CN104143090B - A kind of automobile door opening method based on recognition of face - Google Patents

Abstract

The invention relates to the field of automobiles, in particular to a method for opening doors of automobiles based on face recognition. The present invention provides a method for opening the door of a car based on face recognition, comprising the following steps: 1. Using a mobile phone camera to collect images or video streams containing faces, and then passing the face information to a face processing and recognizer, the face processing 2. Use the sparse feature spherical center classifier to classify the face; 3. Match and recognize the face image according to the classification result. The beneficial effect of the present invention is that a sparse feature center classifier is used to discriminate and authenticate face information, and the car owner only needs to put his face close to the mobile phone camera and pass the authentication to open the door of the car, while other people cannot open the door of the car.

Description

A car door opening method based on face recognition

technical field

The invention relates to the field of automobiles, in particular to a method for opening doors of automobiles based on face recognition.

Background technique

The car door opening system is a subsystem in the current vehicle system, and its main function is to prevent theft and facilitate the owner.

At present, there are two main ways to open the door of a car: one is a car key, and the other is a smart key. These two ways of opening the door often cause the driver a headache because of forgetting to bring the key or losing the key.

The classifier occupies a very important position in the pattern recognition system of the vehicle system. Nearest Neighbor Classifier (NN) and Nearest Center Classifier (NM) are two well-known classifiers. As an extension of the nearest neighbor classifier, the nearest feature plane classifier (NFP) is proposed. Because samples belonging to a specific target class can be represented by the linear subspace of this class, based on the above idea, Linear Regression Classifier (LRC) is proposed. A linear regression classifier can be seen as an extension of the nearest center classifier. When the linear regression classifier was proposed, some improved methods were proposed, including based on kernel-LRC, LDA-LRC, PCA-LRC and so on.

Unlike linear regression classifiers, sparse representation-based classifiers (SRC) use models of all classes to classify test samples. After the SRC classifier was proposed, some other improved methods were proposed, such as Two-Stage Test Sample Sparse Representation (TPTSSR), Collaborative Representation Based Classifier (CRC), Regularized Robust Coding Classifier (RRC) and Relaxed Collaborative Representation Classifier (RCR).

In fact, a linear regression classifier can be viewed as an L2 coefficient representation on the class space. The coefficient indicates that the classifier solves the L1-norm minimization problem for the model of all classes, and then classifies the test samples according to the distance between the test samples and the prediction vectors of each class subspace. But the distance between the test sample and the predicted vector may not be a good measure. So based on the coefficient representation classifier and the nearest eigensurface classifier, we propose the Sparse Feature Spherical Center Classifier (SFSC) classifier.

Contents of the invention

Aiming at the defects or deficiencies in the prior art, the technical problem to be solved by the present invention is to provide a method for opening the door of a car based on face recognition, which can realize the car door unlocking without a key (or other authentication items). The opening of the door makes the driver completely free from the trouble of carrying a lot of items, and also improves the problem of inaccurate expansion of the nearest characteristic line and computational complexity.

The technical solution adopted by the present invention is to provide a method for opening the door of a car based on face recognition, comprising the following steps

Step 1: Use the mobile phone camera to collect images or video streams containing faces, and then pass the face information to the face processing and recognition device, which preprocesses the faces;

Step 2: Classify faces using a sparse feature sphere center classifier;

Step 21: Assuming that each class has at least three samples, a new distance measure is proposed, called the feature center measure, the feature center measure refers to the test sample and the feature center The Euclidean distance between , which can be calculated as

In the formula is a tetrahedron The center of the inscribed ball of , the characteristic center can be calculated as

In the formula with Respectively refer to the triangle with area;

Step 22: Calculate the coefficient weight of each class. The coefficient indicates that the classifier constructs a model of all classes. X is a q-dimensional vector that accumulates all classes, which can be expressed as

Then we normalize the model X of all classes to generate a unit vector; assuming the test vector is x, we can solve the L1-norm minimization problem as:

g＝argmin _g ||g|| ₁ subject to Xg＝x (4)

After we obtain the sparse coefficient vector using formulas ( ₃ ) and (4), we will calculate the sparse coefficient sum Sc for each class as

Because S _c is less than 1, and it may be negative, we change S _c to a positive number, so that as the sparse weight of the c-th class, it can be calculated as

w _c =1-s _c (6)

Using the sparse weights for each class, the sparse feature sphere classifier will compute the test sample x to three training samples with The sparsely weighted feature centroid metric of , which can be computed as

When we get all the sparse weighted feature centroid measures, we sort these distances in ascending order, and each distance corresponds to a class label, and finally the sparse feature centroid classifier will classify the test samples into the one with the smallest distance class, it can be expressed as

Step 3: Match and recognize face images according to classification results.

The beneficial effect of the present invention is that the door can be opened by using the mobile phone carried by the car owner, so that the driver can completely get rid of the annoyance of carrying many items with him, because the face information of each person is different, so the car owner only needs to put the face Close to the mobile phone camera and pass the authentication to open the door of the car, while other people cannot open the door of the car, which increases the safety of the car.

