CN115469665A - Intelligent wheelchair target tracking control method and system suitable for dynamic environment - Google Patents

Abstract

The invention discloses an intelligent wheelchair target tracking control method and system suitable for a dynamic environment, wherein the system comprises a target detection module, an obstacle detection module, a motion planning module and a motion control module; the target detection module is used for identifying a target and predicting the position of the target according to historical information; the obstacle detection module is used for identifying obstacles and predicting the probability distribution of the positions of the obstacles according to historical information; the motion planning module is used for respectively constructing a gravitational potential field and a repulsive potential field according to the predicted target position and the probability distribution of the predicted obstacle position, and the gravitational potential field and the repulsive potential field jointly act on the wheelchair to generate a safe path; and the motion control module is used for tracking the safety path generated by the motion planning module and executing the motion instruction of the intelligent wheelchair. The invention solves the problem that the current intelligent wheelchair is difficult to realize the autonomous following dynamic target in the dynamic obstacle environment.

Description

A method and system for intelligent wheelchair target tracking control adapted to dynamic environment

technical field

The invention relates to the technical field of intelligent wheelchair target tracking, in particular to an intelligent wheelchair target tracking control method and system adapted to a dynamic environment.

Background technique

The aging of society is becoming more and more serious. As a special tool that can improve the quality of life and freedom of movement of the elderly and disabled, smart wheelchairs have extremely broad development prospects and social value. Although some advanced smart wheelchairs have certain functions of navigation and obstacle avoidance, it is still difficult to cope with complex environments and complete complex tasks. The trajectory of smart wheelchairs based on static scene planning is difficult to cope with the ever-changing dynamic environment.

Contents of the invention

In view of the above defects, the present invention proposes an intelligent wheelchair target tracking control method and system adapted to a dynamic environment, the purpose of which is to solve the problem that current intelligent wheelchairs are difficult to autonomously follow a dynamic target in the case of dynamic obstacles.

For reaching this purpose, the present invention adopts following technical scheme:

An intelligent wheelchair target tracking control method adapted to a dynamic environment, comprising the following steps:

Step S1: Identify the target and predict the target position according to the historical information;

Step S2: Identify obstacles and predict the probability distribution of obstacle positions based on historical information;

Step S3: Construct the gravitational potential field and the repulsive potential field respectively according to the predicted target position and the probability distribution of the predicted obstacle position, and the gravitational potential field and the repulsive potential field work together to generate a safe path for the smart wheelchair;

Step S4: Generate an intelligent wheelchair movement instruction based on the intelligent wheelchair safety path controller, and execute the intelligent wheelchair movement instruction.

Preferably, step S1 specifically includes the following steps:

Step S11: Acquire the image data of the target, select the Haar-like feature as the gray feature of the target, use the semi-supervised On-line Boosting algorithm to process the image data, calculate the image space position coordinates of the target, and further obtain the target's position based on the depth image information. Task space position coordinates;

Step S12: Based on the historical data of the target pose coordinates in the task space, use the least squares support vector regression machine (Least Squares Support Vector Machine, LS-SVR) algorithm to predict the task space pose of the target, and obtain the predicted dynamic target Pose information P _p (x _p , y _p );

Step S13: Calculate the final desired position _PD (x _D , y _D ) of the wheelchair according to the predicted dynamic target pose information P _p (x _p , y _p ) and the set following orientation θ _f .

Preferably, in step S11, the On-line Boosting algorithm specifically includes the following steps:

Step S111: Select the target area, enlarge the target area according to a certain ratio to form a search area; the search area includes the possible position of the target after the next movement; when both the target area and the search area are determined, According to the position of the target, select samples to train the prior classifier and tracker in turn, and each training uses 5 samples, 5 samples including the target itself and 4 samples around the target;

Step S112: Divide the search area into a plurality of small blocks, wherein the divided small blocks are the same size as the target area, and there are overlapping parts between the divided small blocks; use the tracker trained in the previous stage The device evaluates all the small blocks that have been divided, calculates the credibility of each small block as the target, and selects the small block with the highest reliability as the position of the target at the next moment;

Step S113: According to the predicted new position of the target, use the tracker to reselect samples for training, and update the parameters of the classifier, where the new target position and 4 surrounding small blocks are used as samples, and the classifier is used to determine the position of the sample Label.

