WO2019085060A1

WO2019085060A1 - Method and system for detecting waving of robot, and robot

Info

Publication number: WO2019085060A1
Application number: PCT/CN2017/112211
Authority: WO
Inventors: 张帆
Original assignee: 南京阿凡达机器人科技有限公司
Priority date: 2017-10-30
Filing date: 2017-11-21
Publication date: 2019-05-09
Also published as: CN107886057A; CN107886057B

Abstract

A method and system for detecting waving of a robot, and a robot. The method comprises: detecting, in a video stream, a palm at a standard position by means of a cascade classifier; extracting angular points of the palm at the standard position so as to obtain an angular point set; tracking and detecting, in the video stream, each angular point in the angular point set, so as to obtain a motion trajectory corresponding to each angular point; and determining whether the palm is waving according to the motion trajectory of each angular point in the angular point set. Also comprised are a system and a robot using the method. The method, system and robot realize waving detection in a complex environment, and have a strong anti-interference capability with regards to dynamic background noise and a high detection accuracy.

Description

Wavelet detecting method and system for robot and robot

The present application claims the priority of the Japanese Patent Application Serial No. No. No. No. No. No. No. No. No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No No.

Technical field

The invention relates to the field of image recognition, in particular to a wave detection method and system for a robot and a robot.

Background technique

With the advancement of technology, artificial intelligence is becoming more and more popular, and people want to interact with robots like real people. However, the current interaction with robots is mainly based on mouse, keyboard and touch screen. People can't satisfy this traditional interaction with many limitations. So, we need to design some more intelligent ways of interacting. Vision is an important channel for understanding interaction information between people. By visually observing the body movements of interacting objects, you can understand each other's interaction intentions. Therefore, in order to increase the anthropomorphic characteristics of the robot-human interaction process, it is necessary for the robot to understand some of the human body language, such as waving a greeting.

The invention patent of the application No. 201610859376.7 discloses a wave detection method based on a motion history image. The invention first detects the approximate area in which the human body is located by the human body detector, and then analyzes the area based on the motion history image to determine whether the person is waving. However, the method based on motion history image analysis is only suitable for static background. When a person is in a complex dynamic background, it is easy to judge the background of the movement as the person is waving. The wave detection algorithm of the prior patent actually performs differential operation on three consecutive images of the detection area, and then uses the threshold method to obtain a binary image of the obtained difference image, and finally uses the acquired series of binary image history information to wave Judge. Since the invention adopts the frame difference method of continuous images, when the background is not fixed but dynamic, for example, when someone in the area to be detected is moving, it is easily interfered to recognize the dynamic background as a person waving.

Therefore, it is necessary to design a method that can accurately recognize waving in a complicated background and improve people. The intelligence of machine interaction.

Summary of the invention

The invention provides a wave detecting method and system for a robot and a robot, which realizes the recognition of the palm of the hand in a complicated background, and tracks the palm of the hand, thereby detecting whether the wave is handed, and the anti-interference against the background noise is strong. . Its technical solutions are as follows:

A method for detecting a wave of a robot includes: detecting, by a cascade classifier, a palm in a standard position in a video stream; extracting a corner point of the palm in a standard position to obtain a corner set; and in the video stream Each corner point in the corner point set is tracked and detected, and the motion track corresponding to each corner point is obtained; whether the palm is waved according to the motion track of each corner point in the corner point set.

Compared with the prior art, the wave detection method can perform palm detection on a separate video frame, and does not need to combine the historical data of the video frames before and after, that is, the wave can be detected in the process of waving. Through the cascaded classifier, the gesture recognition can be completed in complex background, and then the corner points of the palm of the hand are detected and tracked, which realizes the detection of the wave in a complex environment, and has strong anti-interference ability to the dynamic background noise, and the correct rate of the wave detection. high.

Preferably, determining whether the user is waving according to the motion trajectory of each corner point in the corner point set specifically includes: a motion trajectory between two adjacent frames in the video stream according to each corner point in the corner point set and Presetting the pixel distance, and analyzing the number of effective motion corner points in the corner point set; when the number of the effective motion corner points is greater than the first preset number, continuing to perform all the corner points in the corner point set Tracking detection; according to the motion trajectory of the effective motion corner point, analyzing the number of effective wave corner points in the effective motion corner point; when the number of the effective wave corner points is greater than the second preset number, determining that the user is waving.

By detecting the corner trajectory in the video stream, it is first judged whether the corner point has been effectively moved, which improves the correct rate of the wave detection; whether the corner point is effectively moved left to right, can effectively judge whether the palm is waving .

Preferably, according to the motion trajectory of each corner point in the corner point set between two adjacent frames of the video stream and the preset pixel distance, analyzing the number of effective motion corner points in the corner point set specifically includes: Calculating, by the motion trajectory between two adjacent frames in the video stream, the motion distance of the same corner point between two adjacent frames in the video stream; determining whether the motion distance is And greater than the preset pixel distance; when the moving distance is greater than the preset pixel distance, determining that the corner point performs effective motion, and using the corner point as the effective motion corner point; when the motion distance is less than the preset when setting the pixel distance, the corner point determined lost motion, and the motion of the effective corner point N _track number minus 1, N _track finally calculated is the number of active motion of the inner corner angle set point, The initial value of N _track is the total number N of corner points in the corner set;

By comparing the distance between the corner point and the preset pixel distance, the corner points of the effective motion can be effectively screened out, and the number of effective motion corner points N _{track obtained} is also relatively accurate, in order to judge whether the palm is waved or not. .

Preferably, determining, according to the number of the effective motion corner points and the first preset number, the detection of all the corner points in the corner point set comprises: determining whether the number of the effective motion corner points N _track is greater than a first predetermined number; when the number of active motion corners N _track is smaller than said first number, said effective stop motion trajectory is detected corner points, and determines the user or waving; effective when the motion When the number of corner points N _{track is} greater than the first predetermined number, the motion trajectory of the effective motion corner point is continuously detected.

Determining whether the number N _{Track of} the effective motion corner points is greater than the first preset number, can initially determine whether the palm is waving, and if the number of corner points for effective motion is too low, it can be determined that the palm does not move, if the effective motion If the number of corner points N _{track is} greater than the first preset number, the subsequent judgment process is continued.

Preferably, when the number N _{track of} the effective motion corner points is greater than the first preset number, detecting a left moving distance and a right moving motion between the two adjacent frames in the video stream Determining whether the left moving distance and the right moving distance of the effective moving corner between the adjacent two frames in the video stream are greater than a preset distance; when the effective moving corner between two adjacent frames in the video stream When the left movement distance and the right movement distance are both greater than the preset distance, the effective movement corner point is recorded as the effective wave corner point, and the effective wave corner point number N _{move is} increased by 1, and the finally calculated N _{move is} The number of effective wave corner points in the set of corner points, and the initial value of the number of effective wave corner points N _move is 0.

