into a binary segmentation mask to optimize the forward optical flow; and the pose estimation network is used for outputting the camera pose changes according to the optimized forward optical flow. Compared with the prior art, the application can further improve the accuracy and reliability of visual odometry in a dynamic environment.

Description

Technical Field

[0001] This invention belongs to the field of computer vision technology, and in particular relates to a visual odometry measurement method, system and medium for dynamic object suppression. Background Technology

[0002] Visual odometry (VO) refers to the process of calculating camera motion trajectories by analyzing visual data streams, and it is widely used in various fields such as 3D reconstruction, autonomous driving, and virtual reality. Currently, research on monocular visual odometry methods based on deep learning is relatively mature and has achieved excellent performance. However, visual odometry is generally based on static assumptions. In dynamic scenes, camera motion and dynamic objects in the scene jointly cause optical flow. Many existing visual odometry methods directly use optical flow maps as visual odometry outputs, which may lead to feature point matching errors, thus affecting the accurate estimation of camera pose. To address this problem, the mainstream solution is to use semantic or motion information to identify dynamic object regions in the video and then reduce the contribution weight of dynamic object regions in pose updates. Chinese invention patent application number 2019110573430 discloses a visual odometry method based on dynamic and static scene separation. Through inter-frame change analysis, it utilizes the inconsistency of depth information and optical flow information in dynamic scenes to achieve dynamic and static scene separation. Based on this, it uses more reliable static scene information to achieve camera pose estimation. However, the above methods still have some problems: On the one hand, although the method can achieve separation of dynamic and static scenes through inter-frame change analysis, the detection and segmentation of dynamic objects may not be accurate enough in extremely dynamic or complex scenes, which affects the extraction of static scene information and thus the accuracy of camera pose estimation; on the other hand, the method relies on deep convolutional neural networks for depth estimation, camera pose estimation, and dynamic object detection. These computational processes may be relatively complex and have high requirements for computing resources. When running on resource-constrained devices, it may be difficult to guarantee real-time performance. Therefore, it is necessary to redesign a visual odometry method to further improve the accuracy and reliability of visual odometry. Summary of the Invention

[0003] The purpose of this invention is to overcome the shortcomings of the prior art by providing a visual odometry method, system, and medium for dynamic object suppression, thereby further improving the accuracy and reliability of visual odometry in dynamic environments.

[0004] The objective of this invention can be achieved through the following technical solutions:

[0005] This invention provides a visual odometry measurement method for dynamic object suppression, comprising the following steps:

[0006] Two consecutive input image frames are acquired, and a visual odometry model for dynamic object suppression is input to obtain camera pose changes, including camera rotation R and camera displacement T.

[0007] The visual odometry model for dynamic object suppression includes a matching network, a dynamic detection network, and a pose estimation network connected in sequence. The specific process of the visual odometry model for dynamic object suppression processing input data is as follows:

[0008] Two consecutive input image frames I t and I t+1 Input the matching network to estimate the forward optical flow. and backward optical flow

[0009] According to the forward optical flow and the backward optical flow Optical flow consistency detection is performed, including forward optical flow consistency detection and backward optical flow consistency detection. The forward optical flow consistency is obtained based on the results of the forward and backward optical flow consistency detections. Backward consistency with optical flow These are used to represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively.

[0010] The forward consistency of the optical flow and the backward consistency of the optical flow Input the dynamic detection network to obtain the probability graph. The probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Through the binary segmentation mask Optimize the forward optical flow To remove the forward optical flow Dynamic regions within;

[0011] The binary segmentation mask Optimized forward optical flow The pose estimation network is input along with the intrinsic parameter layer to obtain the camera pose change.

[0012] Furthermore, the matching network obtains the forward optical flow through GMFlow estimation. and the backward optical flow

[0013] Furthermore, the forward optical flow consistency detection process is as follows: using the forward optical flow... For the backward optical flow Perform deformation, and then the forward optical flow If the summation exceeds a set threshold, the forward optical flow consistency detection result is considered inconsistent.

[0014] Furthermore, the specific formula for calculating the set threshold is as follows:

[0015] threshold = α * flow_mag + β

[0016] Where threshold is the set threshold, α and β are threshold influence parameters, and flow_mag is the optical flow magnitude.

