US-12626500-B2 - Device and method for filtering out a misrecognized object

Abstract

A device for filtering out an object includes a camera sensor that captures an image of the surroundings of a vehicle. The device includes a controller that recognizes the object and a road area in the image, determines a target area corresponding to the object, determines whether or not the object is a misrecognized object based on the target area, and deletes the misrecognized object from the image.

Inventors

• Seo Won Lee
• Young Hyun Kim

Assignees

• HYUNDAI MOTOR COMPANY
• KIA CORPORATION

Dates

Publication Date

20260512

Application Date

20230524

Priority Date

20230119

Claims (16)

1. 1 . A device for filtering out an object, the device comprising: a camera sensor configured to capture an image of surroundings of a vehicle; and a controller configured to recognize the object and a road area in the image, determine a target area corresponding to the object, determine whether or not the object is a misrecognized object based on the target area, and delete the misrecognized object from the image, wherein the controller is configured to determine the target area corresponding to the object based on a separation distance from the object.
2. 2 . The device of claim 1 , wherein the controller is configured to determine an area located downward of the object as the target area.
3. 3 . The device of claim 1 , wherein the controller is configured to determine a margin and a scale of the target area corresponding to the object based on the separation distance from the object.
4. 4 . The device of claim 1 , wherein the controller is configured to determine whether the object is misrecognized by performing two-dimensional (2D) Gaussian filtering on the target area.
5. 5 . The device of claim 1 , wherein the controller is configured to determine whether the object is the misrecognized object by performing 2D Gaussian filtering on the target area of the object when a reliability value of the object is less than or equal to a preset value.
6. 6 . The device of claim 1 , wherein the controller is configured to: calculate a sum of results obtained by sequentially applying a 2D Gaussian mask within the target area; reduce a score of the object corresponding to the target area in proportion to an average of the sum when the average of the sum is less than a first threshold; and determine that the object is the misrecognized object when the score of the object is less than a second threshold.
7. 7 . The device of claim 6 , wherein the controller is configured to change the first threshold according to the separation distance from the object.
8. 8 . The device of claim 6 , wherein the controller is configured to reduce the first threshold as the separation distance from the object decreases.
9. 9 . A method for filtering out an object, the method comprising: capturing, by a camera sensor, an image of surroundings of a vehicle; recognizing, by a controller, the object and a road area in the image; determining, by the controller, a target area corresponding to the object; determining, by the controller, whether or not the object is a misrecognized object based on the target area; and deleting, by the controller, the misrecognized object from the image, wherein the determining of the target area corresponding to the object includes determining, by the controller, the target area corresponding to the object based on a separation distance from the object.
10. 10 . The method of claim 9 , wherein determining the target area corresponding to the object includes determining, by the controller, an area located downward of the object as the target area.
11. 11 . The method of claim 9 , wherein determining the target area corresponding to the object includes determining, by the controller, a margin and a scale of the target area corresponding to the object based on the separation distance from the object.
12. 12 . The method of claim 9 , wherein determining whether or not the object is the misrecognized object includes determining, by the controller, whether the object is misrecognized by performing two-dimensional (2D) Gaussian filtering on the target area.
13. 13 . The method of claim 9 , wherein determining whether or not the object is the misrecognized object includes determining, by the controller, whether the object is the misrecognized object by performing 2D Gaussian filtering on the target area of the object when a reliability value of the object is less than or equal to a preset value.
14. 14 . The method of claim 9 , wherein determining whether or not the object is the misrecognized object includes: calculating, by the controller, a sum of results obtained by sequentially applying a 2D Gaussian mask within the target area; reducing, by the controller, a score of the object corresponding to the target area in proportion to an average of the sum when the average of the sum is less than a first threshold; and determining, by the controller, that the object is the misrecognized object when the score of the object is less than a second threshold.
15. 15 . The method of claim 14 , wherein reducing the score of the object corresponding to the target area in proportion to the average of the sum includes changing, by the controller, the first threshold according to the separation distance from the object.
16. 16 . The method of claim 14 , wherein reducing the score of the object corresponding to the target area in proportion to the average of the sum includes reducing, by the controller, the first threshold as the separation distance from the object decreases.

