CN112329254A - Automatic driving method for butting simulation environment image and real environment image - Google Patents
Automatic driving method for butting simulation environment image and real environment image Download PDFInfo
- Publication number
- CN112329254A CN112329254A CN202011266665.9A CN202011266665A CN112329254A CN 112329254 A CN112329254 A CN 112329254A CN 202011266665 A CN202011266665 A CN 202011266665A CN 112329254 A CN112329254 A CN 112329254A
- Authority
- CN
- China
- Prior art keywords
- image
- vehicle
- semantic segmentation
- segmentation network
- automatic driving
- 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
- 238000004088 simulation Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 41
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 53
- 230000011218 segmentation Effects 0.000 claims abstract description 48
- 238000013135 deep learning Methods 0.000 claims abstract description 47
- 230000002787 reinforcement Effects 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 41
- 230000008569 process Effects 0.000 claims description 13
- 238000003032 molecular docking Methods 0.000 claims description 9
- 238000012805 post-processing Methods 0.000 claims description 9
- 238000002372 labelling Methods 0.000 claims description 7
- 230000000694 effects Effects 0.000 claims description 5
- 238000001914 filtration Methods 0.000 claims description 4
- 238000005070 sampling Methods 0.000 claims description 4
- 238000003709 image segmentation Methods 0.000 abstract 1
- 238000004364 calculation method Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 210000001503 joint Anatomy 0.000 description 3
- 241000282414 Homo sapiens Species 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000018109 developmental process Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30248—Vehicle exterior or interior
- G06T2207/30252—Vehicle exterior; Vicinity of vehicle
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Traffic Control Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an automatic driving method for butting a simulation environment image and a real environment image. Firstly, acquiring a vehicle camera image in a simulation platform, and training a deep learning semantic segmentation network 1; secondly, performing image segmentation on the simulation platform by adopting a trained deep learning semantic segmentation network 1, and extracting a characteristic value; then, training a deep reinforcement learning algorithm model by adopting the extracted characteristic values to realize that the vehicle runs according to an appointed lane in a simulation environment; and finally, acquiring a vehicle camera image in a real environment, training the deep learning semantic segmentation network 2, and realizing automatic driving of the vehicle on an actual lane. The invention obviously reduces the difficulty of transferring the trained algorithm model in the simulation environment to the real environment, so that the reinforcement learning algorithm trained in the simulation environment can be directly used on a real vehicle without additional reinforcement learning training.
Description
Technical Field
The invention belongs to the field of automatic driving, and particularly relates to an automatic driving method applying a deep reinforcement learning algorithm.
Background
With the continuous development of the artificial intelligence technology, the problem of automatic driving is solved more realistically by applying the artificial intelligence technology in the automobile industry. At present, automatic driving is mainly realized in two directions, wherein one direction is to decompose an automatic driving task and use different algorithms to solve the decomposed problem, for example, deep learning is used for detecting and identifying road information; the other direction is to realize automatic driving by directly using an end-to-end algorithm without subdivision solution of automatic driving, such as reinforcement learning and imitation learning. The end-to-end algorithm can simplify the design of automatic driving theoretically, after all, automatic driving is realized from a sensing section to a control end, the work content is reduced by a mode of decomposing problems, meanwhile, the method also conforms to the driving mode of human, information is obtained from the sensing end image, control is realized from the control end, accurate surrounding information does not need to be known, and therefore, a plurality of enterprises and students are engaged in end-to-end automatic driving research. The reinforcement learning also belongs to an end-to-end automatic driving algorithm in automatic driving, and the principle of the reinforcement learning also conforms to the evolution mode of human beings, namely punishment is obtained from interaction with the environment, and knowledge is learned. After the reinforcement learning is combined with the deep learning, the deep reinforcement learning is generated, and the performance is obviously improved.
Although the deep reinforcement learning algorithm can learn new knowledge in the environment, the current development of the algorithm is still insufficient to enable the algorithm to learn the wanted knowledge in the complex vehicle-mounted camera original image, and the algorithm depends on a manually set reward function, and the setting of the function limits the learning effect of the reinforcement learning algorithm. Meanwhile, the deep reinforcement learning algorithm has the problem of low sample utilization rate, and a large amount of repeated training is needed to learn the knowledge desired by people. In a real environment, the time cost and the vehicle cost are high and a great safety risk exists when a real vehicle is used for deep reinforcement learning training.