Preferably, in step S12, the LS-SVR algorithm specifically includes the following steps:

选择适当的模型参数γ>0；其中，d是训练集大小，m是测试集大小；l是数据集大小；x_i代表输入样本的第i行，yⁱ表示相应输出值；Step S121: given data set

Select an appropriate model parameter γ>0; where, d is the size of the training set, m is the size of the test set; l is the size of the data set; x _i represents the i-th row of the input sample, and y ⁱ represents the corresponding output value;

作为核函数；Step S122: Select radial basis function

as a kernel function;

Step S123: Calculate p=H ^-1 y, q=H ^-1 L and s=L ^T q; where p is the wheelchair pose, H is a positive definite matrix, q is a transformation matrix, L=(1,..., 1) ^T ∈ R ^l ;

Step S124: Calculate b ^* = ^ηT y/s and a ^* =p-bq; wherein b is an offset vector; n is a parameter matrix; a ^* is a vector composed of Lagrangian multipliers;

Step S125: Constructing a regression function

Preferably, in step S13, the final desired position of the wheelchair can be obtained by formula (1):

Step S13: Let the predicted target task space pose be P _p (x _p , y _p ), set the following orientation θ _f , and obtain the expected position P _D (x _D , y _D ) of the smart wheelchair according to formula (1) ):

where d is the relative distance expected to be maintained during the follow-up.

Preferably, in step S2, the following steps are specifically included:

和当前时刻该障碍物的位置

计算得到下一时刻该障碍物的位置

Step S21: pass the position of the i-th obstacle at the previous moment

and the position of the obstacle at the current moment

Calculate the position of the obstacle at the next moment

Step S22: Record the displacement of the obstacle at two adjacent moments as (o _x,i ,o _y,i ), conduct a probabilistic analysis on the obstacle, and obtain the expected value μ _x and μ _y of the obstacle, and the variance σ _x and σ _y , covariance σ _xy , and correlation coefficient ρ _xy :

Among them, N is the amount of historical data involved in the prediction, and the most recent N historical sequences are intercepted as predictions, which can avoid data expansion in the later stage of planning;

Step S23: divide the coordinate system into several grids, and the probability density U _ob (m, n) of each grid (m, n) can be obtained by the following formula:

Among them, U _ob (m,n) is the probability density of each grid (m,n).

Preferably, step S3 specifically includes the following steps:

Step S31: According to the calculation principle of step S13, the final expected position P _D (x _D , y _D ) of the smart wheelchair is obtained, and the gravitational potential field is formed according to the final expected position of the smart wheelchair as the target point, wherein the gravitational potential field is in the grid The potential energy U _att (m,n) generated by the lattice position (m,n) is expressed by formula (8):

Among them, S represents the size of the actual motion environment, (x _D , y _D ) represents the target position;

Step S32: Obtain the resultant potential field U _t (m, n) of the current point (x, y) according to the superposition calculation of the gravitational potential field and the repulsive potential field; the specific calculation is:

U _t (m,n)=U _att (m,n)+U _ob (m,n) (9)

Step S33: Perform a partial derivative operation on the resultant potential field to obtain the following potential field gradient:

According to the potential field gradient, the expected position of the smart wheelchair at the next moment is calculated, and the specific calculation is as follows:

Among them, x _d and y _d represent the desired X-axis coordinates and Y-axis coordinates of the wheelchair at the next moment; θ _d represents the expected posture of the wheelchair at the next moment, and x and y represent the X-axis coordinates and Y-axis coordinates of the wheelchair at the current moment; D Indicates the moving distance of the wheelchair in one cycle, which can be regarded as a reference speed value. The smaller the value, the safer the movement of the smart wheelchair; R is the reference value of the moving distance of the smart wheelchair in one cycle.

R is expressed as follows:

Preferably, in step S4, the generation of the intelligent wheelchair motion command is based on the trajectory planned by the potential field method tracked by the inversion controller of the kinematic model, specifically including the following steps:

Step S41: Introduce virtual input α, according to the equation of robot kinematics, take

Where v is the linear velocity of the robot, and the Lyapunov function is used to judge the stability of the nonlinear system; let the Lyapunov function V ₁ be

Where e _x represents the position error of the wheelchair in the X direction, and e _y represents the position error of the wheelchair in the Y direction;

From formula (16) can get

By designing the virtual quantity α, so that

Among them, x _d and y _d represent the desired X-axis coordinates and Y-axis coordinates of the wheelchair at the next moment, then

Among them, c ₁ and c ₂ are adjustable parameters;

将线速度v和虚拟控制律α设计为：make

The linear velocity v and virtual control law α are designed as:

Then guarantee formula (20) is established;

Step S42: let e=α-θ, define the Lyapunov function V ₂ as:

but

The angular velocity control law ω is designed as:

where c ₃ is an adjustable parameter, then

Where C _m is a constant, C _m ≤ min(c ₁ ,c ₂ ,c ₃ );

即V₂(t)以指数形式收敛于零，从而t→∞时，e_x→0，e_y→0，θ→θ_d且以指数形式收敛。but

That is, V ₂ (t) converges to zero exponentially, so when t→∞, _{ex→0, e y → 0} , θ→θd converge exponentially.