If the number N _{Track of} the effective motion corner points is greater than the first preset number, it is determined whether the effective motion corner point is effectively moved from left to right, and an effective wave corner point for performing effective motion from left to right can be accurately obtained. The number N _{move is} convenient for the next judgment process.

Preferably, determining whether the user is waving or not according to the number of valid wave corner points and the second preset number comprises: determining whether the number of valid wave corner points is greater than a second preset number; and when the number of valid wave corner points is When it is greater than the second preset number, it is judged that the palm is waving; otherwise, it is judged that the palm is not waved.

Since the palm must perform the left-to-right wave process when waving, the trajectory of the palm corner point will also move to the left and right, thereby detecting the number of effective wave corner points through the effective wave angle The judgment of the number of points, when the number of effective wave corner points is greater than the second preset number, can accurately distinguish whether the palm is waving.

Preferably, before the palm of the standard position is detected in the video stream by the cascade classifier, the palm training sample is generated; the LBP texture feature vector of the palm training sample is extracted, and the palm training sample after the extraction process is obtained; After the palm training sample training cascade categorizer.

Before the wave detection, the Adaboost cascade classifier needs to be trained to perform LBP texture feature extraction on the palm training samples needed to train the Adaboost cascade classifier, which can improve the recognition accuracy of the Adaboost cascade classifier.

Preferably, the preparation of the palm training sample comprises: collecting palm samples of different users in different backgrounds, the palm samples including samples in which the palm rotates within a preset angle range with respect to the standard position.

When screening the palm sample, in order to facilitate the extraction of the LBP feature value, the selected palm sample is the palm sample of the palm relative to the standard position within the preset angle range, so as to better enhance the gesture recognition success rate of the cascade classifier near the standard position. . Collecting palm samples from different users in different backgrounds improves the diversity of samples. When the collected palm training samples have a certain degree of shape and illumination difference, the detection results obtained will be very good, and the cascade classification is improved. The success rate of gesture recognition.

Preferably, the process of extracting the LBP texture feature vector of the palm training sample comprises: dividing the detection window into N×N pixel regions, and centering a central pixel point in the pixel region with adjacent 8 adjacent pixels The gray value of the pixel is compared, where N∈ {64, 32, 16, 8}; if the pixel value of the adjacent pixel is greater than the pixel value of the central pixel, the adjacent pixel is marked as 1, otherwise 0, thereby generating an 8-bit binary As the LBP value of the central pixel point, the LBP value of each pixel in the pixel region can be calculated; and the statistical histogram of the pixel region is calculated according to the LBP value of each pixel in the pixel region. And normalizing the statistical histogram; and connecting the statistical histogram of each pixel region of the palm training sample to form an LBP texture feature vector of the palm training sample.

Before training the cascade classifier, the palm sample is tested for LBP features, and the trained cascade classifier is faster in calculation and can accurately locate the position of the palm on the image.

Preferably, the corner points of the palm area are detected, a set of corner points is obtained, and all corner points in the corner point set are tracked and detected in the video stream, and the motion track corresponding to each corner point is obtained as follows: The point detection algorithm extracts the corner points of the palms in the standard position, obtains a set of corner points, and performs optical flow tracking on all the corner points in the corner point set in the video stream, and detects the corner points according to the optical flow tracking algorithm. The position of all the corner points in the set is calculated, and the motion trajectory corresponding to each corner point is calculated according to the sparse optical flow algorithm.

Through the corner detection algorithm, the corner points of the palm area can be effectively extracted, and the optical flow tracking algorithm can better realize the optical flow tracking of each corner point in the video stream, and can be better detected by the sparse optical flow algorithm. The trajectory of the corner points facilitates the subsequent detection process of the wave.

A wave detection system for a robot includes: a detection module, configured to detect a palm in a standard position in a video stream by a cascade classifier; and a corner point extraction module electrically connected to the detection module for extracting the a corner point of the palm of the standard position, obtaining a set of corner points; a corner point tracking module electrically connected to the corner point extraction module, configured to perform tracking detection on each corner point in the corner point set in the video stream, a motion track corresponding to each corner point; a judging module electrically connected to the corner point tracking module, configured to determine whether the palm is waving according to the motion track of each corner point in the corner point set.

Preferably, the determining module includes: an analysis sub-module, and analyzing, according to a motion track between two adjacent frames in the video stream and a preset pixel distance, The number of effective motion corner points in the corner point set; the analysis sub-module is further configured to continue tracking all corner points in the corner point set when the number of the effective motion corner points is greater than the first preset number Detecting; the analysis sub-module is further configured to analyze, according to the motion trajectory of the effective motion corner point, the number of effective wave corner points in the effective motion corner point; the determining sub-module is electrically connected to the analysis sub-module, For determining that the user is waving when the number of valid wave corner points is greater than the second preset number.

Preferably, the determining module further includes: a calculating submodule, configured to calculate, in the video stream, adjacent to each other according to a motion track between two adjacent frames in the video stream in the corner set The judging sub-module is further electrically connected to the computing sub-module for determining whether the moving distance is greater than a preset pixel distance; the judging sub-module is further used for When the moving distance is greater than the preset pixel distance, determining that the corner point performs effective motion and using the corner point as the effective motion corner point; the determining sub-module is further configured to: when the moving distance is smaller than the when the predetermined pixel distance, the corner point determined lost motion, and the motion of the effective corner point N _track number minus 1, N _track finally calculated is the number of active motion of the inner corner angle set point The initial value of N _track is the total number N of corner points in the corner set.

Preferably, the determination sub-module is further for determining whether said effective amount of N _track motion corners is greater than a first predetermined number; the determining sub-module is further configured to, when moving the active corner point less than the number N _track said first predetermined number, said effective stop motion trajectory is detected corner points, and determines the user does not wave; the determining sub-module is further configured to, when moving the active corner point is greater than the first number N _track When a predetermined number is reached, the motion trajectory of the effective motion corner point is continuously detected.

Preferably, the determining module further includes a detecting submodule, configured to detect, between the adjacent two frames in the video stream, when the number N _{track of} the effective motion corner points is greater than the first preset number The effective motion corner point moves to the left and the distance to the right; the determining sub-module is further configured to determine whether the left moving distance and the right moving distance of the effective moving corner between the adjacent two frames in the video stream are greater than The determining sub-module is further configured to: when the effective moving corner point leftward moving distance and the rightward moving distance are greater than a preset distance between adjacent two frames in the video stream, the effective moving angle The point is recorded as the effective wave corner point, and the number of effective wave corner points N _{move is} increased by 1. The final calculated N _move is the number of effective wave corner points in the corner point set, and the number of valid wave corner points N _move The initial value is 0.