[0017] Furthermore, the process of backward optical flow consistency detection is as follows: using the backward optical flow... For the forward optical flow Perform deformation, and then the backward optical flow If the summation exceeds a set threshold, the backward optical flow consistency detection result is considered inconsistent.

[0018] Furthermore, the dynamic detection network is built on the U-Net network architecture, including an encoder and a decoder, with skip connections between the encoder and the decoder.

[0019] Furthermore, the encoder employs a deep residual network, ResNet34.

[0020] Furthermore, the intrinsic parameter layer is constructed based on camera intrinsic parameters, which include focal length and principal point position.

[0021] The present invention also provides a visual odometry system for dynamic object suppression, comprising an optical flow estimation module, a dynamic detection module and a pose estimation module connected in sequence;

[0022] The optical flow estimation module is used to estimate the optical flow based on two consecutive input image frames I. t and I t+1 Estimate forward optical flow and backward optical flow and the forward optical flow and the backward optical flow Optical flow consistency detection is performed, including forward optical flow consistency detection and backward optical flow consistency detection. The forward optical flow consistency is obtained based on the results of the forward and backward optical flow consistency detections. Backward consistency with optical flow These are used to represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively.

[0023] The dynamic detection module is used to determine the forward consistency of the optical flow. and the backward consistency of the optical flow Generate probability map The probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Through the binary segmentation mask Optimize the forward optical flow To remove the forward optical flow Dynamic regions within;

[0024] The pose estimation module is used to estimate the position based on the binary segmentation mask. Optimized forward optical flow And intrinsic parameter layers, to obtain camera pose changes This includes camera rotation R and camera displacement T.

[0025] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0026] Compared with the prior art, the present invention has the following beneficial effects:

[0027] 1. This invention proposes a visual odometry method with dynamic object suppression, implemented through a visual odometry model with dynamic object suppression including a matching network, a dynamic detection network, and a pose estimation network. The specific process is as follows: First, two consecutive input image frames I... t and I t+1 Input the matching network to estimate the forward optical flow. and backward optical flow Forward optical flow reflects pixel displacement from the current frame to the next frame, while backward optical flow reflects pixel displacement from the current frame to the previous frame. Forward and backward optical flow form the basis for subsequent motion analysis. Secondly, regarding forward optical flow... and backward optical flow Perform optical flow consistency detection to obtain forward optical flow consistency. Backward consistency with optical flow These represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively, which can comprehensively reflect the state of pixels in an image frame; then, the forward consistency of optical flow is... Backward consistency with optical flow Input the dynamic detection network to obtain the probability graph. It can reflect the probability that each pixel in the input image frame is a dynamic pixel, identify dynamic objects, and generate a probability map. Convert to binary segmentation mask Using binary segmentation mask Optimize forward optical flow Forward optical flow can be removed The dynamic region in the middle; finally, the binary segmentation mask. Optimized forward optical flow Input the intrinsic parameter layer into the pose estimation network to obtain the camera pose change. The above methods can reduce the negative impact of dynamic objects on visual odometry, and further improve the accuracy and reliability of visual odometry in dynamic environments.

[0028] 2. In this invention, the dynamic detection network is built based on the U-Net network architecture, including an encoder and a decoder. The encoder adopts the deep residual network ResNet34. U-Net allows the network to capture contextual information in the contraction path and recover image details in the expansion path, which can accurately detect dynamic objects. ResNet34 can improve the depth of the network by introducing residual blocks and cross-layer connections, thereby enhancing the network's representation and feature extraction capabilities. At the same time, ResNet34 can effectively alleviate the gradient vanishing and gradient exploding problems and has high computational efficiency. Combining the precise localization capability of U-Net with the deep feature extraction capability of ResNet-34 for dynamic object detection can improve the performance of dynamic object detection, thereby further optimizing the forward optical flow and improving the accuracy and reliability of visual odometry. Attached Figure Description

[0029] Figure 1 A schematic diagram of the visual odometry measurement model for suppressing dynamic objects;

[0030] Figure 2 This is a schematic diagram of optical flow consistency detection;

[0031] Figure 3 This is a schematic diagram of the outputs of each network in the visual odometry model for dynamic object suppression.

