CN112668449A - Low-risk landform identification method for outdoor autonomous mobile robot - Google Patents
Low-risk landform identification method for outdoor autonomous mobile robot Download PDFInfo
- Publication number
- CN112668449A CN112668449A CN202011550817.8A CN202011550817A CN112668449A CN 112668449 A CN112668449 A CN 112668449A CN 202011550817 A CN202011550817 A CN 202011550817A CN 112668449 A CN112668449 A CN 112668449A
- Authority
- CN
- China
- Prior art keywords
- landform
- risk
- classification
- training
- classifier
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000012549 training Methods 0.000 claims abstract description 37
- 238000002360 preparation method Methods 0.000 claims abstract description 4
- 238000012795 verification Methods 0.000 claims description 19
- 238000012360 testing method Methods 0.000 claims description 13
- 238000011176 pooling Methods 0.000 claims description 8
- 230000007423 decrease Effects 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 5
- 238000010200 validation analysis Methods 0.000 claims description 5
- 230000002238 attenuated effect Effects 0.000 claims description 3
- 125000004122 cyclic group Chemical group 0.000 claims description 3
- 238000010008 shearing Methods 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 5
- 238000012216 screening Methods 0.000 abstract description 3
- 230000000007 visual effect Effects 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 244000025254 Cannabis sativa Species 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- 239000010426 asphalt Substances 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 239000004575 stone Substances 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a low-risk landform identification method of an outdoor autonomous mobile robot, which comprises the following steps: preparation before identification: specifying at least one risky landscape from a number of landscape types; collecting a landform image, and establishing a landform image data set; training and verifying at least two classification networks by using a geomorphic image data set to obtain at least one multi-classification classifier and at least one two-classification classifier; and (3) risk landform identification: and sequentially inputting the landform images to be identified into the classifier, outputting the landform images as the risk landform if the classifier identifies the risk landform, and otherwise, outputting the landform images as the low risk landform. The substantial effects of the invention include: through the mechanism of screening many times, can show the recognition rate that improves the risk landform to reduce the risk that outdoor robot brought because of the landform discernment mistake.
Description
Technical Field
The invention relates to the field of terrain identification, in particular to a low-risk terrain identification method for an outdoor autonomous mobile robot.
Background
An important application scenario for autonomous mobile robots is the outdoor environment. Compared to indoors, outdoor mobile robots face complex terrain-type challenges, such as soft sand, which can cause the robot to tip over, water-logging road surfaces which can cause the robot to short circuit, grass, which can increase the travel resistance of the robot. The current outdoor landform classification and identification methods comprise a contact vibration method, a vision method and the like. The contact vibration method analyzes the type of the landform by collecting signals when the robot is in contact with the landform, but such a method cannot judge the type of the landform in advance. Because the visual sensor is low in price, the acquired environmental information is rich, and the computing capability of a computer is remarkably improved in recent years, the landform identification method based on vision becomes a main research method in the field of landform identification.
The invention of the publication number CN110909637A provides an outdoor mobile robot terrain recognition method based on visual-touch fusion, which comprises the steps of firstly selecting terrain types according to different road surface materials; acquiring two modal data of touch and vision aiming at different terrains, and dividing a training set and a test set; constructing a cascade width learning network and training the network by using a touch training sample set and a visual training sample set, wherein in the training process, firstly, primary touch and visual characteristic extraction is carried out, then, fusion characteristic extraction of touch and visual is carried out, and then, a terrain recognition classification result is obtained through a broad inverse approximation of ridge regression and is used as the output of the cascade width learning network by using a fused touch characteristic matrix and a fused visual characteristic matrix through a width learning classification algorithm; and finally, inputting the test set into a trained cascade width learning network to obtain a classification result of terrain recognition.
However, in the prior art, the landform identification is a single identification result, so that the risk landform is difficult to accurately eliminate, and once the risk is omitted, serious risks are caused.
Disclosure of Invention
Aiming at the problems of low landform recognition rate and difficulty in eliminating risk landforms in the prior art, the invention provides a low-risk landform recognition method of an outdoor autonomous mobile robot.
The technical scheme of the invention is as follows.
A low-risk landform identification method for an outdoor autonomous mobile robot comprises the following steps: preparation before identification: specifying at least one risky landscape from a number of landscape types; collecting a landform image, and establishing a landform image data set; training and verifying at least two classification networks by using a geomorphic image data set to obtain at least one multi-classification classifier and at least one two-classification classifier; and (3) risk landform identification: and sequentially inputting the landform images to be identified into the classifier, outputting the landform images as the risk landform if the classifier identifies the risk landform, and otherwise, outputting the landform images as the low risk landform.
