CN-115393828-B - Signpost identification method, device, computer equipment and storage medium

CN115393828BCN 115393828 BCN115393828 BCN 115393828BCN-115393828-B

Abstract

The invention provides a method, a device, computer equipment and a storage medium for identifying signboards, wherein the method comprises the steps of obtaining an image to be detected, inputting the image to be detected into a pre-trained target neural network, determining an initial feature map corresponding to the image to be detected, determining a first initial detection result, a second initial detection result and a depth detection result corresponding to the image to be detected based on the initial feature map, wherein the first initial detection result is used for representing whether each detection area of the image to be detected contains a signboard or not, the second initial detection result is used for representing the central position of the signboards in the image to be detected, determining the area to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result, adjusting the initial feature map based on target depth information corresponding to the area to be confirmed in the depth detection result, and determining the target detection result corresponding to the image to be detected based on the adjusted target feature map.

Inventors

LU XUPENG
GAO SANYUAN
ZOU YANG
YAO LEI
KANG FENGYI

Assignees

贵州宽凳智云科技有限公司

Dates

Publication Date: 20260505
Application Date: 20220922

Claims (10)

1. A method of identifying a sign, comprising: Acquiring an image to be detected, inputting the image to be detected into a pre-trained target neural network, and determining an initial feature map corresponding to the image to be detected; Determining a first initial detection result, a second initial detection result and a depth detection result corresponding to the image to be detected based on the initial feature map, wherein the first initial detection result is used for representing whether the central point of each type of signboard is contained in each detection area of the image to be detected, the second initial detection result is used for representing the central position of each type of signboard in the image to be detected, the depth detection result is used for representing depth information corresponding to each pixel point in the image to be detected, the second initial detection result comprises a first feature map and a second feature map, wherein the value of each first feature point of the first feature map is used for representing the probability that the central point of each type of signboard is contained in the corresponding detection area of the image to be detected, the value of each second feature point of the second feature map is used for representing the coordinate offset information of the central point of each type of signboard, and the size information of each type of signboard; Determining a region to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result; And adjusting the initial feature map based on the target depth information corresponding to the region to be confirmed in the depth detection result, and determining a target detection result corresponding to the image to be detected based on the adjusted target feature map.
2. The method of claim 1, wherein the determining the area to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result comprises: Determining a first detection area corresponding to the signpost in the image to be detected based on the first initial detection result, and determining a second detection area corresponding to the signpost in the image to be detected based on the second initial detection result; The area to be confirmed is determined based on the first detection area and the second detection area.
3. The method according to claim 1, wherein the adjusting the initial feature map based on the target depth information corresponding to the region to be confirmed in the depth detection result includes: Determining a first target feature point corresponding to the region to be confirmed in the initial feature map; and adjusting the value of the target channel of the first target feature point based on the target depth information.
4. The method of claim 1, wherein the second initial detection result is determined by a central point identification module of the target neural network; The determining the target detection result corresponding to the image to be detected based on the adjusted target feature map comprises the following steps: and inputting the target feature map to the center point identification module, and determining the target detection result.
5. The method according to claim 2, wherein determining a second detection area corresponding to the signpost in the image to be detected based on the second initial detection result includes: determining a second target feature point based on the value of each first feature point of the first feature map and a preset probability threshold; And determining the second detection area based on the value of a third target feature point corresponding to the second target feature point in the second feature map and the position coordinates of the second target feature point in the first feature map.
6. The method of claim 1, wherein the target neural network comprises a feature extraction module, a region identification module, a center point identification module, and a depth identification module, the feature extraction module configured to determine the initial feature map, the region identification module configured to determine the first initial detection result, the center point identification module configured to determine the second initial detection result, and the depth identification module configured to determine the depth detection result; The method further includes training the target neural network according to the steps of: Acquiring a sample image and a sample label corresponding to the sample image, wherein the sample label is used for representing the position and the category of a signpost in the sample image and sample depth information of the sample image; Based on the sample image and the sample label, training the feature extraction module, the region identification module, the center point identification module and the depth identification module simultaneously in sequence, and training the trained feature extraction module and another module to be trained after the feature extraction module and any one module are trained; And performing fine adjustment processing on the trained feature extraction module, the trained region identification module, the trained central point identification module and the trained depth identification module to obtain the target neural network.
7. The method of claim 6, wherein the training the feature extraction module with the region identification module, the center point identification module, and the depth identification module in sequence based on the sample image and the sample tag comprises: training the feature extraction module and the region identification module based on a detection frame in the sample tag for representing the position of the signpost in the sample image; Training the feature extraction module and the center point identification module based on the detection frame and the category of the signpost in the sample tag; And training the feature extraction module and the depth recognition module based on the sample depth information in the sample tag.
8. A sign recognition device, comprising: the acquisition module is used for acquiring an image to be detected, inputting the image to be detected into a pre-trained target neural network and determining an initial feature map corresponding to the image to be detected; The device comprises a first determining module, a second determining module and a first detecting module, wherein the first detecting module is used for determining a first initial detecting result, a second initial detecting result and a depth detecting result corresponding to the image to be detected based on the initial characteristic diagram, the first initial detecting result is used for representing whether a mark plate is contained in each detecting area of the image to be detected, the second initial detecting result is used for representing the central position of the mark plate in the image to be detected, the depth detecting result is used for representing depth information corresponding to each pixel point in the image to be detected, the second initial detecting result comprises a first characteristic diagram and a second characteristic diagram, the value of each first characteristic point of the first characteristic diagram is used for representing the probability that the corresponding detecting area in the image to be detected contains the central point of each type of mark plate, the value of each second characteristic point of the second characteristic diagram is used for representing, the target pixel point corresponding to each second characteristic point is used as coordinate offset information of the central point of each type of mark plate, and the size information of each type of mark plate; the second determining module is used for determining a region to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result; and the adjusting module is used for adjusting the initial feature map based on the target depth information corresponding to the region to be confirmed in the depth detection result, and determining a target detection result corresponding to the image to be detected based on the adjusted target feature map.
9. A computer device comprising a processor, a memory and a bus, the memory storing machine-readable instructions executable by the processor, the processor and the memory in communication via the bus when the computer device is in operation, the machine-readable instructions when executed by the processor performing the steps of the method of identifying signboards of any of claims 1 to 7.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when executed by a processor, performs the steps of the method for identifying a signboard of any one of claims 1 to 7.