[0032] Wherein, (3a) is the input image at time t, (3b) is the input image at time t+1, (3c) is the forward optical flow estimation of the matching network, (3d) is the backward optical flow estimation of the matching network, (3e) is the forward optical flow consistency detection, (3f) is the backward optical flow consistency detection, (3g) is the motion mask detected by the dynamic detection network, and (3h) is the visualization of motion segmentation on the image;

[0033] Figure 4 This diagram illustrates the qualitative results of different models on challenging sequences.

[0034] Among them, (4a), (4b), (4c) and (4d) are the segmentation results of dynamic target detection by the method of the present invention, and (4e), (4f), (4g), (4h), (4i), (4j), (4k) and (41) are the VO test result trajectories of KITTI odometer sequences 00, 01, 02, 03, 05, 07, 09 and 10, respectively. Detailed Implementation

[0035] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0036] Example 1

[0037] The paper "TartanVO: A Generalizable Learning-based VO" (W. Wang, Y. Hu, and S. Scherer, CoRL, volume 155 of Proceedings of Machine Learning Research, pages 1761-1772, PMLR, 2020) designed a two-stage TartanVO network and proposed a method to achieve cross-camera / dataset generalization by directly incorporating camera intrinsic functions into the model. However, in dynamic scenes, this method is easily affected by dynamic objects. This embodiment designs a three-stage network architecture based on the two-stage TartanVO network, providing a visual odometry method for dynamic object suppression. It captures occlusion information caused by the movement of dynamic objects in the scene through optical flow consistency detection and directly removes the dynamic optical flow caused by moving objects from the optical flow image. The process is as follows:

[0038] Two consecutive input image frames are acquired, and a dynamic object suppression visual odometry model (MoMaskVO) is input to obtain camera pose changes, including camera rotation R and camera displacement T.

[0039] like Figure 1 As shown, the visual odometry model with dynamic object suppression includes a matching network, a dynamic detection network, and a pose estimation network connected in sequence. The specific process of the visual odometry model with dynamic object suppression processing the input data is as follows:

[0040] S1, combine two consecutive input image frames I t and I t+1 The input is a matching network, which estimates the forward optical flow using GMFlow. and backward optical flow

[0041] S2, Forward optical flow and backward optical flow Perform optical flow consistency detection to obtain forward optical flow consistency. Backward consistency with optical flow

[0042] Optical flow consistency detection identifies dynamic objects by detecting background changes caused by motion, including forward optical flow consistency detection and backward optical flow consistency detection. For example... Figure 2 As shown, assume P t For image frame I t The pixels on, P t+1 For image frame I t+1 For pixels on the [aspect], if they satisfy the following formula, then the forward optical flow consistency detection result is inconsistent:

[0043]

[0044] in, This represents the optical flow from time t to time t+1. The forward optical flow is added to the forward optical flow of the pixel at time t to find the backward optical flow of the corresponding pixel in the next frame.

[0045] If the following formula is satisfied, then the backward optical flow consistency detection result is inconsistent:

[0046]

[0047] The above formula represents the consistency detection under theoretical conditions. In practice, optical flow always has errors. Therefore, the specific process of forward optical flow consistency detection is as follows: using forward optical flow... Backward optical flow The optical flow field is deformed to generate a new optical flow field. Then the forward optical flow is... The forward optical flow is added to the deformed backward optical flow. Theoretically, the forward optical flow and the deformed backward optical flow should be opposite in direction and equal in magnitude, and their sum should be close to zero. However, due to certain errors in practice, if the sum exceeds a set threshold, the area is considered to have occlusion or the forward optical flow consistency detection result is inconsistent, and the corresponding pixel is represented by gray.

[0048] The specific formula for calculating the threshold is as follows:

[0049] threshold = α * flow_mag + β

[0050] Where threshold is the set threshold, α and β are threshold influence parameters, which can be 0.01 and 0.5 respectively, and flow_mag is the optical flow magnitude.

[0051] Using forward optical flow Backward optical flow The specific process of deformation is as follows: A two-dimensional grid coordinate system is established in the space of the target image, representing the position of each pixel in the image. This is achieved by adjusting the forward optical flow... By adding these coordinates to the original grid, the position of each pixel in image A in image B can be calculated. By sampling the backward optical flow at these new coordinate points using bilinear interpolation, the deformed backward optical flow can be obtained.