The method uses at least two trained classifiers to identify the landform image, and only when all the classifiers are identified as the low-risk landform, the final recognition is carried out, so that the possibility of missing judgment of the risk landform is reduced to the greatest extent, and the safety factor is obviously improved.
Preferably, the process of creating a geomorphic image dataset comprises the data enhancement steps of: performing data expansion on the acquired landform image through the cross combination of cyclic shearing, mirror symmetry and gray level adjustment to obtain a data set after data enhancement; and the data set is divided into a training set, a validation set, and a test set. After the operation, the original image data can be expanded to a plurality of times to meet the requirements of training, verification and testing.
Preferably, the training and verification process of the classification network includes: during training, inputting n images from a training set for training each time, and inputting a verification set to test the network precision after finishing network training for m times; when there is v1If the loss value of the secondary verification set does not decrease or the decrease amplitude is smaller than the threshold value, the learning rate is attenuated to half of the original learning rate until v is provided2The secondary network stops training after the loss values on the validation set no longer drop, where v2 > v 1. In order to enable the network parameters after training to be close to the optimal solution, a method of dynamically attenuating the learning rate is adopted, and the effect is obvious.
Preferably, the process of risk landform identification includes: inputting the acquired landform image into a multi-classification classifier to obtain an identification result, directly outputting the risk landform as a result if the landform of the identification result is the risk landform, and inputting the landform image into a two-classification classifier if the landform of the identification result does not belong to the risk landform; and judging whether the landform image is a risk landform or not by the two-classification classifier, if so, outputting the risk landform image, and if not, outputting the identification result obtained by the multi-classification classifier.
Preferably, the classification network is AlexNet, and the feature extraction layers between the classification networks adopt different structures, wherein the classification network corresponding to the multi-classification classifier includes two convolution layers and two maximum pooling layers, and the classification network corresponding to the two-classification classifier includes three convolution layers and three maximum pooling layers. In order to enable each network to have a higher identification rate, the network can be designed according to the current excellent CNN network structure, and meanwhile, the convolution layer and the pooling layer of the two networks have different structures, so that different image characteristics can be extracted, and further double verification is realized.
Preferably, the ratio of the number of images in the training set, the verification set and the test set is 6: 2: 2.
the substantial effects of the invention include: based on the CNN algorithm, firstly, the landform types are subjected to risk grade division, then an identification method with a 'double verification' thought is provided, the 'potential risk landforms' are identified for many times, and the landforms are identified as 'risk landforms' only by the CNN algorithm once. Through the multiple screening mechanism, the risk landform recognition rate can be obviously improved, and therefore the risk of the outdoor robot caused by the error landform recognition is reduced.
Drawings
FIG. 1 shows Tertain-CNNI network parameters according to an embodiment of the present invention;
FIG. 2 shows Tertain-CNNII network parameters according to an embodiment of the present invention;
fig. 3 is a schematic diagram of an identification process according to an embodiment of the present invention.
Detailed Description
The technical solution of the present application will be described with reference to the following examples. In addition, numerous specific details are set forth below in order to provide a better understanding of the present invention. It will be understood by those skilled in the art that the present invention may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present invention.
Example (b):
a low-risk landform identification method for an outdoor autonomous mobile robot comprises the following steps:
preparation before identification: specifying a risk landscape from a plurality of landscape types; collecting a landform image, and establishing a landform image data set; training and verifying the two classification networks by using a landform image data set to obtain a multi-classification classifier and a two-classification classifier; the embodiment takes cement land, asphalt road, grass land, mud land, stone road and water surface as examples, wherein the water surface is a risk landform.
And (3) risk landform identification: and sequentially inputting the landform images to be identified into the classifier, outputting the landform images as the risk landform if the classifier identifies the risk landform, and otherwise, outputting the landform images as the low risk landform.