Description

CROSS-REFERENCE TO RELATED APPLICATION This application claims the benefit of priority to Korean Patent Application No. 10-2023-0008216, filed in the Korean Intellectual Property Office on Jan. 19, 2023, the entire contents of which are incorporated herein by reference. TECHNICAL FIELD The present disclosure relates to a technology for filtering out an object misrecognized by an image recognition system (or image recognition algorithm) mounted on an autonomous vehicle. BACKGROUND In general, an artificial neural network (ANN) is a field of artificial intelligence and is an algorithm for allowing a machine to be trained by simulating a human neural structure. Recently, ANN technology has been applied to image recognition, voice recognition, natural language processing, and the like, and has shown excellent effects. The artificial neural network consists of an input layer that receives an input, a hidden layer that learns, and an output layer that returns the results of operations. A deep neural network (DNN) with multiple hidden layers is also a kind of artificial neural network. The artificial neural network allows a computer to learn from data. When trying to solve a problem using the artificial neural network, what is to be prepared is a suitable artificial neural network model and data to be analyzed. The artificial neural network model to solve a problem is trained based on data. Before training the model, the data is first properly processed. The reason for this is that input data and output data required by the artificial neural network model are regularized. Therefore, a process of preprocessing acquired raw data to be suitable for required input data is required. After the preprocessing is completed, the processed data needs to be divided into two types. The data needs to be classified into a training dataset and a validation dataset. The training dataset is used to train the model and the validation dataset is used to verify the performance of the model. There are several reasons for validating the artificial neural network model. Artificial neural network developers perform tuning of the model by modifying hyperparameters of the model based on the verification result of the model. In addition, the artificial neural network developers verify models in order to select which model is suitable among several models. The reasons why model verification is necessary are explained in more detail as follows. The first reason is to predict accuracy. The purpose of artificial neural networks is to achieve good performance on out-of-sample data that is not used for training. Therefore, after creating the model, it is necessary to check how well the model performs on out-of-sample data. However, it is required to not validate the model using the training dataset. Thus, the accuracy of the model needs to be measured using the validation dataset separate from the training dataset. The second reason is to improve the performance of the model by tuning the model. For example, overfitting may be prevented. Overfitting refers to a state where a model is overtrained on the training dataset. For example, when the training accuracy is high, but the validation accuracy is low, it may be suspected that overfitting has occurred. In addition, overfitting may be figured out in more detail through training loss and validation loss. Preventing overfitting increases the validation accuracy. Overfitting may be prevented by using methods such as regularization or dropout. On the other hand, an image recognition system (or image recognition algorithm) mounted in an autonomous vehicle often misrecognizes an object on a road. This may weaken the performance stability of an autonomous driving system as well as reduce the user's reliability of the autonomous vehicle. To solve this problem, various methods have been proposed to improve the performance of image recognition algorithms (or image recognition networks) based on deep learning. However, the possibility of misrecognition of objects on real roads cannot be completely ruled out. In particular, even in the case of a method of further performing a vanishing point-based post-processing process, misrecognition of a structure located on an actual road (e.g., a lava cone, a construction sign, a toll gate structure, or the like) may occur. Objects misrecognized around the median of a road cannot be filtered out. The subject matter described in this background section are prepared to enhance understanding of the background of the present disclosure. The background section may include subject matter other than the prior art already known to those of ordinary skill in the field to which this technology belongs. SUMMARY The present disclosure has been made to solve the above-mentioned problems occurring in the prior art while advantages achieved by the prior art are maintained intact. An aspect of the present disclosure is to provide a device and a method for filtering out a misrecognized