[0052] Similarly, the process of backward optical flow consistency detection is as follows: using backward optical flow... Forward optical flow Perform deformation, then backflow optical flow The forward optical flow is added to the deformed forward optical flow. If the sum exceeds the set threshold, the region is occluded or the backward optical flow consistency detection result is inconsistent, and the corresponding pixel is indicated in red.

[0053] Forward optical flow consistency obtained through optical flow consistency detection It can determine which pixels will disappear in the next frame, ensuring backward consistency of optical flow. It's possible to determine which pixels will appear in the next frame, and based on this, the optical flow map can be segmented into dynamic and static regions. Some existing methods use optical flow to detect dynamic objects, but at corners, scene changes are more pronounced at faster speeds, and the changes in optical flow are significant across all pixels, causing the vast majority of pixels to be labeled as dynamic, leading to substantial errors in self-motion estimation at corners. If semantic segmentation is used, it is not only slow, but it can only segment a limited number of categories and cannot generalize to unrecognized categories.

[0054] S3, Ensure forward optical flow consistency Backward consistency with optical flow Input the dynamic detection network to obtain the probability graph. This probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Using binary segmentation mask Optimize forward optical flow To remove forward optical flow The dynamic region within.

[0055] If there are moving objects in the image, inconsistencies will arise in both forward and backward optical flow consistency. Camera motion also causes inconsistencies in forward and backward optical flow consistency detection, but these are less pronounced than with moving objects. Therefore, forward and backward optical flow consistency can be input into a motion detection network to detect moving objects.

[0056] The dynamic detection network is built on the U-Net network architecture, including an encoder and a decoder, with skip connections between them. In a preferred embodiment, the encoder uses a deep residual network, ResNet34. U-Net allows the network to capture contextual information in the contraction path and recover image details in the expansion path, enabling accurate detection of dynamic objects. ResNet34, by introducing residual blocks and cross-layer connections, increases the network's depth, thereby enhancing its representation and feature extraction capabilities. Simultaneously, ResNet34 effectively mitigates the vanishing and exploding gradient problems and has high computational efficiency. Combining the precise localization capabilities of U-Net with the deep feature extraction capabilities of ResNet-34 for dynamic object detection improves dynamic object detection performance, further optimizing forward optical flow and enhancing the accuracy and reliability of visual odometry.

[0057] S4. Divide the binary segmentation mask Optimized forward optical flow Input the intrinsic parameter layer into the pose estimation network to obtain the camera pose change.

[0058] The intrinsic parameter layer is constructed based on the camera intrinsic parameters, which include focal length and principal point position. For details, please refer to tartanvo.

[0059] The outputs of each network in the visual odometry model for dynamic object suppression are as follows: Figure 3 As shown, where, Figure 3 (3a) is the input image at time t, (3b) is the input image at time t+1, (3c) is the forward optical flow estimation of the matching network, (3d) is the backward optical flow estimation of the matching network, (3e) is the forward optical flow consistency detection, (3f) is the backward optical flow consistency detection, (3g) is the motion mask detected by the dynamic detection network, and (3h) is the visualization of motion segmentation on the image.

[0060] The method described above detects background changes by adding optical flow consistency detection, capturing occlusion information caused by the movement of dynamic objects in the scene, and then segments and processes this information using a dynamic detection network. This allows for the direct removal of dynamic optical flow caused by moving objects from the optical flow image, thereby reducing the negative impact of dynamic objects on visual odometry. This improvement enables the visual odometry system to estimate camera pose more accurately and maintain high performance even in scenes with dynamic objects.

[0061] To verify the effectiveness of the above method, this embodiment first compared the inference time of TartanVO, DytanVO which performs 3 iterations, and MoMaskVO (Ours) proposed in this invention on the KITTI odometry dataset. The experimental results are shown in Table 1.

[0062] Table 1. Comparison of inference time for different methods tested on the KITTI odometer dataset.

[0063]

[0064] In Table 1, Flow time represents the inference time for optical flow estimation; Mask time represents the inference time for the dynamic detection network; Pose time represents the inference time for the pose estimation network; and Total represents the total time. TartanVO and DytanVO both use PWC-Net, but due to the need to estimate forward and backward optical flow, this invention uses GMFlow, resulting in a longer Flow time than TartanVO and DytanVO. However, MoMaskVO's dynamic detection network only requires 7.5 milliseconds per computation, while DytanVO's requires 37 milliseconds per computation and two computations. All three models take 5.2 milliseconds to run the pose estimation network once, but DytanVO requires three computations. As shown in Table 1, the total time of this invention is significantly less than DytanVO, only 42.73% of DytanVO's.

