Absolute depth determination method, device, medium and product based on monocular camera

Abstract

The present invention discloses a method, device, medium, and product for determining absolute depth based on a monocular camera, relating to the field of depth estimation technology. The method comprises: obtaining a visible light image and laser radar data of a target area; obtaining a first extrinsic parameter matrix; the first extrinsic parameter matrix being the extrinsic parameter matrix between the monocular camera and the laser radar; inputting the visible light image of the target area into a relative depth map generation model to obtain a relative depth map of the target area; the relative depth map generation model is obtained by training a relative depth estimation network using a time series of visible light images, the relative depth estimation network comprising a depth estimation network and a pose estimation network; determining a scale factor based on the first extrinsic parameter matrix, the laser radar data of the target area, and the relative depth map; and determining an absolute depth map of the target area based on the relative depth map and the scale factor. The present invention solves the problem of being unable to obtain an absolute depth map using a monocular view.

Description

Technical Field

[0001] The present invention relates to the technical field of depth estimation, and in particular to a method, device, medium and product for determining absolute depth based on a monocular camera. Background Art

[0002] Depth estimation plays a crucial role in intelligent transportation, a prerequisite for accurately understanding and adapting to the surrounding environment. Traditional methods typically use binocular cameras for image feature extraction and matching, calculating depth information based on the geometric positional relationship between the images. However, these methods suffer from limitations such as high computational complexity and conflicts between the installation size and the device being used. With the development of deep learning algorithms, single-view depth estimation has become possible using a monocular camera alone, leveraging the powerful fitting capabilities of convolutional networks. These methods offer two advantages over traditional approaches: first, they eliminate the need for complex matching algorithms and directly generate a pixel-level depth map that matches the original input image size; second, their installation size is significantly reduced, making them suitable for use on small smart devices. However, there are still drawbacks. When using a self-supervised framework, the supervisory information is reconstructed from the front and back views. However, the lack of scale information in the camera coordinate system in the pose estimation network results in scale ambiguity in the predicted depth map, resulting in only a relative depth map. Therefore, a method for determining absolute depth estimation based on a monocular camera and a self-supervised architecture is urgently needed. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, device, medium and product for determining absolute depth based on a monocular camera, which solves the problem that an absolute depth map cannot be obtained through a monocular view.

[0004] To achieve the above object, the present invention provides the following solutions:

[0005] A method for determining absolute depth based on a monocular camera, comprising:

[0006] Obtaining visible light images and lidar data of the target area; the visible light image is an RGB image acquired by a monocular camera, and the lidar data is data acquired by a lidar;

[0007] Obtain a first extrinsic parameter matrix; the first extrinsic parameter matrix is ​​an extrinsic parameter matrix between the monocular camera and the laser radar; the extrinsic parameter matrix includes: a transfer matrix and a translation matrix;

[0008] Inputting the visible light image of the target area into a relative depth map generation model to obtain a relative depth map of the target area; the relative depth map generation model is obtained by training a relative depth estimation network using a time series of visible light images, the relative depth estimation network comprising: a depth estimation network and a pose estimation network, both of which are encoder-decoder network architectures;

[0009] Determining a scale factor based on the first extrinsic parameter matrix and the lidar data and relative depth map of the target area;

[0010] An absolute depth map of the target area is determined based on the relative depth map of the target area and the scale factor.

[0011] Optionally, the process of determining the first extrinsic parameter matrix includes:

[0012] Acquire visible light images and lidar data for the joint calibration area;

[0013] Determining multiple position points in the joint calibration area as feature points;

[0014] The first extrinsic parameter matrix is ​​determined based on the pixel coordinates and radar coordinates of each of the feature points; the pixel coordinates are coordinates in the camera coordinate system of the camera, and the radar coordinates are coordinates in the lidar coordinate system of the lidar.