In order to reduce the training cost and risk of the reinforcement learning algorithm, researchers generally adopt a mode of training in a simulation environment to reduce the training cost and risk, and after a model is trained, the model is migrated to a real environment. In the field of automatic driving, an obvious problem exists in doing so, namely, an image in a simulation environment is obviously different from a real environment, so that a result of training by using an original image in the simulation environment by a deep reinforcement learning algorithm is difficult to transfer to the real environment, and the deep reinforcement learning algorithm needs to be retrained again.
At present, researchers reduce training differences between a simulation environment and a real environment mainly by establishing a vivid simulation environment and designing a conversion algorithm of images in the simulation environment and the real environment, the methods can reduce the differences between the two environments, but the reduction is limited, and the methods all need to spend a large amount of time.
Disclosure of Invention
In order to solve the technical problems mentioned in the background art, the invention provides an automatic driving method for butting a simulation environment image and a real environment image, and the difficulty of transferring a trained algorithm model in a simulation environment to the real environment is obviously reduced.
In order to achieve the technical purpose, the technical scheme of the invention is as follows:
an automatic driving method for docking a simulation environment image and a real environment image is characterized by comprising the following steps:
(1) acquiring a vehicle camera image in a simulation platform, labeling a region capable of driving in the image, and training a deep learning semantic segmentation network 1 by adopting a labeled image set;
(2) segmenting a real-time image acquired by a vehicle camera in a simulation platform by adopting the deep learning semantic segmentation network 1 trained in the step (1), post-processing the segmented image, and extracting a characteristic value;
(3) training a deep reinforcement learning algorithm model by adopting the characteristic values extracted in the step (2) to realize that the vehicle runs according to a specified lane in a simulation environment;
(4) and (3) acquiring a vehicle camera image in a real environment, training the deep learning semantic segmentation network 2, and realizing automatic driving of the vehicle on a practical lane by adopting the characteristic value of the image in the real environment extracted by the deep learning semantic segmentation network 2 and the deep reinforcement learning algorithm model trained in the step (3).
Further, the specific process of step (1) is as follows:
(1-1) in a simulation platform, a vehicle runs along a correct lane, and a forward-looking image is acquired by using a vehicle-mounted camera in the running process of the vehicle;
(1-2) labeling areas capable of driving in the image acquired in the step (1-1) and storing the areas in a data set form;
and (1-3) selecting a proper deep learning semantic segmentation network, and training the deep learning semantic segmentation network 1 by utilizing the data set obtained in the step (1-2) until the accuracy of the network meets the requirement.
Further, in the step (1-2), the label is marked in a manual mode.
Further, in the step (1-3), the deep learning semantic segmentation network is selected according to the real-time performance of the network.
Further, in step (1-3), the selected deep learning semantic segmentation network includes, but is not limited to, Bisenet.
Further, in the step (1-3), training the deep learning semantic segmentation network until the accuracy of the network exceeds a set threshold value.
Further, the specific process of step (2) is as follows:
(2-1) inputting real-time images acquired by a vehicle camera in the simulation platform into a trained deep learning semantic segmentation network 1, and outputting segmented semantic images by the deep learning semantic segmentation network 1;
and (2-2) carrying out post-processing on the segmented semantic image to obtain a characteristic value.
Further, in the step (2-2), the post-processing includes down-sampling to reduce the computation amount of the network, filtering inefficient feature information, and image cropping to delete inefficient information and ineffective information; the low-efficiency information is information which has no obvious effect on the running of the vehicle on the road and comprises a green belt, a sidewalk and objects on the sidewalk; the invalid information is completely useless information when the vehicle runs on the road, and comprises background information and interference information, wherein the background information comprises sky, buildings and trees outside the road, and the interference information comprises fallen leaves of the road surface and the difference of light rays.
Further, the specific process of step (3) is as follows:
(3-1) butting the deep reinforcement learning algorithm model with the deep learning semantic segmentation network 1 and a simulation platform, and training the deep reinforcement learning algorithm model through the characteristic values extracted in the step (2);
(3-2) continuously adjusting and optimizing the model to enable the training result to reach the aim that the vehicle can run along the specified lane;
and (3-3) outputting a control signal of automatic driving of the vehicle to the simulation platform by the deep reinforcement learning algorithm, and driving the vehicle in the simulation platform according to the received control signal.