[0065] Secondly, this embodiment compares the accuracy of the above three methods on the KITTI odometer dataset, and the comparison results are shown in Table 2.

[0066] Table 2 Comparison of accuracy of different methods tested on the KITTI odometer dataset

[0067]

[0068] In Table 2, 00-10 represent eleven sequences from the KITTI odometry dataset. The KITTI odometry benchmark measures the performance of visual odometry methods by evaluating trajectory subsequences ranging in length from 100 meters to 800 meters in 100-meter increments. For each trajectory subsequence, the standard reports two evaluation criteria: mean translational error and rotational error. Mean translational error (t...) err Rotational Error (r): Expressed as a percentage, this measures the difference between the algorithm's estimated camera position and the actual position. err): Expressed in degrees per 100 meters, this measures the difference between the algorithm's estimated camera rotation and the actual rotation. These error metrics provide quantitative data for comparing the accuracy of different methods in estimating camera position and rotation. In Table 2, the best results are bolded, and the second best results are underlined. The data in Table 2 show that, tested on the KITTI odometry dataset, the dynamic object suppression visual odometry method provided in this invention significantly outperforms DytanVO in accuracy.

[0069] Figure 4 Qualitative results for the three models are shown in challenging sequences. Figure 4 Images (4a), (4b), (4c), and (4d) show the segmentation results of dynamic target detection using the method of the present invention. It can be seen that even when there are multiple dynamic targets, high-speed vehicles, dynamic targets occupying a large area of ​​the image, and low-speed cyclists in the image, the method of the present invention can still effectively identify and segment these dynamic targets. This demonstrates the robustness of the method of the present invention in handling highly dynamic and complex scenes. Figure 4 (4e), (4f), (4g), (4h), (4i), (4j), (4k), and (4l) are the VO test result trajectories of KITTI odometer sequences 00, 01, 02, 03, 05, 07, 09, and 10, respectively, and GT is the standard trajectory. It can be seen that the method of the present invention (MoMaskVO) has good accuracy.

[0070] This embodiment also conducted experiments on the AirDOS-Shibuya dataset. AirDOS-Shibuya is a highly dynamic scene captured using a handheld camera, exhibiting more complex motion patterns. This embodiment compares the method of this invention with state-of-the-art models on this dataset, and the results are shown in Table 3.

[0071] Table 3 Comparison of accuracy of different methods on the AirDOS-Shibuya dataset

[0072]

[0073] Table 3 shows that DROID-SLAM significantly outperforms existing traditional SLAM systems on many challenging benchmark datasets; AirDOS, VDO-SLAM, and DynaSLAM are three SLAM methods for handling dynamic environmental disturbances. Among visual odometry (VO) methods, DeepVO, Trainflow, and CC perform poorly because these models are only trained on the KITTI dataset, and complex motion patterns make them difficult to infer correctly. The model proposed in this invention significantly improves upon TartanVO and achieves comparable results to DytanVO, particularly achieving the best performance on sequences II, V, and VII. This result demonstrates the effectiveness and robustness of the proposed method in handling highly dynamic scenes.

[0074] If the above methods are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0075] Example 2

[0076] This embodiment provides a visual odometry system for dynamic object suppression, including an optical flow estimation module, a dynamic detection module, and a pose estimation module connected in sequence.

[0077] The optical flow estimation module is used to estimate the optical flow based on two consecutive input image frames I. t and I t+1 Estimate forward optical flow and backward optical flow and forward optical flow and backward optical flow Optical flow consistency testing is performed, including forward optical flow consistency testing and backward optical flow consistency testing. The forward optical flow consistency is obtained based on the results of the forward and backward optical flow consistency testing. Backward consistency with optical flow These are used to represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively.

[0078] The dynamic detection module is used to determine the forward consistency of optical flow. Backward consistency with optical flow Generate probability map This probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Using binary segmentation mask Optimize forward optical flow To remove forward optical flow The dynamic region within.

[0079] The pose estimation module is used to estimate the pose based on the binary segmentation mask. Optimized forward optical flow And intrinsic parameter layers, to obtain camera pose changes This includes camera rotation R and camera displacement T.

