CN111626334B

CN111626334B - Key control target selection method for vehicle-mounted advanced auxiliary driving system

Info

Publication number: CN111626334B
Application number: CN202010351930.7A
Authority: CN
Inventors: 杨航; 张达睿; 王宁; 付垚; 胡铭旭
Original assignee: Dongfeng Motor Corp
Current assignee: Dongfeng Motor Corp
Priority date: 2020-04-28
Filing date: 2020-04-28
Publication date: 2023-07-14
Anticipated expiration: 2040-04-28
Also published as: CN111626334A

Abstract

The invention relates to the technical field of automobile control, in particular to a key control target selection method of a vehicle-mounted advanced auxiliary driving system. Training a standard classifier based on the multivariate information fusion module; the method comprises the steps of performing lowering treatment on a standard threshold value of a standard classifier to obtain a first classifier, performing raising treatment on the standard threshold value of the standard classifier to obtain a second classifier, wherein the threshold value of the first classifier is lower than the standard threshold value, and the threshold value of the second classifier is higher than the standard threshold value; when a front vehicle tracking running function is started, a first classifier is utilized to select a key control target; when the anti-collision active braking function of the vehicle is started, the second classifier is utilized to select a key control target. The classifier is applied to the key control target selection, and the classifier with different thresholds is obtained in a mode of regulating down and regulating up, so that the purposes of improving the user experience of the vehicle-mounted advanced auxiliary driving system and enhancing the comfort and the safety are achieved.

Description

Key control target selection method for vehicle-mounted advanced auxiliary driving system

Technical Field

The invention relates to the technical field of automobile control, in particular to a key control target selection method of a vehicle-mounted advanced auxiliary driving system.

Background

In advanced assisted driving systems, to reduce missed detection and tracking continuity of targets, multiple sensors are typically used to detect traffic-engaging targets. For the perception fusion module of the advanced driving assistance system, according to the information quality fed back by the multisensor, the space position and the physical attribute of the targets can be output, and meanwhile, the existence probability of each target is also output, and the probability of the real existence of the object is generally expressed by the percentage, so that the multisensor fusion module can more easily correlate and manage the tracks of the targets respectively.

For the control system of the advanced auxiliary driving system, since the key control target directly triggers the control action, it is necessary to normalize the existence probability of the key control target to determine whether the selected control target should cause the subsequent action of the planning control module. The multisource sensor fusion module needs to fuse information from multiple sensors, but because of different sources, the measurement dimensions and standards of the information are not uniform, so that one standard cannot be used for judging whether a target really exists. And the probability of missing a target (indicating a truly existing target but not correctly detected by a sensor) and the probability of false detection of a target (indicating a truly non-existing target but incorrectly detected by a sensor) are not the same for different control actions. For comfort and convenience functions, such as ACC (Automated Cruces Control) or TJA (traffic Jam Assistance), if the front vehicle is actually present, the function degradation caused by frequent loss of the front vehicle needs to be avoided, so that poor user experience is caused; however, if the preceding vehicle does not exist for the safety function, for example, AEB (Automated Emergency Braking), it is necessary to avoid false detection of the preceding vehicle without the preceding vehicle, causing an emergency braking operation and causing a user complaint.

Disclosure of Invention

Aiming at the defects of the prior art, the invention provides a key control target selection method of a vehicle-mounted advanced auxiliary driving system, which adopts different standards for different control actions to judge the existence of a target object, so that the undetected probability and the false detection probability of the target object meet the functional requirements of the control actions.

The technical scheme of the invention is as follows: training a standard classifier for judging whether a target object exists or not based on a multivariate information fusion module, and setting a standard threshold value of the standard classifier;

the standard threshold value of the standard classifier is subjected to lowering processing to obtain at least one first classifier, the standard threshold value of the standard classifier is subjected to raising processing to obtain at least one second classifier, the threshold value of the first classifier is lower than the standard threshold value, and the threshold value of the second classifier is higher than the standard threshold value;

when a front vehicle tracking running function is started, selecting a key control target by using the first classifier;

and when the anti-collision active braking function of the vehicle is started, selecting a key control target by utilizing the second classifier.

More preferably, the training the standard classifier for determining whether the target object exists based on the multivariate information fusion module includes:

extracting attribute parameters of each target object based on the multivariate information fusion module, and combining the attribute parameters in parallel to form a training vector;

determining whether each target object exists or not in a manual labeling mode, and forming labels corresponding to the target objects one by one;

and training to obtain a standard classifier by using a machine learning mode based on the training vector and the label.

Preferably, the attribute parameters include:

target intensity information output by a radar, image quality information output by a camera, radar quantity information output by a clustering link, tracking state, tracking period, loss period and target position covariance matrix of a target output by data association and Kalman filtering.

Preferably, the machine learning includes any one of neural network, support vector machine and mean value clustering.

The beneficial effects of the invention are as follows: the classifier is applied to the key control target selection, the classifier with different thresholds is obtained through the mode of lowering and raising, and the key control target selection mode is adopted by different classifiers according to different functional requirements, so that the missed detection probability and the false detection probability of the target object meet respective tolerance conditions, and the purposes of improving the user experience of the vehicle-mounted advanced auxiliary driving system and enhancing the comfort and the safety are achieved. The effective information of each link is extracted through the multi-element information fusion framework to form training vectors, and the existence of the target object is marked in a manual marking mode, so that the standard classifier learned by the machine has higher accuracy.

Drawings

FIG. 1 is a schematic flow chart of the present invention;

FIG. 2 is a schematic diagram of an implementation of the standard classifier of the present invention;

FIG. 3 is a diagram of a system architecture of the present invention;

FIG. 4 is a schematic illustration of ACC following target selection;

fig. 5 is an AEB braking target selection schematic.

Detailed Description

The invention will now be described in further detail with reference to the drawings and specific examples, which are given for clarity of understanding and are not to be construed as limiting the invention.

As shown in fig. 1, a key control target selection method of the vehicle-mounted advanced auxiliary driving system comprises the following steps:

extracting attribute parameters of each target object based on a multivariate information fusion module, combining the attribute parameters in parallel to form a training vector, and carrying out normalization processing on the training vector;

determining the existence of each target object through a manual labeling mode (namely, human eye observation) to form labels corresponding to the target objects one by one;

training to obtain a standard classifier by using a machine learning mode based on the training vector and the label, and setting a standard threshold value of the standard classifier;

the method comprises the steps of performing lowering treatment on a standard threshold value of a standard classifier to obtain at least one first classifier, performing raising treatment on the standard threshold value of the standard classifier to obtain at least one second classifier, wherein the threshold value of the first classifier is lower than the standard threshold value, and the threshold value of the second classifier is higher than the standard threshold value;

when a front vehicle tracking running function is started, a first classifier is utilized to select a key control target;

when the anti-collision active braking function of the vehicle is started, the second classifier is utilized to select a key control target.

In the machine learning process, the larger the data collection amount is, the more accurate the classifier is. Machine learning includes neural networks, support vector machines, mean clustering, but is not limited to the above. Non-linear compression methods that compress the results of training vectors into the 0 to 1 space, including but not limited to logistic regression, should be protected by the present patent, such as the tanh method.

As shown in fig. 2, the attribute parameters include: target intensity information output by a radar, image quality information output by a camera, radar quantity information output by a clustering link, tracking state, tracking period, loss period and target position covariance matrix of a target output by data association and Kalman filtering. The tracking state of the target comprises a new single radar target, a new single camera, a new radar and camera generated target, a temporary lost target, a radar tracking target, a camera tracking target, a completely matched target and the like. The information is combined in parallel to form the training vector, so that the training vector contains rich target information, and a more accurate confidence coefficient model is formed.

As shown in fig. 3, the architecture of the key control target selection method of the vehicle-mounted advanced auxiliary driving system according to the present embodiment includes: the system comprises a radar, a camera, a multi-element information fusion module, a key control target selection module and a vehicle control module. The radar is used for acquiring radar point clouds, generating radar targets through clustering, performing data association with camera targets acquired by a camera, and performing fusion target processing after Kalman filtering. And each attribute parameter output by the multivariate information fusion module forms a training vector, and the training vector and the manually marked label can be subjected to machine learning to obtain a standard classifier. The classifier 1 and the classifier 2 are obtained by respectively turning down and turning up the standard separators. The key control target selection module can select to adopt the classifier 1 or the classifier 2 to select the key control target according to the function of starting the vehicle, so that the vehicle control module can realize corresponding control based on the key control target.

Example 1

As shown in fig. 4, when the vehicle turns on the front vehicle tracking travel function, it is necessary to make the multi-sensor fusion system easier to recognize the front vehicle. Under certain conditions of uneven illumination, limited view, too strong or too weak illumination, the confidence of the preceding vehicle is generated by a confidence model, which is below the standard classifier threshold, but should identify this target as a following target. For the perception detection target in the ACC interest area, the classifier 1 is used, so that the miss rate of the following critical control target is reduced, the vehicle is easier to track the front vehicle, and after the front vehicle is identified and tracked, the critical target is not easy to lose to cause function degradation, and the probability of correctly identifying the front vehicle under the condition of the front vehicle is improved.

Example two

As shown in fig. 5, when the vehicle starts the anti-collision active braking function, the multi-sensor fusion system needs to more accurately identify the target vehicle, so as to avoid the situation that some illumination is uneven, the vision is blocked and limited, the illumination is too strong or too weak, or the radar is misidentified and ghosted due to the reflection of the ground and the well lid. Because the confidence coefficient of the front vehicle is generated by a confidence coefficient model and is above the threshold value of the standard classifier, but the target is caused by a well cover or other clutters, in order to avoid the braking action of the self vehicle, the perception detection target in the AEB interest area is processed by using the classifier 2, so that the probability of braking caused by false detection of the uncertain target is reduced.

To achieve this goal, we use classifier 2 to reduce the false detection rate of the brake key control target, increase the accuracy rate, and increase the probability of the actual front vehicle existence when the braking action occurs, as shown in fig. 4. For example, no brake action should be induced; at this time, if the target is identified as a following target, the AEB target selection threshold is used as a target classifier threshold for the perceived detected target in the AEB interest area, and the AEB target selection threshold should be higher than a standard classifier threshold, so that the probability of braking caused by false detection of the uncertain target is reduced.

When the anti-collision active braking function and the front vehicle tracking and driving function are started simultaneously, the multi-sensor fusion system is required to more accurately identify the target vehicle so as to reduce the probability of false braking and avoid rear-end collision of the rear vehicle and reduction of driving experience. According to the application scene and the function, different key control target selection methods can be adopted for the two different control applications in the same system.

What is not described in detail in this specification is prior art known to those skilled in the art.

Claims

1. A key control target selection method of a vehicle-mounted advanced auxiliary driving system is characterized in that,

training a standard classifier for judging whether a target object exists or not in advance based on a multi-element information fusion module, and setting a standard threshold value of the standard classifier;

the standard threshold value of the standard classifier is subjected to preset lowering treatment to obtain at least one first classifier, the standard threshold value of the standard classifier is subjected to preset raising treatment to obtain at least one second classifier, the threshold value of the first classifier is lower than the standard threshold value, and the threshold value of the second classifier is higher than the standard threshold value;

when a front vehicle tracking running function is started, a first classifier is utilized to select a key control target, so that the omission ratio of the following key control target is reduced;

when the anti-collision active braking function of the vehicle is started, the second classifier is utilized to select key control targets, so that the error detection rate of the uncertain targets is reduced;

the training of the standard classifier for judging whether the target object exists based on the multivariate information fusion module comprises the following steps:

extracting attribute parameters of each target object based on the multivariate information fusion module, combining the attribute parameters in parallel to form a training vector, and carrying out normalization processing on the training vector;

2. The target recognition method of the in-vehicle advanced auxiliary driving system according to claim 1, wherein the attribute parameters include:

3. The method for identifying an object of an in-vehicle advanced driver assistance system according to claim 1, wherein the machine learning includes any one of a neural network, a support vector machine, and a mean value cluster.