Preferably, the determining sub-module is further configured to determine whether the number of valid wave corner points is greater than a second preset number; the determining sub-module is further configured to: when the number of valid wave corner points is greater than the second preset When the quantity is counted, it is judged that the palm is waving; otherwise, it is judged that the palm does not wave.

Preferably, the method further includes: a sample making module, configured to generate a palm training sample; a feature extraction module, extracting an LBP texture feature vector of the palm training sample, and obtaining an extracted palm training sample; a classifier training module, and the The feature extraction module is electrically connected for training the cascade classifier by extracting the processed palm training samples.

Preferably, the feature extraction module includes: a processing sub-module, configured to divide the detection window into N×N pixel regions, and one central pixel point in the pixel region and adjacent eight adjacent pixel points The gray value is compared, wherein N ∈ {64, 32, 16, 8}; the eigenvalue calculation sub-module is electrically connected to the processing sub-module, if the pixel value of the adjacent pixel is greater than the pixel of the central pixel a value, the adjacent pixel is marked as 1, otherwise 0, thereby generating an 8-bit binary number as the LBP value of the central pixel, whereby each pixel in the pixel region can be calculated An LBP value; the processing sub-module is further configured to calculate a statistical histogram of the pixel region according to an LBP value of each pixel in the pixel region, and perform normalization processing on the statistical histogram; The sub-module is further configured to connect a statistical histogram of each pixel region of the palm training sample to form an LBP texture feature vector of the palm training sample.

Preferably, the corner tracking module is further configured to extract a corner point of the palm in a standard position by a corner detection algorithm, obtain a corner point set, and perform all corner points in the corner point set in the video stream. The optical flow tracking method detects the position of all the corner points in the corner point set according to the optical flow tracking algorithm, and calculates the motion track corresponding to each corner point according to the sparse optical flow algorithm.

A robot integrated with a wave detection system of a robot provided by the present invention.

According to the present invention, a wave detection method and system for a robot and a robot can achieve at least one of the following beneficial effects:

1. The present invention is capable of being in a separate video frame relative to prior art detection methods for waving Performing palm detection on the above does not need to combine the historical data of the video frames before and after, that is, the wave can be detected in the process of waving. Through the cascaded classifier, the gesture recognition can be completed in complex background, and then the corner points of the palm of the hand are detected and tracked, which realizes the detection of the wave in a complex environment, and has strong anti-interference ability to the dynamic background noise, and the correct rate of the wave detection. high.

2. Through the corner detection algorithm, the corner points of the palm area can be effectively extracted, and the optical flow tracking algorithm can better realize the optical flow tracking of each corner point in the video stream, which can be better by the sparse optical flow algorithm. The trajectory of the corner points is detected to facilitate the subsequent detection process of the wave.

3. By detecting the corner trajectory in the video stream, it is first determined whether the corner point has been effectively moved, and the correct rate of the wave detection is improved. It is judged whether or not the corner point has an effective movement to the left and right, and it is possible to effectively judge whether the palm is waving.

4. When screening the palm sample, in order to facilitate the extraction of the LBP feature value, the selected palm sample is the palm sample of the palm relative to the standard position within the preset angle range, so as to better enhance the gesture recognition of the cascade classifier near the standard position. Success rate. Collecting palm samples from different users in different backgrounds improves the diversity of samples. When the collected palm training samples have a certain degree of shape and illumination difference, the detection results obtained will be very good, and the cascade classification is improved. The success rate of gesture recognition.

5. Before training the cascade classifier, the LBP feature detection is performed on the palm sample, and the trained cascade classifier is faster in calculation and can accurately locate the position of the palm on the image.

DRAWINGS

The above-described characteristics, technical features, advantages and implementation manners of a robot hand wave detection method, system and robot will be further described below in a clear and understandable manner with reference to the accompanying drawings.

1 is a flow chart of an embodiment of a method for detecting a wave of a robot according to the present invention;

2 is a flow chart showing another embodiment of a method for detecting a wave of a robot according to the present invention;

Figure 3 is a schematic view of the palm of the present invention;

4 is a flow chart showing another embodiment of a method for detecting a wave of a robot according to the present invention;

Figure 5 is a diagram showing the tracking of the corner of the palm of the present invention;

6 is a flow chart of another embodiment of a method for detecting a wave of a robot according to the present invention;

7 is a schematic structural view of a wave detection system of a robot according to the present invention;

8 is another schematic structural view of a wave detection system of a robot according to the present invention;

9 is another schematic structural view of a wave detection system of a robot of the present invention.

Detailed ways

In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the specific embodiments of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings in the following description are only some embodiments of the present invention, and those skilled in the art can obtain other drawings according to the drawings without obtaining creative labor, and obtain Other embodiments.

In order to simplify the drawings, only the parts related to the present invention are schematically shown in the drawings, and they do not represent the actual structure of the product. In addition, in order to make the drawings simple and easy to understand, components having the same structure or function in some of the figures are only schematically illustrated, or only one of them is marked. In the present context, "a" means not only "only one" but also "more than one".

As shown in FIG. 1, the present invention provides an embodiment of a method for detecting a wave of a robot, comprising:

S4 detects the palm in the standard position in the video stream through the cascade classifier;

S5 extracts a corner point of the palm in the standard position to obtain a corner point set;

S6 performs tracking detection on each corner point in the corner point set in the video stream to obtain a motion track corresponding to each corner point;

S7 determines whether the palm is waving according to the motion trajectory of each corner point in the corner point set.

Specifically, in this embodiment, the Adaboost cascade classifier first uses the Adaboost algorithm to detect the palm in the standard position in the video stream, and after detecting the palm in the standard position, the method of corner point acquisition is used to collect the palm of the hand. The corner points are tracked and detected in the video stream to obtain the motion trajectory of each corner point. Through the motion trajectory, it can be judged whether the palm of the video stream is waving.

In this embodiment, the robot can synchronously detect the palm and the palm when collecting the video stream. The motion trajectory of the upper corner point does not need to be waved based on the historical image. The recognition rate of the palm of the hand using the Adaboost cascade classifier is relatively high. Compared with the prior art, the present embodiment can also be accurate under complex background and dynamic background. The palm in the standard position is detected in the video stream, which facilitates the tracking detection of the subsequent opponent's palm corner point.

In this embodiment, by detecting the trajectory of the corner point, whether the palm is waved or not is determined, for example, the motion trajectory of each corner point in the corner point is detected, and if more than half of the corner points are effectively moved left to right, Then you can judge that the user is waving.

As shown in FIG. 2, the present invention further provides another embodiment of a method for detecting a wave of a robot, including:

S1 produces a palm training sample;

S2 extracts an LBP texture feature vector of the palm training sample, and obtains a palm training sample after the extraction process;

S3 trains the cascade classifier by extracting the processed palm training samples;

Preferably, the preparing the palm training sample comprises: S11 collecting palm samples of different users in different backgrounds, wherein the palm samples comprise samples in which the palm rotates within a preset angle range with respect to the standard position.

Preferably, the process of extracting the LBP texture feature vector of the palm training sample specifically includes:

S21 divides the detection window into N×N pixel regions, and compares a central pixel point in the pixel region with gray values of adjacent 8 adjacent pixel points, where N∈{64, 32, 16 ,8};

S22, if the pixel value of the adjacent pixel is greater than the pixel value of the central pixel, the adjacent pixel is marked as 1, otherwise 0, thereby generating an 8-bit binary number as the central image The LBP value of the prime point, whereby the LBP value of each pixel in the pixel region can be calculated;

S23: calculating a statistical histogram of the pixel region according to an LBP value of each pixel in the pixel region, and performing normalization processing on the statistical histogram;

S24 connects the statistical histogram of each pixel region of the palm training sample to form an LBP texture feature vector of the palm training sample.

Specifically, before the cascading classifier is used to detect the palm of the standard position in the video stream, the cascading classifier needs to be trained. This embodiment specifically describes the sample making process of the cascading classifier, and how to train the cascading classifier. process.

As shown in FIG. 3, the user moves back and forth between the 1, 2, and 3 positions during the swinging process. In this embodiment, the Adaboost algorithm is combined with the LBP feature to detect the palm of the video frame. Since the algorithm does not have rotational invariance, it can only recognize the palm of the hand whose rotation angle is within ±15° at a certain standard position. In this embodiment, the gesture of selecting the palm at the 2 position is the standard palm, and the gestures at other positions are not recognized.

The training process of the cascade classifier can be divided into: 1. Making a palm training sample; 2. Extracting the LBP feature of the palm training sample; 3. Training the cascade classifier.

In the process of collecting the palm training samples, in order to facilitate the extraction of the LBP feature values, the selected palm sample is the palm sample of the palm relative to the standard position within the preset angle range, so as to better enhance the gesture of the cascade classifier near the standard position. The recognition success rate, the palm preset angle range may be a range of ±15°, that is, the sample includes a gesture sample with a relative standard position rotation angle within a range of ±15° to better improve the gesture recognition rate of the algorithm near the standard position. In order to improve the diversity of samples when collecting palm samples of different users in different backgrounds, when the collected palm training samples have a certain degree of shape and illumination difference, the detection results obtained will be very good and can be more The success rate of the cascading classifier for gesture recognition is improved.

The extraction process of the LBP feature of the palm training sample is as follows: a. The detection window is divided into 16×16 small cells, and the gray values of the adjacent 8 pixels are compared with one pixel in each cell. If the surrounding pixel value is greater than the center pixel value, the position of the pixel is marked as 1, otherwise 0. Thus, 8 points in the 3*3 neighborhood are compared to produce an 8-bit binary number, that is, an LBP value of the center pixel of the window is obtained; b. a histogram of each cell is calculated, that is, each number (assuming the decimal number LBP value) appears; normalize the histogram; c. Finally, connect the resulting statistical histogram of each cell into a feature vector, that is, the LBP texture of the entire image. Feature vector.

The AdaBoost algorithm used in the training cascade classifier is an iterative algorithm. The core idea is to train different classifiers, ie weak classifiers, for the same training set, and then combine these weak classifiers to construct a stronger one. The final classifier, the algorithm is divided into the following three steps:

a. Initialize the weight distribution of the training data. If there are N samples, each training sample is given the same weight at the beginning: 1/N;

b. Train the weak classifier. If a sample point has been accurately classified, its weight is reduced in constructing the next training set; conversely, if a sample point is not accurately classified, its weight is increased. The weighted updated sample set is then used to train the next classifier, and the entire training process proceeds so iteratively. Using the training dataset with weight distribution to learn, get the basic classifier (choose the threshold with the lowest error rate to design the basic classifier):

G _m (x): x→{-1, +1}

Calculate the classification error rate on the training data set:

It can be seen from the above formula that the error rate on the training data set is the sum of the weights of the misclassified samples.

The calculated coefficient, indicating the degree of importance in the final classifier (purpose: to get the weight of the basic classifier in the final classifier):

As can be seen from the above formula, when

When α _m ≥ 0, and α _m increases as e _m decreases, it means that the smaller the classification error rate, the greater the role of the basic classifier in the final classifier.

Update the weight distribution of the training data set (purpose: get the new weight distribution of the sample) for the next iteration:

D _m+1 =(w _m+1,1 ,w _m+1,2 ,...,w _m+1,N )

The weights that are misclassified by the basic classifier are increased, while the weights of the correctly classified samples are reduced. In this way, in this way, the Adaboost method can "focus" or "focus" on those samples that are more difficult to distinguish.

c. Combine the weak classifiers obtained from each training into a strong classifier. After the training process of each weak classifier is finished, the weight of the weak classifier with small classification error rate is increased, which plays a greater role in determining the final classification function, and reduces the weak classifier with large classification error rate. The weights make it play a smaller role in the final classification function. In other words, a weak classifier with a low error rate has a larger weight in the final classifier, otherwise it is smaller.

Combine each weak classifier to get the final classifier, as follows:

Thanks to the improved palm detection algorithm, it has the following advantages:

1. The gesture detection algorithm performs palm detection on a separate video frame, and does not need to combine historical data of front and rear video frames.

2. The detection algorithm is detected by the LBP feature of the palm image, the calculation speed is relatively fast, and the position of the palm on the image can be accurately located.

3. When the collected palm training samples have a certain degree of shape and difference in illumination, the obtained detection effect will be very good and the recognition rate is high.

As shown in FIG. 4, the present invention further provides another embodiment of a method for detecting a wave of a robot, including:

S11 collects a palm sample of different users in different backgrounds, wherein the palm sample includes a sample in which the palm rotates within a preset angle range with respect to a standard position;

S24, connecting the statistical histogram of each pixel area of the palm training sample to form an LBP texture feature vector of the palm training sample;

S51 extracts a corner point of the palm in the standard position by a corner detection algorithm to obtain a corner point set;

S61 performs optical flow tracking on all the corner points in the corner point set in the video stream, detects the position of all the corner points in the corner point set according to the optical flow tracking algorithm, and calculates each corner according to the sparse optical flow algorithm. Point corresponding motion trajectory;

S71: analyzing, according to the motion trajectory between two adjacent frames in the video stream and the preset pixel distance of each corner point in the corner point set, analyzing the number of effective motion corner points in the corner point set;

S72, when the number of the effective motion corner points is greater than the first preset number, continue to perform tracking detection on all corner points in the corner point set;

S73 analyzes, according to the motion trajectory of the effective motion corner point, the number of effective wave corner points in the effective motion corner point;

S74 determines that the user is waving when the number of valid wave corner points is greater than the second preset number.

This embodiment specifically describes the detection, tracking of the palm corner points, and how to detect the waved motion according to the motion trajectory of the corner points.

Specifically, as shown in FIG. 5, the process of detecting and tracking the movement of the palm. 1 is the palm in the video frame detected in the standard position; 2 is the corner point extracted into the palm area by the corner detection algorithm, and the extracted corner points are multiple, and a part of the figure is schematically shown in the figure. The black point is the detected corner point; 3 is the initial position of the corner detected when the palm is in the standard position in the video frame, and the optical frame is tracked by using the point as the initial position; 4, 5 are The position of the corner point detected by the optical flow tracking algorithm in the next video frame; L1, L2 are the motion paths tracked to the hand corner point by the sparse optical flow algorithm during the palm waving process.

In this embodiment, the corner point extraction method adopts a corner point detection algorithm, which can effectively extract the corner points of the palm area, and in addition, other feature point extraction methods, such as the FAST feature point extraction method, etc., can be used. I will not repeat them here. The tracking of diagonal points in the video stream uses an optical flow tracking algorithm, which can effectively track each corner point in the video stream, which improves the effective degree of tracking, and can better detect the angle by the sparse optical flow algorithm. The trajectory of the point is convenient for the subsequent detection process of the wave.

The motion trajectory of each corner point between two adjacent frames in the video stream is detected, and the motion distance of each corner point can be obtained. When the motion distance of the corner point in the adjacent two frames is greater than the preset pixel distance, the corner point is used as Effectively move the corner points while counting the number of effective motion corner points.

According to the number of the effective motion corner points and the first preset number, the detection situation of all the corner points in the corner point set is analyzed, and the specific detection condition is: one is that the number of effective motion corner points is smaller than the first The preset number, the value of the first preset number is relatively small, and can be set to 5 to 10, because the number of extracted palm corner points is far more than 5 to 10, when the number of effective motion corner points is smaller than the first The preset number, that is, most of the corner points are not effectively exercised, and it can be judged that the palm is not waved.

When the number of effective motion corner points is greater than the first preset number, that is, when the detection condition satisfies the preset requirement, the effective wave angle in the effective motion corner point is obtained according to the motion trajectory of the effective motion corner point. The number of points, when waving, the palm will definitely move to the left and right, the corners of the palm will also do the same movement, which will form a corresponding movement trajectory, according to the movement trajectory of the corner point can judge the palm of the hand. When tracking and detecting the effective motion corner point, the target will be lost inevitably, so the number of effective motion corner points obtained by tracking will not have so many original effective motion corner points, when the effective motion corner point is left to right. After the effective exercise is performed, the effective motion corner point is recorded as an effective wave corner point. When the number of valid wave corner points is greater than the second preset number, it can be determined that the user is waving.

As shown in FIG. 6, the present invention further provides another embodiment of a method for detecting a wave of a robot, including:

S22, if the pixel value of the adjacent pixel is greater than the pixel value of the central pixel, the adjacent pixel is marked as 1, otherwise 0, thereby generating an 8-bit binary number as the LBP of the central pixel a value, whereby an LBP value of each pixel in the pixel region can be calculated;

S711: calculating a motion distance of the same corner point between two adjacent frames in the video stream according to a motion trajectory between two adjacent frames in the video stream according to each corner point in the corner point set;

S712 determines whether the moving distance is greater than a preset pixel distance;

S713: when the moving distance is greater than a preset pixel distance, determining that the corner point performs effective motion, and using the corner point as the effective motion corner point;

S714, when the movement distance is less than the predetermined pixel distance, the corner point determined lost motion, and the motion of the effective corner point N _track number minus 1, the final calculated the angle is the N _track The number of effective motion corner points in the point set, the initial value of N _track is the total number of corner points in the corner point set;

S721 determines whether the number N _{Track of} the effective motion corner points is greater than the first preset number;

S722 stops detecting the motion _{track of} the effective motion corner point when the number N _{track of} the effective motion corner points is smaller than the first preset number, and determines that the user does not wave;

S723: when the number of effective motion corner points N _{track is} greater than the first preset number, continue to detect a motion trajectory of the effective motion corner point;

S731: when the number N _{track of} the effective motion corner points is greater than the first preset number, detecting a left moving distance and a right moving distance between the adjacent two moving frames in the video stream;

S732 determines whether the left moving distance and the right moving distance of the effective moving corner between two adjacent frames in the video stream are greater than a preset distance;

S733, when the effective moving corner point leftward moving distance and the rightward moving distance between the adjacent two frames in the video stream are greater than a preset distance, the effective moving corner point is recorded as an effective wave corner point, and is effective The number of wave corner points N _move plus 1, the final calculated N _move is the number of effective wave corner points in the corner point set, the initial value of the number of effective wave corner points N _move is 0;

S741 determines whether the number of valid wave corner points N _move is greater than a second preset number;

S742: when the number of effective wave corner points N _{move is} greater than the second preset number, determining that the palm is waving;

S743 Otherwise, judge that the palm does not wave.

Specifically, the present embodiment describes how to determine whether the palm is waving or not according to the motion trajectory of each corner point in the corner point set.

The wave detection process is specifically as follows:

1 Starting from the detection of the standard gesture in the video frame, the opponent's palm is tracked, and the motion path of all tracking corner points A _i {i=1, 2, ..., N} in the palm area is counted as L _i {i=1, 2,.. .N}, N is the total number of corner points in the palm area. It is specified that when L _i >0 indicates that the corner point moves to the right of the standard gesture, when L _i <0 indicates that the corner point moves to the left of the standard gesture.

2 When L _i <-D _left indicates that the corner point has been effectively moved to the left, the corresponding flag of the corner point is M _i,left =true, where D _left is the threshold of the leftward moving distance, which can be set to two pixels. Distance; when L _i >D _right indicates that the corner point has been effectively moved to the right, the corresponding flag of the corner point is M _i,right =true, where D _right is the threshold of the rightward moving distance, which can be set to two pixels distance.

③ When the video stream path of motion between two corner points A _i L _i is smaller than the distance of two pixels, depending on the failure point for the angle tracking, tracking the number of active corners subtracting N _track 1, the initial value of N _track The total number of corner points in the corner points is N.

4 When the total number of corner points that are valid is tracked, N _track <5, the tracking is stopped. When the total number of effective corner points is tracked, N _track ≥ 5, it is determined whether the effective motion corner point in the video stream is valid both left and right. motion.

5 When a certain corner point A _i has been effectively moved to the left and to the right, that is, the motion flag M _i,left =true, M _i,right =true, the number of valid wave corner points N _{move is} increased by one. The initial value of the number of valid wave corner points N _move is zero.

6 real-time statistics of the number of effective wave corners N _move , when

It can be judged that the user is waving, while stopping the corner tracking and starting a new round of palm detection.

7 If you wave the total number of effective sports corners

Then this test is invalid and a new round of testing is started.

This embodiment proposes a method for accurately recognizing a human hand waving in a dynamic background, and the camera can detect whether a person is waving through a continuous video frame. The invention is essentially different from the recognition based on the frame difference method. Firstly, the palm is detected on the picture of each frame by the palm detector. Once the palm is detected, the palm is tracked, and the motion trajectory of the palm is used for the wave judgment. Since the palm detection is performed on a single frame picture and does not need to combine the upper and lower historical image frames, it is not interfered by the complex background and the dynamic background. That is to say, the detection method is accurate and stable, and can be free from the influence of complex backgrounds, and can accurately identify whether it is waving even if the motion of the background or other objects interferes.

As shown in FIG. 7, the present invention provides an embodiment of a robot's wave detection system, including:

a detecting module 4, configured to detect a palm in a standard position in the video stream by using a cascade classifier;

a corner point extraction module 5, electrically connected to the detection module 4, for extracting a corner point of the palm in a standard position, to obtain a corner point set;

The corner tracking module 6 is electrically connected to the corner point extraction module 5 for performing tracking detection on each corner point in the corner point set in the video stream to obtain a motion track corresponding to each corner point;

The judging module 7 is electrically connected to the corner tracking module 6 for judging whether the palm is waving according to the motion trajectory of each corner point in the corner point set.

First, the robot call detection module detects the palm in the standard position in the video stream through the cascade classifier; secondly, the corner point extraction module is called to extract the corner point of the palm in the standard position to obtain the corner point set; again, the call angle is The point tracking module performs tracking detection on each corner point in the corner point set in the video stream to obtain a motion track corresponding to each corner point; finally, the judgment module is called, according to the motion of each corner point in the corner point set. The trajectory determines whether the palm is waving.

In this embodiment, the robot can synchronously detect the motion trajectory of the palm and the corner of the palm when collecting the video stream, and does not perform the wave detection based on the historical image. The recognition rate of the palm of the hand using the Adaboost cascade classifier is relatively high, compared with In the prior art, in the complex background and the dynamic background, the palm of the standard position can be accurately detected in the video stream, which facilitates the tracking detection of the subsequent palm corner points.

As shown in FIG. 8, the present invention provides an embodiment of a wave detection system for a robot. Based on the previous embodiment, the present invention further includes:

Sample production module 1 for making a palm training sample;

The feature extraction module 2 extracts an LBP texture feature vector of the palm training sample, and obtains a palm training sample after the extraction process;

The classifier training module 3 is electrically connected to the feature extraction module 2 for training the cascade classifier by extracting the processed palm training samples.

Preferably, the feature extraction module 2 includes:

The processing sub-module 21 is configured to divide the detection window into N×N pixel regions, and compare one central pixel point in the pixel region with gray values of adjacent eight adjacent pixel points, where N∈ {64,32,16,8};

The feature value calculation sub-module 22 is electrically connected to the processing sub-module 21, and if the pixel value of the adjacent pixel point is greater than the pixel value of the central pixel point, the adjacent pixel point is marked as 1, otherwise 0. Thereby generating an 8-bit binary number as the LBP value of the central pixel point, whereby the LBP value of each pixel in the pixel region can be calculated;

The processing sub-module 21 is further configured to obtain an LBP value according to each pixel in the pixel region. Calculating a statistical histogram of the pixel region, and normalizing the statistical histogram;

The processing sub-module 21 is further configured to connect a statistical histogram of each pixel region of the palm training sample to form an LBP texture feature vector of the palm training sample;

The corner tracking module 6 is further configured to extract a corner point of the palm in a standard position by a corner detection algorithm, obtain a corner point set, and perform optical flow on all corner points in the corner point set in the video stream. Tracking, detecting the position of all the corner points in the corner point set according to the optical flow tracking algorithm, and calculating the motion trajectory corresponding to each corner point according to the sparse optical flow algorithm.

As shown in FIG. 3, the user moves back and forth between the 1, 2, and 3 positions during the waving process. In this embodiment, the Adaboost algorithm is combined with the LBP feature to detect the palm of the video frame. Since the algorithm does not have rotational invariance, it can only recognize the palm of the hand whose rotation angle is within ±15° at a certain standard position. In this embodiment, the gesture of selecting the palm at the 2 position is the standard palm, and the gestures at other positions are not recognized.

In the process of collecting the palm training samples, in order to facilitate the extraction of the LBP feature values, the selected palm sample is the palm sample of the palm relative to the standard position within the preset angle range, so as to better enhance the gesture of the cascade classifier near the standard position. The recognition success rate, the palm preset angle range may be a range of ±15°, that is, the sample includes a gesture sample with a relative standard position rotation angle within a range of ±15° to better improve the gesture recognition rate of the algorithm near the standard position. In order to improve the diversity of the sample when collecting the palm samples of different users in different backgrounds, when the collected palm training samples have a certain degree of shape and illumination difference, the obtained detection effect will be very good and can be more The success rate of the cascading classifier for gesture recognition is improved.

The process of extracting the LBP feature of the palm training sample by the feature extraction module is: a. dividing the detection window into 16×16 small cells, and for each pixel in each cell, the gray value of the adjacent 8 pixels In contrast, if the surrounding pixel value is greater than the center pixel value, the position of the pixel is marked as 1, otherwise 0. Thus, 8 points in the 3*3 neighborhood are compared to produce an 8-bit binary number, that is, the LBP value of the center pixel of the window is obtained; b. The histogram of each cell is calculated, that is, each number (assumed to be a decimal number) The frequency at which the LBP value appears; normalize the histogram Finally, c. Finally, the obtained statistical histogram of each cell is connected into a feature vector, that is, an LBP texture feature vector of the entire image.

G _m (x): x→{-1, +1}

Calculate the classification error rate on the training data set:

As can be seen from the above formula, when

D _m+1 =(w _m+1,1 ,w _m+1,2 ,...,w _m+1,N )

Combine each weak classifier to get the final classifier, as follows:

As shown in FIG. 9, the present invention provides an embodiment of a robot's wave detection system, including:

The judging module 7 is electrically connected to the corner tracking module 6 for judging whether the palm is waving according to the motion trajectory of each corner point in the corner point set;

The determining module 7 includes:

The analyzing sub-module 71 analyzes, according to the motion trajectory between two adjacent frames in the video stream and the preset pixel distance, the number of effective motion corner points in the corner point set is obtained;

The analysis sub-module 71 is further configured to analyze, according to the number of the effective motion corner points and the first preset number, the detection status of all the corner points in the corner point set;

The analysis sub-module 71 is further configured to analyze, according to the motion trajectory of the effective motion corner point, the number of effective wave corner points in the effective motion corner point when the detection condition satisfies a preset requirement;

The determining sub-module 72 is electrically connected to the analyzing sub-module 71, and is configured to determine whether the user is waving according to the number of valid wave corner points and the second preset number;

The determining module 7 further includes:

a calculation sub-module 73, configured to calculate, according to a motion trajectory between two adjacent frames in the video stream, the motion of the same corner between the adjacent two frames in the video stream distance;

The determining sub-module 72 is further electrically connected to the calculating sub-module 73, and configured to determine whether the moving distance is greater than a preset pixel distance;

The determining sub-module 72 is further configured to: when the moving distance is greater than a preset pixel distance, determine that the corner point performs effective motion, and use the corner point as the effective motion corner point;

Determining sub-module 72 is further configured, when the movement distance is less than the predetermined pixel distance, the corner point determined lost motion, and the motion of the effective number of corners _Track N minus 1, finally calculated N _Track is the number of effective motion corner points in the corner set, and the initial value of N _track is the total number N of corner points in the corner set;

The determining sub-module 72 is further configured to determine whether the number N _{Track of} the effective motion corner points is greater than the first preset number;

The determining sub-module 72 is further configured to stop detecting the motion _{track of} the effective motion corner point when the number N _{track of} the effective motion corner points is less than the first preset number, and determine that the user does not wave;

The determining sub-module 72 is further configured to continue to detect a motion trajectory of the effective motion corner point when the number N _{track of} the effective motion corner points is greater than the first preset number.

The determining module 7 further comprising detecting sub-module 74, configured to, when moving the active corners N _track number is greater than the first predetermined number, the video stream is detected in between two adjacent ones of said active The moving angle of the moving corner to the left and the distance to the right;

The determining sub-module 72 is further configured to determine whether the left moving distance and the right moving distance of the effective moving corner point between the adjacent two frames in the video stream are greater than a preset distance;

The determining sub-module 72 is further configured to: when the effective moving corner point leftward moving distance and the rightward moving distance between the adjacent two frames in the video stream are greater than a preset distance, the effective moving corner point is recorded as Effectively wave the corner point and increase the number of effective wave corner points N _move by 1. The final calculated N _move is the number of effective wave corner points in the corner point set, and the initial value of the number of valid wave corner points N _move Is 0;

The determining sub-module 72 is further configured to determine whether the number of valid wave corner points N _move is greater than a second preset number;

The determining sub-module 72 is further configured to: when the number of valid wave corner points N _{move is} greater than the second preset number, determine that the palm is waving; otherwise, determine that the palm has not waved;

a sample making module for making a palm training sample;

The feature extraction module 1 extracts an LBP texture feature vector of the palm training sample, and obtains a palm training sample after the extraction process;

a classifier training module 3, electrically connected to the feature extraction module 2, for training the cascade classifier by extracting the processed palm training samples;

The feature extraction module 2 includes:

The processing sub-module 21 is further configured to calculate a statistical histogram of the pixel region according to an LBP value of each pixel in the pixel region, and perform normalization processing on the statistical histogram;

According to the number of the effective motion corner points and the first preset number, the detection situation of all the corner points in the corner point set is obtained, and the specific detection condition is: one is the effective motion corner point The quantity is smaller than the first preset quantity, and the value of the first preset quantity is relatively small, and can be set to 5 to 10, because the number of extracted palm corner points is far more than 5-10, when the effective movement corner point The number is smaller than the first preset number, that is, most of the corner points are not effectively moved, and it can be judged that the palm is not waved.

Specifically, in this embodiment, how to determine whether the palm is waved according to the motion track of each corner point in the corner point set is described in more detail. The wave detection process is specifically:

2 When L _i <-D _left indicates that the corner point has been effectively moved to the left, the corresponding flag of the corner point is M _i,left =true, where D _left is the threshold of the leftward moving distance, which can be set to two pixels. Distance; when L _i >D _right indicates that the corner point has been effectively moved to the right, the corresponding flag of the corner point is M _i,right =true, where D _right is the threshold of the rightward moving distance, which can be set to two pixels distance;

③ When the video stream path of movement of the corner points A _i L _i between two pixel distance is less than two, considered the failure point tracking angle, the tracking number of corners effectively subtracting N _track 1, the initial value of N _track The total number of corner points in the corner points is N;

4 When the total number of corner points that are valid is tracked, N _track <5, the tracking is stopped. When the total number of effective corner points is tracked, N _track ≥ 5, it is determined whether the effective motion corner point in the video stream is valid both left and right. motion,

5 When a certain corner point A _i has been effectively moved to the left and to the right, that is, the motion flag M _i,left =true, M _i,right =true, the number of valid wave corner points N _{move is} increased by one. The initial value of the number of effective wave corner points N _move is 0;

6 real-time statistics of the number of effective wave corners N _move , when

7 If you wave the total number of effective sports corners

Then this test is invalid and a new round of testing is started.

It should be noted that the above embodiments can be freely combined as needed. The above description is only a preferred embodiment of the present invention, and it should be noted that those skilled in the art can also make several improvements and retouchings without departing from the principles of the present invention. It should be considered as the scope of protection of the present invention.