Description

Signpost identification method, device, computer equipment and storage medium Technical Field The disclosure relates to the technical field of computers, and in particular relates to a method and a device for identifying a signpost, computer equipment and a storage medium. Background The traffic sign is used as an important road infrastructure, and the detection and identification of the traffic sign are important components in public digital assets of roads and also important components in high-precision technologies such as automatic driving technology, high-precision map making and the like. In the related art, when identifying traffic signboards, an image to be detected is often input into a convolutional neural network trained in advance, so that a signboard identification result output by the convolutional neural network is obtained, but when the situation that the signboards in the image to be detected are blocked exists, the situation that the traffic signboards are missed can exist by using the convolutional neural network, so that how to reduce the missed detection of the traffic signboards is a problem to be solved in the field. Disclosure of Invention The embodiment of the disclosure at least provides a method, a device, computer equipment and a storage medium for identifying signboards. In a first aspect, an embodiment of the present disclosure provides a method for identifying a signboard, including: Acquiring an image to be detected, inputting the image to be detected into a pre-trained target neural network, and determining an initial feature map corresponding to the image to be detected; Determining a first initial detection result, a second initial detection result and a depth detection result corresponding to the image to be detected based on the initial feature map, wherein the first initial detection result is used for representing whether a marker is contained in each detection area of the image to be detected, the second initial detection result is used for representing the central position of the marker in the image to be detected, and the depth detection result is used for representing depth information corresponding to each pixel point in the image to be detected; Determining a region to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result; And adjusting the initial feature map based on the target depth information corresponding to the region to be confirmed in the depth detection result, and determining a target detection result corresponding to the image to be detected based on the adjusted target feature map. In a possible implementation manner, the determining the area to be confirmed in the image to be detected based on the first initial detection result and the second initial detection result includes: Determining a first detection area corresponding to the signpost in the image to be detected based on the first initial detection result, and determining a second detection area corresponding to the signpost in the image to be detected based on the second initial detection result; The area to be confirmed is determined based on the first detection area and the second detection area. In a possible implementation manner, the adjusting the initial feature map based on the target depth information corresponding to the to-be-confirmed region in the depth detection result includes: Determining a first target feature point corresponding to the region to be confirmed in the initial feature map; and adjusting the value of the target channel of the first target feature point based on the target depth information. In a possible implementation manner, the second initial detection result is determined by a central point identification module of the target neural network; The determining the target detection result corresponding to the image to be detected based on the adjusted target feature map comprises the following steps: and inputting the target feature map to the center point identification module, and determining the target detection result. In a possible implementation manner, the second initial detection result includes a first feature map and a second feature map, wherein the value of each first feature point of the first feature map is used for representing, the corresponding detection area of the first feature point in the image to be detected includes the probability of the center point of each type of signboard, the value of each second feature point of the second feature map is used for representing, and the target pixel point corresponding to each second feature point is used as coordinate offset information of the center point of each type of signboard, and size information of each type of signboard. In a possible implementation manner, the determining, based on the second initial detection result, a second detection area corresponding to the signpost in the image to be detected includes: determining a