Description of drawings

Fig. 1 is a schematic diagram of the characteristic sphere center measurement of the method of opening the car door based on face recognition in the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

The face recognition-based car door opening system of the present invention collects face images or videos through the camera of the mobile phone of the car owner. After the face image or video information is collected, the new car door opening system will use the proposed sparse feature center classifier to authenticate the face information to determine whether the owner is opening the door.

The driver uses a mobile phone camera to collect face images or videos, and then transmits the collected face images or videos to the face processing and recognition device. When the face processing and recognition device receives the face image or video from the collector chip, it will perform some preprocessing on the face image or video signal, and then use the sparse feature center classifier we proposed to classify the face image or video signal. It performs classification certification, and when the system in the car is certified, the door will be opened.

As shown in Figure 1, the present invention proposes a face recognition-based car door opening method, which includes two steps: in the first step, the sparse feature center classifier assumes that each class has at least 3 samples, which is the closest The eigenfaces are similar. Sparse feature centroid classifier Instead of using feature face metric, a new distance metric is proposed, called feature centroid metric. The characteristic centroid measure refers to the test sample and the characteristic centroid The Euclidean distance between , which can be calculated as

in is a tetrahedron The center of the inscribed ball. The center of the characteristic ball can be calculated as

in with Respectively refer to the triangle with area;

After the first step, we start the second step to calculate the coefficient weights of each class. The coefficient indicates that the classifier constructs a model X of all classes as a q-dimensional vector that accumulates all classes, which can be expressed as

We then normalize the model X across all classes, resulting in a unit vector. Assuming the test vector is x, we can solve the L1-norm minimization problem as:

g＝argmin _g ||g|| ₁ subject to Xg＝x (4)

For Equation (4), we know that the purpose of a sparse representation classifier is to produce an ideal parameter vector So the class with the largest coefficient sum represents the most similar class. Based on the above situation, we will use the sparse coefficient to weight the feature centroid measure. After we obtain the sparse coefficient vectors using formulas ( ₃ ) and (4), we will calculate the sparse coefficients and sc for each class as

Because S _c is less than 1, and it may be negative, we change S _c to a positive number, so that as the sparse weight of the c-th class, it can be calculated as

w _c =1-s _c (6)

Using the sparse weights for each class, the sparse feature sphere classifier will compute the test sample x to three training samples with The sparsely weighted feature centroid metric of , which can be computed as

When we get all the sparse weighted feature centroid measures, we sort these distances in ascending order, and each distance corresponds to a class label, and finally the sparse feature centroid classifier will classify the test samples into the one with the smallest distance class, it can be expressed as

The third step is to match and recognize the face images according to the classification results.

The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be assumed that the specific implementation of the present invention is limited to these descriptions. For those of ordinary skill in the technical field of the present invention, without departing from the concept of the present invention, some simple deduction or replacement can be made, which should be regarded as belonging to the protection scope of the present invention.

Claims (1)

1. A method for opening the door of a car based on face recognition, characterized in that: comprising the following steps Step 1: Use the mobile phone camera to collect images or video streams containing faces, and then pass the face information to the face processing and recognition device, which preprocesses the faces; Step 2: Classify faces using a sparse feature sphere center classifier; Step 21: Assuming that each class has at least three samples, a distance metric is proposed, called the feature sphere center metric, and the feature sphere center metric refers to the test sample and the feature sphere center The Euclidean distance between , is calculated as <mrow><msub><mi>d</mi><mrow><mi>F</mi><mi>S</mi><mi>C</mi></mrow></msub><mrow><mo>(</mo><mi>x</mi><mo>,</mo><mover><mrow><msubsup><mi>x</mi><mi>i</mi><mi>c</mi></msubsup><msubsup><mi>x</mi><mi>j</mi><mi>c</mi></msubsup><msubsup><mi>x</mi><mi>k</mi><mi>c</mi></msubsup></mrow><mo>&amp;OverBar;</mo></mover><mo>)</mo></mrow><mo>=</mo><mo>|</mo><mo>|</mo><mi>x</mi><mo>-</mo><msubsup><mi>x</mi><mi>o</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi><mo>,</mo><mi>c</mi></mrow></msubsup><mo>|</mo><mo>|</mo><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>1</mn><mo>)</mo></mrow></mrow> In the formula is a tetrahedron The center of the inscribed ball of , the characteristic center can be calculated as <mrow><msubsup><mi>x</mi><mi>o</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi><mo>,</mo><mi>c</mi></mrow></msubsup><mo>=</mo><mfrac><mrow><msubsup><mi>s</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>&amp;times;</mo><mi>x</mi><mo>+</mo><msubsup><mi>s</mi><mrow><mi>y</mi><mi>j</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>&amp;times;</mo><msubsup><mi>x</mi><mi>i</mi><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>s</mi><mrow><mi>y</mi><mi>i</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>&amp;times;</mo><msubsup><mi>x</mi><mi>j</mi><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>s</mi><mrow><mi>y</mi><mi>i</mi><mi>j</mi></mrow><mi>c</mi></msubsup><mo>&amp;times;</mo><msubsup><mi>x</mi><mi>k</mi><mi>c</mi></msubsup></mrow><mrow><msubsup><mi>s</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>s</mrow>mi><mrow><mi>y</mi><mi>j</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>s</mi><mrow><mi>y</mi><mi>i</mi><mi>k</mi></mrow><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>s</mi><mrow><mi>y</mi><mi>i</mi><mi>j</mi></mrow><mi>c</mi></msubsup></mrow></mfrac><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>2</mn><mo>)</mo></mrow></mrow> In the formula with Respectively refer to the triangle with the area of with Represent three training samples respectively; Step 22: Calculate the coefficient weight of each class. The coefficient indicates that the classifier constructs a model X of all classes. X is a q-dimensional vector that accumulates all classes. The coefficient weight of each class is expressed as <mrow><mi>X</mi><mo>=</mo><mo>&amp;lsqb;</mo><mo>...</mo><mo>,</mo><msubsup><mi>x</mi><mi>i</mi><mi>c</mi></msubsup><mo>,</mo><mo>...</mo><mo>&amp;rsqb;</mo><mo>,</mo><mi>c</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>M</mi><mo>,</mo><mi>i</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>N</mi><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>3</mn><mo>)</mo></mrow></mrow> Then we normalize the model X of all classes to generate a unit vector; assuming the test vector is x, the solution to the L1-norm minimization problem is: g＝arg min _g ||g|| ₁ subject to Xg＝x (4) After obtaining the sparse coefficient vectors using formulas ( ₃ ) and (4), calculate the sparse coefficient sum Sc of each class as <mrow><msub><mi>s</mi><mi>c</mi></msub><mo>=</mo><msubsup><mi>g</mi><mn>1</mn><mi>c</mi></msubsup><mo>+</mo><msubsup><mi>g</mi><mn>2</mn><mi>c</mi></msubsup><mo>+</mo><mo>...</mo><mo>+</mo><msubsup><mi>g</mi><mi>N</mi><mi>c</mi></msubsup><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>5</mn><mo>)</mo></mrow></mrow> S _c is less than 1, and it may be a negative number, change S _c to a positive number, so as to serve as the sparse weight of the c-th class, S _c as the sparse weight of the c-th class is calculated as w _c =1-s _c (6) Using the sparse weights for each class, the sparse feature sphere classifier will compute the test sample x to three training samples with The sparsely weighted feature sphere centroid measure of the test sample x to three training samples with The sparsely weighted characteristic centroid metric of is computed as <mrow><msub><mi>d</mi><mrow><mi>S</mi><mi>F</mi><mi>S</mi><mi>C</mi></mrow></msub><mrow><mo>(</mo><mi>x</mi><mo>,</mo><mover><mrow><msubsup><mi>x</mi><mi>i</mi><mi>c</mi></msubsup><msubsup><mi>x</mi><mi>j</mi><mi>c</mi></msubsup>msubsup><msubsup><mi>x</mi><mi>k</mi><mi>c</mi></msubsup></mrow><mo>&amp;OverBar;</mo></mover><mo>)</mo></mrow><mo>=</mo><mo>|</mo><mo>|</mo><mi>x</mi><mo>-</mo><msubsup><mi>x</mi><mi>o</mi><mrow><mi>i</mi><mi>j</mi><mi>k</mi><mo>,</mo><mi>c</mi></mrow></msubsup><mo>|</mo><mo>|</mo><mo>&amp;times;</mo><msub><mi>w</mi><mi>c</mi></msub><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>7</mn><mo>)</mo></mrow></mrow> After obtaining all the sparse weighted feature centroid measures, these distances are sorted in ascending order, and each distance corresponds to a category label. Finally, the sparse feature centroid classifier will classify the test samples into the class with the smallest distance, which is Expressed as <mrow><munder><mrow><mi>m</mi><mi>i</mi><mi>n</mi></mrow><mrow><mi>c</mi><mo>*</mo></mrow></munder><msub><mi>d</mi><mrow><mi>S</mi><mi>F</mi><mi>S</mi><mi>C</mi></mrow></msub><mrow><mo>(</mo><mi>x</mi><mo>,</mo><mover><mrow><msubsup><mi>x</mi><mi>i</mi><mi>c</mi></msubsup><msubsup><mi>x</mi><mi>j</mi><mi>c</mi></msubsup><msubsup><mi>x</mi><mi>k</mi><mi>c</mi></msubsup></mrow><mo>&amp;OverBar;</mo></mover><mo>)</mo></mrow><mo>,</mo><mi>c</mi><mo>=</mo><mn>1</mn><mo>,</mo><mn>2</mn><mo>,</mo><mo>...</mo><mo>,</mo><mi>M</mi><mo>-</mo><mo>-</mo><mo>-</mo><mrow><mo>(</mo><mn>8</mn><mo>)</mo></mrow></mrow> Step 3: Match and recognize face images according to classification results.