Further, the specific process of step (4) is as follows:
(4-1) collecting a vehicle camera image in a real environment, marking the image in the same way as the step (1), and training a deep learning semantic segmentation network 2 by using the marked image set;
(4-2) extracting the characteristic value of the image in the real environment in the same way as the step (2) based on the deep learning semantic segmentation network 2 trained in the step (4-1);
and (4-3) using the deep reinforcement learning algorithm model trained in the step (3) and the characteristic value extracted in the step (4-2) to realize automatic driving of the vehicle on the actual lane.
Adopt the beneficial effect that above-mentioned technical scheme brought:
the method has the advantages that the key point of the method is the butt joint of the simulation environment and the real environment, the key point of the method is that the deep reinforcement learning algorithm is trained by using the characteristic values extracted by the semantic segmentation network, and the efficiency of realizing automatic driving by the deep reinforcement learning algorithm is improved through the two aspects. The invention greatly reduces the difficulty of transferring the trained algorithm model in the simulation environment to the real environment by designing the images used in the algorithm training: the conventional image of the camera used in the deep reinforcement learning training is changed into the semantic segmentation image, invalid and inefficient image information in the reinforcement learning algorithm training automatic driving is shielded, only road information required by the vehicle is reserved, and the information is processed, so that the difficulty of transferring the trained algorithm in the simulation environment to the real environment can be obviously reduced, the process of carrying out massive retraining on the actual vehicle by the deep reinforcement learning algorithm is avoided, and a large amount of time and money cost are saved. Through actual tests, the algorithm trained in the simulation environment can realize automatic driving without retraining in a real environment.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a schematic diagram of the feature extraction results of the straight-driving lane simulation environment and the real environment in the present invention;
fig. 3 is a schematic diagram of the feature extraction results of the turning lane simulation environment and the real environment in the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below are exemplary embodiments with reference to the drawings, are only for illustrating the technical idea of the present invention, and the scope of the present invention should not be limited thereby, and any modifications made on the technical solution based on the technical idea of the present invention are within the scope of the present invention
As used herein, the singular forms "a", "an", "the" and "the" may include the plural forms as well, unless expressly stated otherwise. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or coupled. As used herein, the term "and/or" includes any and all combinations of one or more of the associated listed items.
It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.
The invention provides an automatic driving method for butting a simulation environment image and a real environment image, which comprises the following steps as shown in figure 1:
step 1: acquiring a vehicle camera image in a simulation platform, labeling a region capable of driving in the image, and training a deep learning semantic segmentation network 1 by adopting a labeled image set;
step 2: adopting the deep learning semantic segmentation network 1 trained in the step 1 to segment real-time images acquired by a vehicle camera in a simulation platform, carrying out post-processing on the segmented images, and extracting characteristic values;
and step 3: training a deep reinforcement learning algorithm model by adopting the characteristic values extracted in the step (2) to realize that the vehicle runs according to an appointed lane in a simulation environment;
and 4, step 4: and (3) acquiring a vehicle camera image in a real environment, training the deep learning semantic segmentation network 2, and realizing automatic driving of the vehicle on a practical lane by adopting the characteristic value of the image in the real environment extracted by the deep learning semantic segmentation network 2 and the deep reinforcement learning algorithm model trained in the step 3.
In this embodiment, the step 1 can be implemented by the following preferred scheme:
step 1-1: in the simulation platform, a vehicle runs along a correct lane, and a forward-looking image is acquired by using a vehicle-mounted camera in the running process of the vehicle;
step 1-2: labeling areas capable of driving in the image acquired in the step 1-1, and storing the areas into a data set form;
step 1-3: and (3) selecting a proper deep learning semantic segmentation network 1, and training the deep learning semantic segmentation network 1 by utilizing the data set obtained in the step (1-2) until the accuracy of the network meets the requirement.
Further, in step 1-2, the label is manually applied.
Further, in step 1-3, a deep learning semantic segmentation network, such as Bisenet, is selected according to the real-time performance of the network.
At present, more than ten public algorithms of a semantic segmentation network in deep learning used for segmenting images can be rapidly realized. The criteria for selecting a network is that the real-time performance is good, for example: under the condition of not using an acceleration tool TensorRT provided by NVIDIA official, the time for processing a single 1024 x 1024 color image on NVIDIA GeForce RTX 2080 is not more than 30ms, the segmentation accuracy is not lower than the level of the current mainstream algorithm, and the calculation cost of the algorithm is low.