Wherein the process of creating a geomorphic image dataset comprises the steps of data enhancement: performing data expansion on the acquired landform image through the cross combination of cyclic shearing, mirror symmetry and gray level adjustment to obtain a data set after data enhancement; and the data set is divided into a training set, a validation set, and a test set. After the operation, the original image data can be expanded to a plurality of times to meet the requirements of training, verification and testing. The ratio of the number of images in the training set, the verification set and the test set in this embodiment is 6: 2: 2.
the training and verification process of the classification network comprises the following steps: during training, inputting n images from a training set for training each time, and inputting a verification set to test the network precision after finishing network training for m times; when there is v1Loss value of the secondary verification set does not decrease or decreases in magnitudeIf the degree is less than the threshold value, the learning rate is attenuated to half of the original learning rate until v exists2The secondary network stops training after the loss values on the validation set no longer drop, where v2 > v 1. In order to enable the network parameters after training to be close to the optimal solution, a method of dynamically attenuating the learning rate is adopted, and the effect is obvious.
The classification network of this embodiment is AlexNet, the feature extraction layers between the classification networks adopt different structures, and Terrain-cnni are respectively designed, and specific parameters are shown in fig. 1 and fig. 2, where Terrain-cnni is used for multi-classification of a landform type, and Terrain-cnni is used for two-classification of a landform type, where Terrain-cnni includes two convolution layers and two maximum pooling layers, and Terrain-cnni includes three convolution layers and three maximum pooling layers. In order to enable each network to have a higher identification rate, the network can be designed according to the currently mainstream excellent CNN network structure, and meanwhile, the convolution layer and the pooling layer of the two networks should have different structures, so that different image features can be extracted, and further double verification is realized.
As shown in fig. 3, the process of risk landform identification in this embodiment includes: inputting the acquired landform image into Tertain-CNNI to obtain an identification result, if the landform of the identification result is a risk landform, directly outputting the risk landform as a result, and if the landform in the identification result does not belong to the risk landform, inputting the landform image into Tertain-CNNI; and Tertain-CNNI judges whether the landform image is a risk landform, if so, the risk landform image is output, and if not, the identification result obtained by the multi-classification classifier is output.
The substantial effects of the present embodiment include: based on the CNN algorithm, firstly, the landform types are subjected to risk grade division, then an identification method with a 'double verification' thought is provided, the 'potential risk landforms' are identified for many times, and the landforms are identified as 'risk landforms' only by the CNN algorithm once. Through the multiple screening mechanism, the risk landform recognition rate can be obviously improved, and therefore the risk of the outdoor robot caused by the error landform recognition is reduced.
The technical solution of the embodiments of the present application may be essentially or partially contributed to the prior art, or all or part of the technical solution may be embodied in the form of a software product, where the software product is stored in a storage medium, and includes several instructions to enable a device (which may be a single chip, a chip, or the like) or a processor (processor) to execute all or part of the steps of the method of the embodiments of the present application. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (6)
1. A low-risk landform identification method of an outdoor autonomous mobile robot is characterized by comprising the following steps:
preparation before identification: specifying at least one risky landscape from a number of landscape types; collecting a landform image, and establishing a landform image data set; training and verifying at least two classification networks by using a geomorphic image data set to obtain at least one multi-classification classifier and at least one two-classification classifier;
and (3) risk landform identification: and sequentially inputting the landform images to be identified into the classifier, outputting the landform images as the risk landform if the classifier identifies the risk landform, and otherwise, outputting the landform images as the low risk landform.
2. The low-risk terrain recognition method of an outdoor autonomous mobile robot of claim 1, characterized in that the process of establishing a terrain image dataset comprises the data enhancement steps of: performing data expansion on the acquired landform image through the cross combination of cyclic shearing, mirror symmetry and gray level adjustment to obtain a data set after data enhancement; and the data set is divided into a training set, a validation set, and a test set.
3. The method for low-risk landform recognition of an outdoor autonomous mobile robot according to claim 2, characterized in that the training and verification process of the classification network comprises: during training, inputting n images from a training set for training each time, and inputting a verification set to test the network precision after finishing network training for m times; when there is v1If the loss value of the secondary verification set does not decrease or the decrease amplitude is smaller than the threshold value, the learning rate is attenuated to half of the original learning rate until v is provided2Stopping the training of the secondary network after the loss value of the secondary network on the verification set does not decrease any more, wherein v2>v1。
4. A low risk landform recognition method for an outdoor autonomous mobile robot according to claim 1 or 2, characterized in that the risk landform recognition process comprises: inputting the acquired landform image into a multi-classification classifier to obtain an identification result, directly outputting the risk landform as a result if the landform of the identification result is the risk landform, and inputting the landform image into a two-classification classifier if the landform of the identification result does not belong to the risk landform; and judging whether the landform image is a risk landform or not by the two-classification classifier, if so, outputting the risk landform image, and if not, outputting the identification result obtained by the multi-classification classifier.
