Head positioning method based on Adaboost

Abstract

The invention provides an Adaboost-based head positioning method. The method comprises the following steps of: 1, performing real-time video acquisition on head information by using a camera; 2, receiving real-time data transmitted when the head information is acquired by the video in the step 1 through an image processing module, and preprocessing an image; and step 3, head identification is carried out by using an Adaboost-based method, so that the method has the advantage of realizing positioning of the head of the human body.

Description

Technical Field

[0001] The present invention belongs to the technical field of positioning methods, and particularly relates to a head positioning method based on Adaboost. Background Art

[0002] In the field of human body three-dimensional modeling and intelligent analysis, the feature recognition and extraction of special parts are the core basis for realizing high-order applications. Among them, the three-dimensional point cloud positioning, feature recognition and extraction technology of the human head, as a key link in face depth analysis and processing, has become a frontier topic explored by academia and industry. This technology not only concerns the breakthrough of basic research in computer vision, but also plays an irreplaceable role in multiple scenarios such as face advanced analysis, automotive safety driving monitoring, immersive human-computer interaction systems, and medical auxiliary diagnosis.

[0003] The positioning, feature recognition and feature extraction of the head are widely used in many fields such as face advanced analysis, automotive safety driving, immersive human-computer interaction, and medical assistance. The positioning methods of the head are mainly divided into two categories, one is the positioning method based on two-dimensional images, and the other is the positioning method based on three-dimensional data. Most of the existing methods are implemented by combining two-dimensional images with machine learning / deep learning methods to achieve the positioning, feature recognition and feature extraction of the head. Most of the three-dimensional data-based head positioning methods are based on RGBD data. Such methods can effectively solve many problems based on two-dimensional images, such as solving the problem that the two-dimensional image-based head positioning method can only reflect the basic orientation, and solving the problem of accurate estimation of feature points in the case of large head rotation. However, they still cannot meet the needs of practical applications. Summary of the Invention

[0004] The present invention proposes a head positioning method based on Adaboost, which solves the problems in the prior art.

[0005] The technical solution of the present invention is realized as follows: A head positioning method based on Adaboost includes:

[0006] Step 1: Use a camera to collect real-time video of the target head;

[0007] Step 2: The image processing module receives the real-time video data collected in step (1) and preprocesses the video frames;

[0008] Step 3: Perform head detection and positioning on the preprocessed video frames based on the AdaBoost algorithm.

[0009] As a preferred implementation, the camera in step (1) includes an autofocus color camera and a depth camera, which respectively obtain the RGB image and depth image of the target head in real time.

[0010] As a preferred embodiment, both the color camera and the depth camera are devices with a wide angle of 90°, 5 million pixels, and a frame rate of 15 fps, and the real-time video data collected is transmitted to the image processing module through video conversion and transmission technology.

[0011] As a preferred embodiment, the AdaBoost-based head detection in step (3) specifically includes:

[0012] 3.1 Initialize the sample weights. For non-face sample i, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2m; for face samples, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2.

[0013] Among them, face sample i and non-face sample i are initialized to different values, m is the total number of non-face samples, i is the total number of face samples, and the following steps are performed for the loop of t = 1, 2,..., T, where T is the number of loops.

[0014] As a preferred embodiment, the loop performs the following steps:

[0015] 3.2 Normalize the current weight Dt;

[0016] 3.3 Train weak classifiers based on all candidate features;

[0017] 3.4 Select the optimal weak classifier according to the minimum weighted error rate criterion;

[0018] 3.5 Update the sample weights according to the classification results of the selected weak classifier;

[0019] 3.6 Construct a strong classifier according to the weak classifiers in each round;

[0020] 3.7 Combine the weak classifiers in a cascaded manner to form an AdaBoost strong classifier.

[0021] As a preferred embodiment, the formula used for the weight normalization process is:

[0022]

[0023] Among them, q i is the normalized weight value.

[0024] As a preferred embodiment, for each feature f, a weak classifier h(x, f, p, θ) is trained, and the weighted error rate ε f of all feature weak non-classifiers is calculated by the formula:

[0025]

[0026] Among them, p and θ are respectively the threshold and bias parameter of the weak classifier.

[0027] As a preferred embodiment, in step 3, the optimal weak classifier h t (x)

[0028]

[0029] As a preferred embodiment, in step 4, the weights are adjusted and updated according to the optimal weak classifier

[0030]

[0031] Among them

[0032] As a preferred embodiment, the formula for selecting the strong classifier in step 5 is:

[0033]

[0034] After adopting the above technical solution, the beneficial effects of the present invention are:

[0035] Useful information such as RGB image information and depth information of the head are obtained in real time through the autofocus color camera and depth camera used in video acquisition, and the video data will be transmitted to the image processing module through video conversion and transmission technology. Then, the image processing module receives the real-time video data transmitted from the video acquisition step through video frame acquisition, preprocesses the image, and finally uses the Adaboost-based method for head recognition to realize the positioning process of the human head, making the estimation of head feature points more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] like Figure 1 As shown, a head positioning method based on Adaboost includes:

[0040] Step 1: Use a camera to collect real-time video of the target head;

[0041] Step 2: The image processing module receives the real-time video data collected in step (1) and pre-processes the video frame;

[0042] Step 3: Perform head detection and positioning on the preprocessed video frames based on the AdaBoost algorithm.

[0043] The camera described in step (1) includes an auto-focus color camera and a depth camera, which respectively acquire an RGB image and a depth image of the target head in real time.