Further, in step 1-3, training the deep learning semantic segmentation network until the accuracy of the network exceeds a set threshold, which is adjusted according to the actual situation.
In this embodiment, the step 2 can be implemented by the following preferred scheme:
step 2-1: inputting real-time images acquired by a vehicle camera in a simulation platform into a trained deep learning semantic segmentation network 1, and outputting segmented semantic images by the deep learning semantic segmentation network 1;
step 2-2: and carrying out post-processing on the segmented semantic image to obtain a characteristic value.
Further, in step 2-2, the post-processing includes down-sampling to reduce the computation amount of the network, filtering inefficient feature information, and image cropping to delete inefficient information and ineffective information; the low-efficiency information is information which has no obvious effect on the running of the vehicle on the road and comprises a green belt, a sidewalk and objects on the sidewalk; the invalid information is completely useless information when the vehicle runs on the road, and comprises background information and interference information, wherein the background information comprises sky, buildings and trees outside the road, and the interference information comprises fallen leaves of the road surface and the difference of light rays.
The purpose of downsampling is as follows: the length and width of the segmented image usually exceed 1024 × 1024, which means that the segmented image contains a large amount of repeated information, on one hand, the large amount of repeated information increases the calculation times/calculation amount of the network, meanwhile, the repeated information may cause adverse effects, and the deep reinforcement learning algorithm may explore whether the repeated information has special meanings. The length and width after down sampling is less than 100 multiplied by 100, the number grade is reduced by more than 100 times, and the training complexity and the time consumption can be obviously reduced.
After downsampling and filtering of ineffective and invalid information, the remaining characteristic value is useful information for deep reinforcement learning algorithm training, on one hand, the training amount and the calculation amount can be reduced by reducing the number, so that the efficiency is improved, on the other hand, the learning and exploring range is greatly reduced during deep reinforcement learning algorithm training after the ineffective and invalid information is reduced, and the training efficiency can also be improved.
In this embodiment, the step 3 can be implemented by the following preferred scheme:
step 3-1: the deep reinforcement learning algorithm model is in butt joint with the deep learning semantic segmentation network 1 and the simulation platform, and the deep reinforcement learning algorithm model is trained through the characteristic values extracted in the step 2;
step 3-2: continuously adjusting and optimizing the deep reinforcement learning algorithm model to enable the training result to reach the aim that the vehicle can run along the specified lane;
step 3-3: and outputting a control signal of automatic driving of the vehicle to the simulation platform by the deep reinforcement learning algorithm, and driving the vehicle in the simulation platform according to the received control signal.
In this embodiment, the step 4 can be implemented by the following preferred scheme:
step 4-1: collecting vehicle camera images in a real environment, marking the images in the same way as the step 1, and training a deep learning semantic segmentation network 2 by using the marked image set;
step 4-2: extracting the characteristic value of the image in the real environment by adopting the same mode as the step 2 based on the deep learning semantic segmentation network 2 trained in the step 4-1;
step 4-3: and 3, using the deep reinforcement learning algorithm model trained in the step 3 and the characteristic value extracted in the step 4-2 to realize automatic driving of the vehicle on the actual lane.
As shown in fig. 2 and 3, the results of extracting the features of the straight lane and the turning lane are respectively shown in the schematic diagrams, where (a) in fig. 2 and 3 is the result of extracting the feature value of the lane in the simulation environment, (b) is the result of extracting the feature value of the lane in the real environment, the white area in the diagram is the extracted feature information of the lane, and the black area is the information unrelated to the lane. It can be seen that the lane characteristic value extracted in the simulation environment is almost the same as the characteristic value extracted in the actual environment, that is, the deep reinforcement learning algorithm model trained by using the image characteristic value in the simulation environment can directly use the characteristic value extracted by using the image in the real environment to complete the image butt joint work in the simulation environment and the real environment, thereby realizing automatic driving.
Those of skill in the art will appreciate that the various operations, methods, steps in the processes, acts, or solutions discussed in the present application can be interchanged, modified, combined, or eliminated. Further, various operations, methods, steps in the flows, which have been discussed in the present application, may be interchanged, modified, rearranged, decomposed, combined, or eliminated. Further, steps, measures, schemes in the various operations, methods, procedures disclosed in the prior art and the present invention can also be alternated, changed, rearranged, decomposed, combined, or deleted.
Claims (10)
1. An automatic driving method for docking a simulation environment image and a real environment image is characterized by comprising the following steps:
(1) acquiring a vehicle camera image in a simulation platform, labeling a region capable of driving in the image, and training a deep learning semantic segmentation network 1 by adopting a labeled image set;
(2) segmenting a real-time image acquired by a vehicle camera in a simulation platform by adopting the deep learning semantic segmentation network 1 trained in the step (1), post-processing the segmented image, and extracting a characteristic value;
(3) training a deep reinforcement learning algorithm model by adopting the characteristic values extracted in the step (2) to realize that the vehicle runs according to a specified lane in a simulation environment;
(4) and (3) acquiring a vehicle camera image in a real environment, training the deep learning semantic segmentation network 2, and realizing automatic driving of the vehicle on a practical lane by adopting the characteristic value of the image in the real environment extracted by the deep learning semantic segmentation network 2 and the deep reinforcement learning algorithm model trained in the step (3).
2. The automatic driving method for docking simulation environment image and real environment image according to claim 1, wherein the specific process of step (1) is as follows:
(1-1) in a simulation platform, a vehicle runs along a correct lane, and a forward-looking image is acquired by using a vehicle-mounted camera in the running process of the vehicle;
(1-2) labeling areas capable of driving in the image acquired in the step (1-1) and storing the areas in a data set form;
and (1-3) selecting a proper deep learning semantic segmentation network, and training the deep learning semantic segmentation network 1 by utilizing the data set obtained in the step (1-2) until the accuracy of the network meets the requirement.
3. The automatic driving method for docking simulation environment image with real environment image according to claim 2, wherein in the step (1-2), labeling is performed manually.
4. The automated driving method for interfacing an image of a simulated environment with an image of a real environment according to claim 2, wherein in step (1-3), a deep learning semantic segmentation network is selected according to the real-time nature of the network.
5. The method for automated driving of an interface between an image of a simulated environment and an image of a real environment according to claim 4, wherein in step (1-3) the selected deep learning semantic segmentation network includes, but is not limited to, Bisenet.
6. The automatic driving method for docking simulation environment image and real environment image according to claim 2, wherein in the step (1-3), the deep learning semantic segmentation network is trained until the accuracy of the network exceeds a set threshold.
7. The automatic driving method for docking simulation environment image and real environment image according to claim 2, wherein the specific process of step (2) is as follows:
(2-1) inputting real-time images acquired by a vehicle camera in the simulation platform into a trained deep learning semantic segmentation network 1, and outputting segmented semantic images by the deep learning semantic segmentation network 1;
and (2-2) carrying out post-processing on the segmented semantic image to obtain a characteristic value.
8. The automatic driving method for docking simulation environment image with real environment image according to claim 7, wherein in the step (2-2), the post-processing comprises down-sampling to reduce the amount of computation of the network, filtering inefficient feature information, and image cropping to delete inefficient information and ineffective information; the low-efficiency information is information which has no obvious effect on the running of the vehicle on the road and comprises a green belt, a sidewalk and objects on the sidewalk; the invalid information is completely useless information when the vehicle runs on the road, and comprises background information and interference information, wherein the background information comprises sky, buildings and trees outside the road, and the interference information comprises fallen leaves of the road surface and the difference of light rays.
9. The automatic driving method for docking simulation environment image and real environment image according to claim 2, wherein the specific process of step (3) is as follows:
(3-1) butting the deep reinforcement learning algorithm model with the deep learning semantic segmentation network 1 and a simulation platform, and training the deep reinforcement learning algorithm model through the characteristic values extracted in the step (2);
(3-2) continuously adjusting and optimizing the model to enable the training result to reach the aim that the vehicle can run along the specified lane;
and (3-3) outputting a control signal of automatic driving of the vehicle to the simulation platform by the deep reinforcement learning algorithm, and driving the vehicle in the simulation platform according to the received control signal.