[0015] Optionally, the process of determining the relative depth map generation model includes:

[0016] Acquire the visible light image time series sequence; the visible light image time series sequence includes: visible light images of the training area acquired at multiple consecutive sampling moments;

[0017] Constructing the relative depth estimation network based on the depth estimation network and the pose estimation network;

[0018] Initializing the relative depth estimation network;

[0019] The relative depth estimation network is iteratively trained multiple times using each visible light image in the visible light image time series sequence to obtain the relative depth map generation model.

[0020] Optionally, performing multiple iterative training on the relative depth estimation network using each visible light image in the visible light image time series sequence to obtain the relative depth map generation model includes:

[0021] Determine any sampling moment among all sampling moments corresponding to the visible light image time series as the current moment;

[0022] Determine the visible light image at the current moment as the first visible light image, and determine the visible light image at the next moment as the second visible light image;

[0023] determining the first visible light image and the second visible light image as an image group;

[0024] The relative depth estimation network is iteratively trained multiple times using each of the image groups to obtain the relative depth map generation model; wherein the training process under any current training times includes:

[0025] Determine any image group as the current image group;

[0026] Inputting the current image group into the pose estimation network at the current training number to obtain a second extrinsic parameter matrix of the current image group at the current training number; the second extrinsic parameter matrix is ​​an extrinsic parameter matrix between the first visible light image and the second visible light image;

[0027] Inputting the second visible light image in the current image group into the depth estimation network at the current training times to obtain a relative depth map of the current image group at the current training times;

[0028] Determining a reconstructed image of the current image group at the current training number based on the second extrinsic parameter matrix at the current training number and the relative depth map at the current training number;

[0029] Calculating a loss function value at the current training times based on the reconstructed image and the first visible light image of each image group at the current training times;

[0030] Determine whether the training stop condition is met; the training stop condition is that the loss function value under the current training number is less than a preset threshold or the preset training number is reached;

[0031] If yes, the depth estimation network under the current training times is determined as the relative depth map generation model;

[0032] If not, the back propagation method is used to update the parameters of the pose estimation network under the current training times and the parameters of the depth estimation network under the current training times, and return to "inputting the current image group into the pose estimation network under the current training times to obtain the second external parameter matrix of the current image group under the current training times" until the training stop condition is met.

[0033] Optionally, determining a scale factor based on the first extrinsic parameter matrix and the lidar data and relative depth map of the target area includes:

[0034] Reprojecting the lidar data of the target area onto a relative depth map of the target area according to the first extrinsic parameter matrix;

[0035] Determining a set pixel point in the laser radar data of the target area as a first scale factor calculation point;

[0036] Determining, according to the first extrinsic parameter matrix, a pixel point in the relative depth map of the target area that matches the first scale factor calculation point as a second scale factor calculation point;

[0037] The scale factor is calculated based on the depth value corresponding to the first scale factor calculation point and the relative depth value corresponding to the second scale factor calculation point.

[0038] Optionally, calculating the scale factor based on the depth value corresponding to the first scale factor calculation point and the relative depth value corresponding to the second scale factor calculation point includes:

[0039] The scale factor is determined as a ratio of a depth value corresponding to the first scale factor calculation point to a relative depth value corresponding to the second scale factor calculation point.

[0040] Optionally, determining the absolute depth map of the target area based on the relative depth map of the target area and the scale factor includes:

[0041] The ratio of the relative depth value at each pixel point in the relative depth map of the target area to the scale factor is determined as the absolute depth value at the corresponding pixel point, thereby obtaining the absolute depth map of the target area.

[0042] A computer device comprises: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the above-mentioned methods for determining absolute depth based on a monocular camera.

[0043] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements any of the above-mentioned methods for determining absolute depth based on a monocular camera.

[0044] A computer program product comprises a computer program, which, when executed by a processor, implements any of the above-mentioned methods for determining absolute depth based on a monocular camera.