Claims

A method for detecting a wave of a robot, comprising:

The palm of the standard position is detected in the video stream by the cascade classifier;

Extracting a corner point of the palm in the standard position to obtain a corner point set;

Tracking and detecting each corner point in the corner point set in the video stream, and obtaining a motion track corresponding to each corner point;

Whether the palm is waving or not is determined according to the motion trajectory of each corner point in the corner point set.
The method for detecting a wave of a robot according to claim 1, wherein determining whether the user is waving according to the motion trajectory of each corner point in the corner point set comprises:

Obtaining, according to the motion trajectory between adjacent two frames in the video stream and the preset pixel distance of each corner point in the corner point set, the number of effective motion corner points in the corner point set is obtained;

When the number of the effective motion corner points is greater than the first preset number, tracking detection is continued for all corner points in the corner point set;

Obtaining, according to the motion trajectory of the effective motion corner point, the number of effective wave corner points in the effective motion corner point;

When the number of valid wave corner points is greater than the second preset number, it is determined that the user is waving.
The method for detecting a wave of a robot according to claim 2, wherein the trajectory of each corner point in the corner point set between two adjacent frames of the video stream and the preset pixel distance are analyzed. The number of effective motion corner points in the set of corner points specifically includes:

Calculating a motion distance of the same corner point between adjacent two frames in the video stream according to a motion trajectory between adjacent two frames in the video stream in each corner point of the video point set;

Determining whether the moving distance is greater than a preset pixel distance;

When the moving distance is greater than a preset pixel distance, determining that the corner point performs effective motion, and using the corner point as the effective motion corner point;

When the movement distance is less than the predetermined pixel distance, the corner point determined lost motion, and said effective amount of N track motion corner point minus 1, the final calculated the corner point is the N track The number of effective motion corner points in the set, and the initial value of N track is the total number N of corner points in the corner point set.
The method for detecting a wave of a robot according to any one of claims 1 to 3, wherein before the palm of the standard position is detected in the video stream by the cascade classifier, the method comprises:

Make a palm training sample;

Extracting an LBP texture feature vector of the palm training sample, and obtaining a palm training sample after the extraction process;

The cascade classifier is trained by extracting the processed palm training samples.
The method for detecting a wave of a robot according to claim 4, wherein the preparing the palm training sample comprises:

A sample of the palm of a different user in a different background is captured, the sample of the palm including a sample of the palm that rotates within a predetermined range of angles relative to the standard position.
The method of claim 4, wherein the process of extracting the LBP texture feature vector of the palm training sample comprises:

Dividing the detection window into N×N pixel regions, comparing a central pixel point in the pixel region with gray values of adjacent 8 adjacent pixel points, where N∈{64, 32, 16, 8};

If the pixel value of the adjacent pixel is greater than the pixel value of the central pixel, the adjacent pixel is marked as 1, otherwise 0, thereby generating an 8-bit binary number as the LBP value of the central pixel Thereby, the LBP value of each pixel in the pixel region can be calculated;

Calculating a statistical histogram of the pixel region according to an LBP value of each pixel in the pixel region, and normalizing the statistical histogram;

A statistical histogram of each pixel region of the palm training sample is connected to form an LBP texture feature vector of the palm training sample.
A method for detecting a wave of a robot according to any one of claims 1 to 3, wherein a corner point of a palm at a standard position is extracted, a corner point set is obtained, and the corner point is obtained in the video stream. Each corner point in the set is tracked and detected, and the motion track corresponding to each corner point is obtained as follows:

Extracting a corner point of the palm in the standard position by a corner detection algorithm to obtain a corner point set, and performing optical flow tracking on each corner point in the corner point set in the video stream, and detecting the optical flow tracking algorithm according to the optical flow tracking algorithm The position of each corner point in the corner point set is calculated, and the motion trajectory corresponding to each corner point is calculated according to the sparse optical flow algorithm.
A wave detection system for a robot, comprising:

a detecting module, configured to detect a palm in a standard position in the video stream by using a cascade classifier;

a corner point extraction module, electrically connected to the detection module, for extracting a corner point of the palm in a standard position, to obtain a corner point set;

The corner tracking module is electrically connected to the corner point extraction module, and is configured to perform tracking detection on each corner point in the corner point set in the video stream to obtain a motion track corresponding to each corner point;

The judging module is electrically connected to the corner tracking module for determining whether the palm is waving according to the motion trajectory of each corner point in the corner point set.
The wave detection system of a robot according to claim 8, wherein the determining module comprises:

The analysis sub-module analyzes the number of effective motion corner points in the corner point set according to the motion track between the adjacent two frames in the video stream and the preset pixel distance of each corner point in the corner point set;

The analysis sub-module is further configured to continue tracking detection of all corner points in the corner point set when the number of the effective motion corner points is greater than the first preset number;

The analysis sub-module is further configured to analyze according to the motion trajectory of the effective motion corner point The number of effective wave corners within the effective motion corner;

The determining sub-module is electrically connected to the analyzing sub-module, and is configured to determine that the user is waving when the number of valid waving corner points is greater than the second preset number.
The wave detection system of the robot according to claim 9, wherein the determining module further comprises:

a calculation submodule, configured to calculate a moving distance of the same corner point between adjacent two frames in the video stream according to a motion trajectory between two adjacent frames in the video stream according to each corner point in the corner point set ;

The determining sub-module is further electrically connected to the computing sub-module, and configured to determine whether the moving distance is greater than a preset pixel distance;

The determining sub-module is further configured to: when the moving distance is greater than a preset pixel distance, determine that the corner point performs effective motion, and use the corner point as the effective motion corner point;

The determining sub-module is further configured to: when the moving distance is less than the preset pixel distance, determine that the corner point performs an invalid motion, and decrement the number N Track of the effective motion corner point by a final calculation. N track is the number of effective motion corner points in the corner point set, and the initial value of N track is the total number N of corner points in the corner point set.
The method for detecting a wave of a robot according to any one of claims 8 to 10, further comprising:

a sample making module for making a palm training sample;

a feature extraction module, extracting an LBP texture feature vector of the palm training sample, and obtaining a palm training sample after the extraction process;

The classifier training module is electrically connected to the feature extraction module for training the cascade classifier by extracting the processed palm training samples.
A wave detection system for a robot according to claim 11, wherein the feature extraction module comprises:

a processing sub-module, configured to divide the detection window into N×N pixel regions, and compare one central pixel point in the pixel region with gray values of adjacent eight adjacent pixel points, where N∈{ 64,32,16,8};

The feature value calculation sub-module is electrically connected to the processing sub-module. If the pixel value of the adjacent pixel point is greater than the pixel value of the central pixel point, the adjacent pixel point is marked as 1, otherwise 0. Generating an 8-bit binary number as the LBP value of the central pixel point, thereby calculating an LBP value of each pixel in the pixel region;

The processing sub-module is further configured to calculate a statistical histogram of the pixel region according to an LBP value of each pixel in the pixel region, and perform normalization processing on the statistical histogram;

The processing sub-module is further configured to connect a statistical histogram of each pixel region of the palm training sample to form an LBP texture feature vector of the palm training sample.
A wave detection system for a robot according to any one of claims 8 to 10, characterized in that:

The corner tracking module is further configured to extract a corner point of the palm in a standard position by a corner detection algorithm, obtain a corner point set, and perform optical flow on each corner point in the corner point set in the video stream. Tracking, detecting the position of all the corner points in the corner point set according to the optical flow tracking algorithm, and calculating the motion trajectory corresponding to each corner point according to the sparse optical flow algorithm.
A robot characterized by integrating a wave detecting system of a robot according to any one of claims 8 to 13.