5. The method of claim 4, wherein the classification network is AlexNet, and the feature extraction layers between the classification networks have different structures, wherein the classification network corresponding to the multi-classification classifier includes two convolution layers and two maximum pooling layers, and the classification network corresponding to the two-classification classifier includes three convolution layers and three maximum pooling layers.
6. The method of claim 2, wherein the ratio of the number of images in the training set, the verification set and the test set is 6: 2: 2.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011550817.8A CN112668449A (en) | 2020-12-24 | 2020-12-24 | Low-risk landform identification method for outdoor autonomous mobile robot |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011550817.8A CN112668449A (en) | 2020-12-24 | 2020-12-24 | Low-risk landform identification method for outdoor autonomous mobile robot |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112668449A true CN112668449A (en) | 2021-04-16 |
Family
ID=75410000
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011550817.8A Pending CN112668449A (en) | 2020-12-24 | 2020-12-24 | Low-risk landform identification method for outdoor autonomous mobile robot |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112668449A (en) |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
WO2020164278A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Image processing method and device, electronic equipment and readable storage medium |
-
2020
- 2020-12-24 CN CN202011550817.8A patent/CN112668449A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107748895A (en) * | 2017-10-29 | 2018-03-02 | 北京工业大学 | UAV Landing landforms image classification method based on DCT CNN models |
WO2020164278A1 (en) * | 2019-02-14 | 2020-08-20 | 平安科技(深圳)有限公司 | Image processing method and device, electronic equipment and readable storage medium |
Non-Patent Citations (2)
Title |
---|
ZHUIMENGSHAONIAN66: "训练技巧_验证集准确度震荡", pages 1 - 2, Retrieved from the Internet <URL:https://blog.csdn.net/zhuimengshaonian66/article/details /83650576> * |
张琪安等: "采用卷积神经网络的低风险可行地貌分类方法", 控制理论与应用, vol. 37, no. 9, 30 September 2020 (2020-09-30), pages 1945 - 1948 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109145939B (en) | Semantic segmentation method for small-target sensitive dual-channel convolutional neural network | |
KR102516360B1 (en) | A method and apparatus for detecting a target | |
EP3686779B1 (en) | Method and device for attention-based lane detection without post-processing by using lane mask and testing method and testing device using the same | |
CN108492281B (en) | Bridge crack image obstacle detection and removal method based on generation type countermeasure network | |
KR20200094652A (en) | Learning method and learning device, and testing method and testing device for detecting parking spaces by using point regression results and relationship between points to thereby provide an auto-parking system | |
CN112052787A (en) | Target detection method and device based on artificial intelligence and electronic equipment | |
CN112016464A (en) | Method and device for detecting face shielding, electronic equipment and storage medium | |
CN108805016B (en) | Head and shoulder area detection method and device | |
KR20200095356A (en) | Method for recognizing face using multiple patch combination based on deep neural network with fault tolerance and fluctuation robustness in extreme situation | |
CN109118504B (en) | Image edge detection method, device and equipment based on neural network | |
CN104537647A (en) | Target detection method and device | |
CN109919252A (en) | The method for generating classifier using a small number of mark images | |
CN109902018A (en) | A kind of acquisition methods of intelligent driving system test cases | |
JP6916548B2 (en) | Learning methods and devices that segment images with at least one lane using embedding loss and softmax loss to support collaboration with HD maps needed to meet level 4 of autonomous vehicles, and use them. Test method and test equipment | |
CN111476247B (en) | CNN method and device using 1xK or Kx1 convolution operation | |
WO2021144943A1 (en) | Control method, information processing device, and control program | |
JP2020119529A (en) | Method and device using masking parameters for pooling roi usable for optimizing hardware applicable to mobile devices or compact networks, and testing method and testing device using the same | |
CN110909741A (en) | Vehicle re-identification method based on background segmentation | |
CN110555461A (en) | scene classification method and system based on multi-structure convolutional neural network feature fusion | |
CN111242176B (en) | Method and device for processing computer vision task and electronic system | |
CN114626042B (en) | Face verification attack method and device | |
KR20200093418A (en) | Method and device for verifying integrity of parameters of cnn by using test pattern to enhance fault tolerance and fluctuation robustness in extreme situations for functional safety | |
CN110135428A (en) | Image segmentation processing method and device | |
CN116343159B (en) | Unstructured scene passable region detection method, device and storage medium | |
CN115294162B (en) | Target identification method, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |