CN-115223127-B - Automatic driving perception method and device for vehicle, vehicle and storage medium

CN115223127BCN 115223127 BCN115223127 BCN 115223127BCN-115223127-B

Abstract

The application relates to the technical field of intelligent application of automobiles, in particular to an automatic driving sensing method and device of a vehicle, the vehicle and a storage medium, wherein the method comprises the steps of obtaining initial picture data of surrounding environment of the vehicle; and if the initial picture data does not have the preset countercheck sample, obtaining a perception result of the initial picture data after finishing the reasoning, otherwise, omitting the preset countercheck sample of the initial picture data by using a preset denoising strategy to obtain denoised first picture data, and reasonedly obtaining the perception result of the initial picture data based on the first picture data. Therefore, the problems that in the related art, all data are subjected to denoising processing, so that the workload of data processing is large, the perception efficiency is low, noise of the input data of a perception task cannot be effectively avoided, and the like are solved.

Inventors

LUO YONGGANG
MA JINYAN

Assignees

重庆长安汽车股份有限公司

Dates

Publication Date: 20260508
Application Date: 20220729

Claims (8)

1. An automatic driving perception method of a vehicle, comprising the steps of: acquiring initial picture data of the surrounding environment of the vehicle; At least one target task is inferred according to the initial picture data, and whether a preset countermeasure sample exists in the initial picture data is detected in the inference process; If the preset countermeasures do not exist in the initial picture data, obtaining a perception result of the initial picture data after finishing reasoning, otherwise, omitting the preset countermeasures for denoising the initial picture data by using a preset denoising strategy to obtain denoised first picture data, and carrying out reasonement on the perception result of the initial picture data based on the first picture data; the step of reasoning at least one target task according to the initial picture data and detecting whether a preset countermeasure sample exists in the initial picture data in the reasoning process comprises the following steps: Inputting the initial picture data into a preset neural network of an automatic driving perception algorithm to perform reasoning of at least one target task; In the reasoning process, a multi-layer leave-one strategy is used for detecting preset countermeasure samples of a plurality of middle layers of the preset neural network, when a preset countermeasure sample exists in any middle layer, the preset countermeasure sample exists in the initial picture data, otherwise, the preset countermeasure sample does not exist in the initial picture data; denoising a preset countermeasure sample of the initial picture data by using a preset denoising strategy to obtain denoised first picture data, wherein the denoising method comprises the following steps of: inputting the picture data into a preset high-level guiding denoising device to denoise a preset countering sample, and obtaining the denoised first picture data; the preset high-rise guiding denoising device is characterized by being a characteristic guiding denoising device and a logarithmic guiding denoising device; the specific process of the multi-layer leave-on strategy is as follows: the difference between the reserve value and the original value of a single node is first recorded, Refers to the difference between a value and the original value; Leaving a value for; is the original value; Calculating the tetrad difference 。
2. The method of claim 1, further comprising, prior to reasonedly deriving a perceived result of the initial picture data based on the first picture data: continuously detecting whether the preset countermeasure sample exists in the first picture data in the reasoning process; If the preset countermeasures are not existed in the first picture data, obtaining a perception result of the initial picture data after finishing reasoning, otherwise, utilizing a preset denoising strategy to omit the preset countermeasures of the first picture data, obtaining denoised second picture data until the preset countermeasures are not existed or the number of reasoning times reaches the preset number of times, finishing the reasoning, and obtaining the perception result of the initial picture data.
3. An autopilot sensing apparatus of a vehicle for performing the method of any one of claims 1-2, comprising: The acquisition module is used for acquiring initial picture data of the surrounding environment of the vehicle; The reasoning module is used for reasoning at least one target task according to the initial picture data and detecting whether a preset countermeasure sample exists in the initial picture data in the reasoning process; And the perception module is used for obtaining a perception result of the initial picture data after finishing reasoning if the preset countermeasure sample does not exist in the initial picture data, or omitting the preset countermeasure sample of the initial picture data by using a preset denoising strategy to obtain denoised first picture data, and carrying out reasonement to obtain the perception result of the initial picture data based on the first picture data.
4. The apparatus of claim 3, wherein the inference module is further configured to: Inputting the initial picture data into a preset neural network of an automatic driving perception algorithm to perform reasoning of at least one target task; In the reasoning process, a multi-layer leave-one strategy is used for detecting preset countermeasure samples of a plurality of middle layers of the preset neural network, when a preset countermeasure sample exists in any middle layer, the preset countermeasure sample exists in the initial picture data, and otherwise, the preset countermeasure sample does not exist in the initial picture data.
5. The apparatus of claim 3, wherein the sensing module is further to: And inputting the picture data into a preset high-level guiding denoising device to perform preset countersample denoising, and obtaining the denoised first picture data.
6. The apparatus according to any one of claims 3-5, further comprising, prior to reasonedly deriving a perception result of the initial picture data based on the first picture data: the detection module is used for continuously detecting whether the preset countermeasure sample exists in the first picture data in the reasoning process; And the processing module is used for obtaining a perception result of the initial picture data after finishing reasoning if the preset countermeasures are not existed in the first picture data, otherwise, utilizing a preset denoising strategy to omit the preset countermeasures of the first picture data, obtaining denoised second picture data until the preset countermeasures are not existed or the reasoning times reach the preset times, finishing the reasoning and obtaining the perception result of the initial picture data.
7. A vehicle comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the method of autonomous driving awareness of a vehicle according to any of claims 1-2.
8. A computer-readable storage medium, on which a computer program is stored, characterized in that the program is executed by a processor for implementing the automatic driving awareness method of a vehicle according to any one of claims 1-2.

Description

Automatic driving perception method and device for vehicle, vehicle and storage medium Technical Field The application relates to the technical field of intelligent application of automobiles, in particular to an automatic driving sensing method and device of a vehicle, the vehicle and a storage medium. Background Currently, a great deal of artificial intelligence algorithms given to the neural network are adopted for the automatic driving perception task, and researches show that the neural network is not robust under the condition of tiny disturbance (namely, against sample attack), so that great potential safety hazards are brought. In the related art, denoising processing is carried out on an image to be processed to obtain a denoised image, and image data processing is carried out according to the denoised image to realize anti-sample defense and optimize the effect of image data processing. However, in the related art, all data is usually subjected to denoising processing, which not only has large workload and reduces the perception efficiency, but also cannot effectively avoid the noise problem of the input data of the perception task. Disclosure of Invention The application provides an automatic driving sensing method and device for a vehicle, the vehicle and a storage medium, and aims to solve the problems that in the related art, generally, all data are subjected to denoising processing, so that the workload of data processing is large, the sensing efficiency is low, noise of sensing task input data cannot be effectively avoided, and the like. An embodiment of a first aspect of the application provides an automatic driving perception method of a vehicle, which comprises the following steps of obtaining initial picture data of a surrounding environment of the vehicle, deducing at least one target task according to the initial picture data, detecting whether a preset countermeasures sample exists in the initial picture data in a reasoning process, obtaining a perception result of the initial picture data after finishing the reasoning if the preset countermeasures sample does not exist in the initial picture data, otherwise, eliminating the preset countermeasures sample of the initial picture data by using a preset denoising strategy, obtaining denoised first picture data, and reasonedly obtaining the perception result of the initial picture data based on the first picture data. According to the technical means, the image data is subjected to the countersample detection, the image data is subjected to the denoising processing based on the countersample, so that accurate denoising is realized, huge workload caused by unified denoising of all data is avoided, the workload of denoising processing is reduced, the processing efficiency is improved, and the perception efficiency is improved at the same time, so that the countersample detection and denoising are combined, the problem of noise of perception task input data is systematically solved, the reinforcement of an automatic driving perception system is effectively realized, and the interference of noise to the input data is effectively resisted. Optionally, in one embodiment of the present application, the step of reasoning at least one target task according to the initial picture data and detecting whether a preset challenge sample exists in the initial picture data in a reasoning process includes the steps of inputting the initial picture data into a preset neural network of an autopilot sensing algorithm to perform reasoning of the at least one target task, and in the reasoning process, using a multi-layer leave-one strategy to perform preset challenge sample detection on a plurality of middle layers of the preset neural network, and determining that the preset challenge sample exists in the initial picture data when a preset challenge sample exists in any middle layer, otherwise determining that the preset challenge sample does not exist in the initial picture data. According to the technical means, the method and the device can input the initial picture data of the surrounding environment of the vehicle into the plurality of middle layers in the neural network of the automatic driving perception algorithm to detect the countermeasure sample, judge whether the initial picture data has the preset countermeasure sample or not, so that the problem of unnecessary denoising of the data without noise is avoided, and the robust reinforcement of the perception algorithm is realized. Optionally, in an embodiment of the present application, denoising the preset challenge sample of the initial picture data by using a preset denoising policy to obtain denoised first picture data includes inputting the picture data into a preset high-level guiding denoising device to denoise the preset challenge sample, so as to obtain denoised first picture data. According to the technical means, when the preset countermeasure sample exists in the initial picture data, the p