10. The automatic driving method for docking simulation environment image and real environment image according to claim 2, wherein the specific process of step (4) is as follows:
(4-1) collecting a vehicle camera image in a real environment, marking the image in the same way as the step (1), and training a deep learning semantic segmentation network 2 by using the marked image set;
(4-2) extracting the characteristic value of the image in the real environment in the same way as the step (2) based on the deep learning semantic segmentation network 2 trained in the step (4-1);
and (4-3) using the deep reinforcement learning algorithm model trained in the step (3) and the characteristic value extracted in the step (4-2) to realize automatic driving of the vehicle on the actual lane.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011266665.9A CN112329254A (en) | 2020-11-13 | 2020-11-13 | Automatic driving method for butting simulation environment image and real environment image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202011266665.9A CN112329254A (en) | 2020-11-13 | 2020-11-13 | Automatic driving method for butting simulation environment image and real environment image |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112329254A true CN112329254A (en) | 2021-02-05 |
Family
ID=74317695
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202011266665.9A Pending CN112329254A (en) | 2020-11-13 | 2020-11-13 | Automatic driving method for butting simulation environment image and real environment image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112329254A (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109101907A (en) * | 2018-07-28 | 2018-12-28 | 华中科技大学 | A kind of vehicle-mounted image, semantic segmenting system based on bilateral segmentation network |
CN110647839A (en) * | 2019-09-18 | 2020-01-03 | 深圳信息职业技术学院 | Method and device for generating automatic driving strategy and computer readable storage medium |
CN110795821A (en) * | 2019-09-25 | 2020-02-14 | 的卢技术有限公司 | Deep reinforcement learning training method and system based on scene differentiation |
CN110837811A (en) * | 2019-11-12 | 2020-02-25 | 腾讯科技(深圳)有限公司 | Method, device and equipment for generating semantic segmentation network structure and storage medium |
-
2020
- 2020-11-13 CN CN202011266665.9A patent/CN112329254A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109101907A (en) * | 2018-07-28 | 2018-12-28 | 华中科技大学 | A kind of vehicle-mounted image, semantic segmenting system based on bilateral segmentation network |
CN110647839A (en) * | 2019-09-18 | 2020-01-03 | 深圳信息职业技术学院 | Method and device for generating automatic driving strategy and computer readable storage medium |
CN110795821A (en) * | 2019-09-25 | 2020-02-14 | 的卢技术有限公司 | Deep reinforcement learning training method and system based on scene differentiation |
CN110837811A (en) * | 2019-11-12 | 2020-02-25 | 腾讯科技(深圳)有限公司 | Method, device and equipment for generating semantic segmentation network structure and storage medium |
Non-Patent Citations (1)
Title |
---|
周苏;易然;郑淼;: "基于知识蒸馏的车辆可行驶区域分割算法研究", 汽车技术, no. 01 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110210551B (en) | Visual target tracking method based on adaptive subject sensitivity | |
CN109800736A (en) | A kind of method for extracting roads based on remote sensing image and deep learning | |
CN109635744A (en) | A kind of method for detecting lane lines based on depth segmentation network | |
CN105069779B (en) | A kind of architectural pottery surface detail pattern quality detection method | |
CN113240691A (en) | Medical image segmentation method based on U-shaped network | |
CN112837344B (en) | Target tracking method for generating twin network based on condition countermeasure | |
CN104392212A (en) | Method for detecting road information and identifying forward vehicles based on vision | |
CN110009095A (en) | Road driving area efficient dividing method based on depth characteristic compression convolutional network | |
CN113902915A (en) | Semantic segmentation method and system based on low-illumination complex road scene | |
CN112488025B (en) | Double-temporal remote sensing image semantic change detection method based on multi-modal feature fusion | |
CN111414954B (en) | Rock image retrieval method and system | |
CN108304786A (en) | A kind of pedestrian detection method based on binaryzation convolutional neural networks | |
CN110009648A (en) | Trackside image Method of Vehicle Segmentation based on depth Fusion Features convolutional neural networks | |
CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
CN110503613A (en) | Based on the empty convolutional neural networks of cascade towards removing rain based on single image method | |
CN111476285B (en) | Training method of image classification model, image classification method and storage medium | |
CN114048822A (en) | Attention mechanism feature fusion segmentation method for image | |
CN109766823A (en) | A kind of high-definition remote sensing ship detecting method based on deep layer convolutional neural networks | |
CN110276378A (en) | The improved method that example is divided based on unmanned technology | |
CN110490189A (en) | A kind of detection method of the conspicuousness object based on two-way news link convolutional network | |
CN112419333A (en) | Remote sensing image self-adaptive feature selection segmentation method and system | |
CN111046723B (en) | Lane line detection method based on deep learning | |
CN112598684A (en) | Open-pit area ground feature segmentation method based on semantic segmentation technology | |
CN108961270B (en) | Bridge crack image segmentation model based on semantic segmentation | |
CN105893941A (en) | Facial expression identifying method based on regional images |
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 |