[0045] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0046] The present invention discloses a method, device, medium and product for determining absolute depth based on a monocular camera. The method of determining the absolute depth map using a monocular view simplifies the previous binocular system, greatly reduces the installation size, and can meet the requirements of application on small smart devices. At the same time, a method is proposed to convert the relative depth map into an absolute depth map by calculating the scale factor through lidar data, which solves the current problem that the absolute depth map cannot be obtained through a monocular view, and improves the ability to detect obstacles using a monocular camera. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0048] Figure 1 A schematic flow chart of a method for determining absolute depth based on a monocular camera according to embodiment 1 of the present invention;

[0049] Figure 2 Schematic diagram of the relative depth of field estimation network training principle;

[0050] Figure 3 Schematic diagram of the depth estimation network structure;

[0051] Figure 4 Schematic diagram of the intermediate feature aggregation node structure. DETAILED DESCRIPTION

[0052] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0053] The purpose of the present invention is to provide a method, device, medium and product for determining absolute depth based on a monocular camera, aiming to improve the accuracy of absolute depth determination based on a monocular camera.

[0054] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] Example 1

[0056] like Figure 1As shown, the absolute depth determination method based on a monocular camera in this embodiment includes:

[0057] Step 1: Obtain visible light images and lidar data of the target area.

[0058] Among them, the visible light image is the RGB image captured by the monocular camera, and the lidar data is the data collected by the lidar.

[0059] Step 2: Get the first external parameter matrix.

[0060] Among them, the first extrinsic parameter matrix is ​​the extrinsic parameter matrix between the monocular camera and the laser radar; the extrinsic parameter matrix includes: a transfer matrix and a translation matrix.

[0061] As an optional implementation manner, the process of determining the first external parameter matrix includes:

[0062] Step 21: Obtain visible light images and lidar data for the joint calibration area.

[0063] Step 22: Determine multiple position points in the joint calibration area as feature points.

[0064] Step 23: Determine a first extrinsic parameter matrix based on the pixel coordinates and radar coordinates of each feature point; the pixel coordinates are coordinates in the camera coordinate system of the camera, and the radar coordinates are coordinates in the lidar coordinate system of the lidar.

[0065] Specifically, determining the first external parameter matrix is ​​to realize the joint calibration of the monocular camera and the lidar. The joint calibration formula is:

[0066]

[0067] Among them, Z c is the Z-axis coordinate of the feature point in the laser radar coordinate system; (u, v) is the two-dimensional coordinate of the feature point in the camera coordinate system; K is the intrinsic parameter matrix of the monocular camera; R is the transfer matrix; X c is the X-axis coordinate of the feature point in the laser radar coordinate system; Y c is the Y-axis coordinate of the feature point in the lidar coordinate system; T is the translation matrix.

[0068] Joint calibration mainly includes three steps: (1) data feature extraction: using breakpoint features (such as ground planes, building edges, etc.) as feature points; (2) feature matching: matching the extracted feature points with the corresponding pixel points in the visible light image and lidar data based on the manual feature method to establish a correspondence between them and obtain matching point pairs; (3) calibration parameter optimization: the extracted matching point pairs are calculated using the Levenberg–Marquardt (LM) algorithm to obtain R and T.

[0069] Step 3: Input the visible light image of the target area into the relative depth map generation model to obtain the relative depth map of the target area.

[0070] Among them, the relative depth map generation model is obtained by training the relative depth estimation network using the time series of visible light images. The relative depth estimation network includes: a depth estimation network and a pose estimation network. The architectures of the depth estimation network and the pose estimation network are both encoder-decoder network architectures.

[0071] As an optional implementation, Figure 2 As shown in FIG, the process of determining the relative depth map generation model includes:

[0072] Step 31: Acquire a time series of visible light images; the time series of visible light images includes: visible light images of the training area collected at multiple consecutive sampling moments.

[0073] Step 32: Construct a relative depth estimation network based on the depth estimation network and the pose estimation network.

[0074] Specifically, such as Figure 3 As shown in Figure 2, the depth estimation network adopts an encoder-decoder architecture consisting of three parts:

[0075] (1) Encoder: Using the pre-trained ResNet-18 convolutional neural network as the backbone network can effectively avoid the gradient vanishing and gradient exploding problems during training. The encoder gradually reduces the spatial size of the input image through continuous downsampling and convolution operations to extract deep feature information of different dimensions. While reducing the size, the number of feature channels is increased to ensure that the network can more easily extract abstract deep feature information. The specific parameters of the ResNet-18 convolutional neural network are shown in Table 1.

[0076] Table 1 ResNet18 model parameters and structure

[0077]

[0078] Among them, Conv1 represents the first convolutional layer; Conv2 represents the second convolutional layer; Conv3 represents the third convolutional layer; Conv4 represents the fourth convolutional layer; Conv5 represents the fifth convolutional layer;

[0079] The input of the network is set to an image with a resolution of 640×480. The encoder extracts feature maps of five different dimensions. The lower the dimension, the lower the feature space resolution, and the better it can represent the deep semantic features.

[0080] (2) Intermediate feature aggregation nodes: Add different numbers of intermediate feature aggregation nodes in different dimensions. Figure 4The intermediate aggregation node shown is used to integrate feature information at different levels. Feature information at the same level includes selective connections and direct connections, which retains shallow information to the greatest extent and helps to achieve more refined texture segmentation. At the same time, it also accepts the feature information of the next level obtained by upsampling, which can pay more attention to deep semantic information, help better positioning and thus achieve multi-scale feature fusion.

[0081] (3) Decoder: After receiving feature information from the encoder and intermediate feature aggregation nodes, the decoder gradually restores the image size to 1 / 8, 1 / 4, 1 / 2, and 1 of the original size by upsampling the underlying feature information. At the same time, it gradually restores the number of channels from 256 to 128, 64, 32, and 16 through deconvolution. Finally, the fused four-layer feature information is restored to a single-channel depth map, ensuring that the dimensions and size coordinates are completely consistent with the input. In short, the main function of the decoder is to convert the information extracted by the encoder into a pixel-level depth map to provide an estimate of the scene depth.

[0082] The pose estimation network is used to predict the pose transformation between visible light images at different sampling times to construct supervision information. Similar to the encoder and decoder structure of the depth estimation network, it also consists of two parts: an encoder and a decoder, except that the output becomes the predicted pose transformation relationship between the front and back views. Similarly, the encoder is responsible for extracting semantic features from the two visible light images, while the decoder is responsible for feature point matching and mapping these features into a pose transformation matrix. In this part, the decoder first uses a 1×1 convolutional layer for channel dimensionality reduction, followed by two 3×3 convolutional layers to further extract high-level semantic features related to the pose transformation. Finally, a 1×1 convolutional layer converts the features into a 1×1×6 six-degree-of-freedom pose transformation information.

[0083] Step 33: Initialize the relative depth estimation network.

[0084] Step 34: Perform multiple iterative training on the relative depth estimation network using each visible light image in the visible light image time series sequence to obtain a relative depth map generation model.

[0085] As an optional implementation, step 34 includes:

[0086] Step 341: Determine any sampling moment among all sampling moments corresponding to the visible light image time series as the current moment.

[0087] Step 342: Determine the visible light image at the current moment as the first visible light image, and determine the visible light image at the next moment as the second visible light image.

[0088] Step 343: Determine the first visible light image and the second visible light image as an image group.

[0089] Step 344: Perform multiple iterations of training on the relative depth estimation network using each image group to obtain a relative depth map generation model. The training process for any current number of training times includes:

[0090] Any image group is determined as the current image group.

[0091] The current image group is input into the pose estimation network under the current training number to obtain a second extrinsic parameter matrix of the current image group under the current training number; the second extrinsic parameter matrix is ​​an extrinsic parameter matrix between the first visible light image and the second visible light image.

[0092] The second visible light image in the current image group is input into the depth estimation network under the current training number to obtain a relative depth map of the current image group under the current training number.

[0093] Based on the second extrinsic parameter matrix at the current training number and the relative depth map at the current training number, a reconstructed image of the current image group at the current training number is determined.

[0094] Based on the reconstructed image and the first visible light image of each image group at the current training number, the loss function value at the current training number is calculated.

[0095] Determine whether the training stop condition is met; the training stop condition is that the loss function value under the current training number is less than the preset threshold or the preset training number is reached.

[0096] If so, the depth estimation network under the current training times is determined as the relative depth map generation model.

[0097] If not, the backpropagation method is used to update the parameters of the pose estimation network under the current training times and the parameters of the depth estimation network under the current training times, and return to "input the current image group into the pose estimation network under the current training times to obtain the second external parameter matrix of the current image group under the current training times" until the training stop condition is met.

[0098] Step 4: Determine a scale factor based on the first extrinsic parameter matrix and the lidar data and relative depth map of the target area.

[0099] As an optional implementation, step 4 includes:

[0100] Step 41: Reproject the lidar data of the target area onto a relative depth map of the target area according to the first extrinsic parameter matrix.

[0101] Step 42: Determine a set pixel point in the laser radar data of the target area as a first scale factor calculation point.

[0102] Step 43: Determine, according to the first extrinsic parameter matrix, a pixel point in the relative depth map of the target area that matches the first scale factor calculation point as a second scale factor calculation point.

[0103] Step 44: Calculate a scale factor based on the depth value corresponding to the first scale factor calculation point and the relative depth value corresponding to the second scale factor calculation point.

[0104] As an optional implementation, step 44 includes:

[0105] The ratio of the depth value corresponding to the first scale factor calculation point to the relative depth value corresponding to the second scale factor calculation point is determined as the scale factor.

[0106] Step 5: Based on the relative depth map of the target area and the scale factor, determine the absolute depth map of the target area.

[0107] As an optional implementation, step 5 includes:

[0108] The ratio of the relative depth value at each pixel point in the relative depth map of the target area to the scale factor is determined as the absolute depth value at the corresponding pixel point, thereby obtaining the absolute depth map of the target area.

[0109] Specifically, the calculation formula for the absolute depth value at any pixel is:

[0110]

[0111] Among them, D abs is the absolute depth value; D rel is the relative depth value; is the scale factor.

[0112] Example 2

[0113] A computer device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the absolute depth determination method based on a monocular camera in embodiment 1.

[0114] Example 3

[0115] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the absolute depth determination method based on a monocular camera in embodiment 1.

[0116] Example 4

[0117] A computer program product includes a computer program, which, when executed by a processor, implements the absolute depth determination method based on a monocular camera in embodiment 1.

[0118] Example 5

[0119] A computer device, which may be a database. The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected via a system bus, and the communication interface is connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store pending transactions. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the absolute depth determination method based on a monocular camera in Example 1 is implemented.

[0120] It should be noted that the object information (including but not limited to object device information, object personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the object or fully authorized by all parties, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of relevant countries and regions.

[0121] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided by the present invention can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided by the present invention may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided by the present invention may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0122] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0123] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for determining absolute depth based on a monocular camera, characterized in that: The method comprises: Obtaining visible light images and lidar data of the target area; the visible light image is an RGB image acquired by a monocular camera, and the lidar data is data acquired by a lidar; Obtain a first extrinsic parameter matrix; the first extrinsic parameter matrix is ​​an extrinsic parameter matrix between the monocular camera and the laser radar; the extrinsic parameter matrix includes: a transfer matrix and a translation matrix; Inputting the visible light image of the target area into a relative depth map generation model to obtain a relative depth map of the target area; the relative depth map generation model is obtained by training a relative depth estimation network using a time series of visible light images, the relative depth estimation network comprising: a depth estimation network and a pose estimation network, both of which are encoder-decoder network architectures; Determining a scale factor based on the first extrinsic parameter matrix and the lidar data and relative depth map of the target area; An absolute depth map of the target area is determined based on the relative depth map of the target area and the scale factor.

2. The method for determining absolute depth based on a monocular camera according to claim 1, wherein: The process of determining the first external parameter matrix includes: Acquire visible light images and lidar data for the joint calibration area; Determining multiple position points in the joint calibration area as feature points; The first extrinsic parameter matrix is ​​determined based on the pixel coordinates and radar coordinates of each of the feature points; the pixel coordinates are coordinates in the camera coordinate system of the camera, and the radar coordinates are coordinates in the lidar coordinate system of the lidar.

3. The method for determining absolute depth based on a monocular camera according to claim 1, wherein: The process of determining the relative depth map generation model includes: Acquire the visible light image time series sequence; the visible light image time series sequence includes: visible light images of the training area acquired at multiple consecutive sampling moments; Constructing the relative depth estimation network based on the depth estimation network and the pose estimation network; Initializing the relative depth estimation network; The relative depth estimation network is iteratively trained multiple times using each visible light image in the visible light image time series sequence to obtain the relative depth map generation model.

4. The method for determining absolute depth based on a monocular camera according to claim 3, wherein: The relative depth estimation network is trained multiple times iteratively using each visible light image in the visible light image time series sequence to obtain the relative depth map generation model, including: Determine any sampling moment among all sampling moments corresponding to the visible light image time series as the current moment; Determine the visible light image at the current moment as the first visible light image, and determine the visible light image at the next moment as the second visible light image; determining the first visible light image and the second visible light image as an image group; The relative depth estimation network is iteratively trained multiple times using each of the image groups to obtain the relative depth map generation model; wherein the training process under any current training times includes: Determine any image group as the current image group; Inputting the current image group into the pose estimation network at the current training number to obtain a second extrinsic parameter matrix of the current image group at the current training number; the second extrinsic parameter matrix is ​​an extrinsic parameter matrix between the first visible light image and the second visible light image; Inputting the second visible light image in the current image group into the depth estimation network at the current training times to obtain a relative depth map of the current image group at the current training times; Determining a reconstructed image of the current image group at the current training number based on the second extrinsic parameter matrix at the current training number and the relative depth map at the current training number; Calculating a loss function value at the current training times based on the reconstructed image and the first visible light image of each image group at the current training times; Determine whether the training stop condition is met; the training stop condition is that the loss function value under the current training number is less than a preset threshold or the preset training number is reached; If yes, the depth estimation network under the current training times is determined as the relative depth map generation model; If not, the back propagation method is used to update the parameters of the pose estimation network under the current training times and the parameters of the depth estimation network under the current training times, and return to "input the current image group into the pose estimation network under the current training times to obtain the second external parameter matrix of the current image group under the current training times" until the training stop condition is met.

5. The method for determining absolute depth based on a monocular camera according to claim 1, wherein: Determining a scale factor based on the first extrinsic parameter matrix and the laser radar data and relative depth map of the target area includes: Reprojecting the lidar data of the target area onto a relative depth map of the target area according to the first extrinsic parameter matrix; Determining a set pixel point in the laser radar data of the target area as a first scale factor calculation point; Determining, according to the first extrinsic parameter matrix, a pixel point in the relative depth map of the target area that matches the first scale factor calculation point as a second scale factor calculation point; The scale factor is calculated based on the depth value corresponding to the first scale factor calculation point and the relative depth value corresponding to the second scale factor calculation point.

6. The method for determining absolute depth based on a monocular camera according to claim 5, wherein: Calculating the scale factor based on the depth value corresponding to the first scale factor calculation point and the relative depth value corresponding to the second scale factor calculation point includes: The scale factor is determined as a ratio of a depth value corresponding to the first scale factor calculation point to a relative depth value corresponding to the second scale factor calculation point.

7. The method for determining absolute depth based on a monocular camera according to claim 1, wherein: Determining an absolute depth map of the target area based on the relative depth map of the target area and the scale factor includes: The ratio of the relative depth value at each pixel point in the relative depth map of the target area to the scale factor is determined as the absolute depth value at the corresponding pixel point, thereby obtaining the absolute depth map of the target area.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the absolute depth determination method based on a monocular camera according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for determining absolute depth based on a monocular camera according to any one of claims 1 to 7 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the method for determining absolute depth based on a monocular camera according to any one of claims 1 to 7 is implemented.

Images