[0080] The above description of the embodiments is provided to enable those skilled in the art to understand and use the invention. It will be apparent to those skilled in the art that various modifications can be made to these embodiments, and the general principles described herein can be applied to other embodiments without inventive effort. Therefore, the present invention is not limited to the above embodiments, and any improvements and modifications made by those skilled in the art based on the disclosure of the present invention without departing from the scope of the invention should be within the protection scope of the present invention.

Claims

1. A method for visual odometry measurement with dynamic object suppression, characterized in that, Includes the following steps: Two consecutive input image frames are acquired, and a visual odometry model for dynamic object suppression is input to obtain camera pose changes, including camera rotation R and camera displacement T. The visual odometry model for dynamic object suppression includes a matching network, a dynamic detection network, and a pose estimation network connected in sequence. The specific process of the visual odometry model for dynamic object suppression processing input data is as follows: Two consecutive input image frames I t and I t+1 Input the matching network to estimate the forward optical flow. and backward optical flow According to the forward optical flow and the backward optical flow Optical flow consistency detection is performed, including forward optical flow consistency detection and backward optical flow consistency detection. Forward optical flow consistency is obtained based on the results of the forward and backward optical flow consistency detections. Backward consistency with optical flow These are used to represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively. The forward consistency of the optical flow and the backward consistency of the optical flow Input the dynamic detection network to obtain the probability graph. The probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Through the binary segmentation mask Optimize the forward optical flow To remove the forward optical flow Dynamic regions within; The binary segmentation mask Optimized forward optical flow The pose estimation network is input along with the intrinsic parameter layer to obtain the camera pose change.

2. The visual odometry measurement method for dynamic object suppression according to claim 1, characterized in that, The matching network obtains the forward optical flow through GMFlow estimation. and the backward optical flow 3. The visual odometry measurement method for dynamic object suppression according to claim 1, characterized in that, The forward optical flow consistency detection process is as follows: using the forward optical flow... For the backward optical flow Perform deformation, and then the forward optical flow If the summation exceeds a set threshold, the forward optical flow consistency detection result is considered inconsistent.

4. The visual odometry measurement method for dynamic object suppression according to claim 3, characterized in that, The specific formula for calculating the set threshold is as follows: theshold = α * flow_mag + β Where theshold is the set threshold, α and β are threshold influence parameters, and flow_mag is the optical flow magnitude.

5. The visual odometry method for dynamic object suppression according to claim 1, characterized in that, The process of backward optical flow consistency detection is as follows: using the backward optical flow For the forward optical flow Perform deformation, and then the backward optical flow If the summation exceeds a set threshold, the backward optical flow consistency detection result is considered inconsistent.

6. The visual odometry method for dynamic object suppression according to claim 1, characterized in that, The dynamic detection network is built on the U-Net network architecture and includes an encoder and a decoder, with skip connections between the encoder and the decoder.

7. The visual odometry method for dynamic object suppression according to claim 6, characterized in that, The encoder uses a deep residual network, ResNet34.

8. The visual odometry method for dynamic object suppression according to claim 1, characterized in that, The intrinsic parameter layer is constructed based on camera intrinsic parameters, which include focal length and principal point position.

9. A visual odometry system for dynamic object suppression, characterized in that, It includes an optical flow estimation module, a dynamic detection module, and a pose estimation module connected in sequence; The optical flow estimation module is used to estimate the optical flow based on two consecutive input image frames I. t and I t+1 Estimate the forward optical flow and backward optical flow and the forward optical flow and the backward optical flow Optical flow consistency detection is performed, including forward optical flow consistency detection and backward optical flow consistency detection. Forward optical flow consistency is obtained based on the results of the forward and backward optical flow consistency detections. Backward consistency with optical flow These are used to represent pixels that disappear in the next frame and pixels that appear in the next frame, respectively. The dynamic detection module is used to determine the forward consistency of the optical flow. and the backward consistency of the optical flow Generate probability map The probability map is used to reflect the probability that each pixel in the input image frame is a dynamic pixel. Convert to binary segmentation mask Through the binary segmentation mask Optimize the forward optical flow To remove the forward optical flow Dynamic regions within; The pose estimation module is used to estimate the position based on the binary segmentation mask. Optimized forward optical flow And intrinsic parameter layers, to obtain camera pose changes This includes camera rotation R and camera displacement T.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-8.

Images