[0044] The autofocus color camera and depth camera used in video acquisition are used to obtain useful information such as RGB image information and depth information of the head in real time, and the video data will be transmitted to the image processing module through video conversion transmission technology. Then, the image processing module receives the real-time video data transmitted from the video acquisition step through the video frame acquisition step, and pre-processes the image. Finally, the Adaboost-based method is used for head recognition to realize the positioning process of the human head, making the estimation of head feature points more accurate.

[0045] The color camera and depth camera are both devices with a wide angle of 90°, 5 million pixels, and a frame rate of 15fps, and the collected real-time video data is transmitted to the image processing module through video conversion transmission technology.

[0046] In the present invention, it is realized through a head positioning system, which consists of a software system and a hardware system. The hardware system is composed of a color camera, a depth camera, a digital signal processor, a synchronous dynamic random access memory, and a flash memory. The software system is composed of a video acquisition module and a data processing module. In the video acquisition step, useful information such as RGB image information and depth information of the head is obtained in real time using an autofocus color camera and a depth camera. In this module, a camera with a wide angle of 90°, 5 million pixels, and a frame rate of 15 frames per second will be selected to collect real-time video data. The real-time video data will be transmitted to the image processing module through video conversion and transmission technology. In the video frame acquisition step, the image processing module will receive the real-time video data transmitted from the video acquisition step and preprocess the image.

[0047] The head detection based on AdaBoost in step (3) specifically includes:

[0048] 3.1 Initialize the sample weights. For a non-face sample i, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2m; for a face sample, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2;

[0049] Among them, the face sample i and the non-face sample i are initialized to different values, m is the total number of non-face samples, and i is the total number of face samples.

[0050] For t = 1, 2,..., T loops, the following steps are performed, where T is the number of loops, and the loop performs the following steps:

[0051] 3.2 Normalize the current weight Dt;

[0052] 3.3 Train weak classifiers based on all candidate features;

[0053] 3.4 Select the optimal weak classifier according to the minimum weighted error rate criterion;

[0054] 3.5 Update the sample weights according to the classification results of the selected weak classifier;

[0055] 3.6 Construct a strong classifier according to the weak classifiers in each round;

[0056] 3.7 Combine the weak classifiers in a cascaded manner to form an AdaBoost strong classifier.

[0057] The formula used for the weight normalization process is:

[0058]

[0059] Among them, q i is the normalized weight value.

[0060] For each feature f, a weak classifier h(x, f, p, θ) is trained, and the weighted error rate ε of all feature weak non-classifiers is calculated f The calculation formula is:

[0061]

[0062] where p and θ are the threshold and bias parameter of the weak classifier respectively.

[0063] In step 3, the optimal weak classifier h t (x)

[0064]

[0065] In step 4, the weights are adjusted and updated according to the optimal weak classifier

[0066]

[0067] where

[0068] The formula for selecting the strong classifier in step 5 is:

[0069]

[0070] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. In the description of the present invention, unless otherwise specified and defined, it should be noted that the terms "installed", "connected", "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the communication inside two elements. It can be directly connected or indirectly connected through an intermediate medium. For those of ordinary skill in the art, the specific meanings of the above terms can be understood according to specific situations.

[0071] The above are only the preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A head positioning method based on Adaboost, characterized in that, Including: Step 1: Use a camera to collect real-time video of the target head; Step 2: The image processing module receives the real-time video data collected in step (1) and preprocesses the video frames; Step 3: Perform head detection and localization on the preprocessed video frames based on the AdaBoost algorithm.

2. The head positioning method based on Adaboost according to claim 1, characterized in that, In step (1), the camera includes an auto-focus color camera and a depth camera, which respectively obtain the RGB image and depth image of the target head in real time.

3. The head positioning method based on Adaboost according to claim 2, wherein Both the color camera and the depth camera are devices with a wide angle of 90°, 5 million pixels, and a frame rate of 15 fps, and the collected real-time video data is transmitted to the image processing module through video conversion and transmission technology.

4. A head positioning method based on Adaboost according to claim 1, characterized in that The head detection based on AdaBoost in step (3) specifically includes: 3.1 Initialize the sample weights. For non-face sample i, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2m; for face samples, the error weight Dt(i) of the i-th sample in the t-th loop is 1 / 2; Among them, face sample i and non-face sample i are initialized to different values, m is the total number of non-face samples, and i is the total number of face samples.

5. The head positioning method based on Adaboost according to claim 4, characterized in that, For t = 1, 2,..., T loops, perform the following steps, where T is the number of loops, and the loop performs the following steps: 3.2 Normalize the current weight Dt; 3.3 Train weak classifiers based on all candidate features; 3.4 Select the optimal weak classifier according to the minimum weighted error rate criterion; 3.5 Update the sample weights according to the classification results of the selected weak classifier; 3.6 Construct a strong classifier according to the weak classifiers in each round; 3.7 Combine the weak classifiers in a cascaded manner to form an AdaBoost strong classifier.

6. The head positioning method based on Adaboost according to claim 5, wherein, The formula used for the weight normalization process is: where q i is a normalized weight value.

7. The head positioning method based on Adaboost according to claim 5, characterized in that For each feature f, a weak classifier h(x, f, p, θ) is trained, and the weighted error rate ε of all weak non-classifiers of features is calculated f The calculation formula of Among them, p and θ are the threshold and bias parameters of the weak classifier respectively.

8. A head positioning method based on Adaboost according to claim 5, characterized in that, In step 3, the optimal weak classifier h is selected according to the minimum error rate t (x) 9. The head positioning method based on Adaboost according to claim 5, wherein In step 4, update the weights according to the optimal weak classifier Among them 10. A head positioning method based on Adaboost according to claim 1, characterized in that, The formula for selecting the strong classifier in step 5 is:

Images