Another aspect of the present application provides an intelligent wheelchair target tracking control system adapted to a dynamic environment, the system comprising a target detection module, an obstacle detection module, a motion planning module and a motion control module;

The target detection module is used to identify the target and predict the position of the target according to historical information;

The obstacle detection module is used to identify obstacles and predict the probability distribution of obstacle positions according to historical information;

The motion planning module is used to construct the gravitational potential field and the repulsive force potential field respectively according to the probability distribution of the predicted target position and the predicted obstacle position, and the gravitational potential field and the repulsive force potential field work together to generate a safe path for the intelligent wheelchair;

The motion control module is used to track the safe path generated by the motion planning module, and execute the motion command of the intelligent wheelchair.

The technical solutions provided by the embodiments of the present application may include the following beneficial effects:

In this program, the target detection module is combined with the semi-supervised On-lineBoosting algorithm to identify the target, and the least squares support vector regression method is used to analyze the predicted target position according to the historical data; at the same time, the obstacle detection module identifies the obstacle, and according to The historical data analysis predicts the probability distribution of the obstacle position, generates the gravitational potential field based on the predicted target position, generates the repulsive potential field based on the predicted probability distribution of the obstacle position, and plans a safe following through the joint action of the two potential fields. path. The smart wheelchair moves according to the planned following path, realizing the stable, safe and efficient following of the target object by the smart wheelchair in a dynamic environment.

Description of drawings

Fig. 1 is a functional block diagram of an intelligent wheelchair target tracking control system adapted to a dynamic environment;

Figure 2 is a frame diagram of the hardware structure of the intelligent wheelchair.

detailed description

Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the figures are exemplary only for explaining the present invention and should not be construed as limiting the present invention.

An intelligent wheelchair target tracking control method adapted to a dynamic environment, comprising the following steps:

Step S1: Identify the target and predict the target position according to the historical information;

Step S2: Identify obstacles and predict the probability distribution of obstacle positions based on historical information;

Step S3: Construct the gravitational potential field and the repulsive potential field respectively according to the predicted target position and the probability distribution of the predicted obstacle position, and the gravitational potential field and the repulsive potential field work together to generate a safe path for the smart wheelchair;

Step S4: Generate an intelligent wheelchair movement instruction based on the intelligent wheelchair safety path controller, and execute the intelligent wheelchair movement instruction.

This method uses the target detection module, that is, the visual sensor combined with the semi-supervised On-line Boosting algorithm to identify the target, and uses the least squares support vector regression method to analyze and predict the target position according to the historical data; at the same time, the obstacle detection module, That is, the laser sensor identifies obstacles, analyzes and predicts the probability distribution of obstacle positions based on historical data, generates a gravitational potential field based on the predicted target position, and generates a repulsive potential field based on the predicted probability distribution of obstacle positions. Work together to plan a safe following path. The smart wheelchair moves according to the planned following path, realizing the stable, safe and efficient tracking of the target object by the smart wheelchair in a dynamic environment.

Preferably, step S1 specifically includes the following steps:

Step S11: Acquire the image data of the target, select the Haar-like feature as the gray feature of the target, use the semi-supervised On-line Boosting algorithm to process the image data, calculate the image space position coordinates of the target, and further obtain the target's position based on the depth image information. Task space position coordinates;

Step S12: Based on the historical data of the target pose coordinates in the task space, use the least squares support vector regression machine (Least Squares Support Vector Machine, LS-SVR) algorithm to predict the task space pose of the target, and obtain the predicted dynamic target Pose information P _p (x _p , y _p );

Step S13: Calculate the final desired position _PD (x _D , y _D ) of the wheelchair according to the predicted dynamic target pose information P _p (x _p , y _p ) and the set following orientation θ _f .

In this embodiment, the image data is processed through a semi-supervised On-line Boosting algorithm, the image space position coordinates of the target are calculated, and then the position coordinates of the target point in the task space are obtained based on the depth image information. Then, based on the historical data of the target position coordinates, the LS-SVR algorithm is used to predict the target position, and a reliable prediction of the dynamic target is obtained. This is beneficial to reduce the time lag of the system and improve the response speed of the system to follow the dynamic target.

Preferably, in step S11, the On-line Boosting algorithm specifically includes the following steps:

Step S111: Select the target area, enlarge the target area according to a certain ratio to form a search area; the search area includes the possible position of the target after the next movement; when both the target area and the search area are determined, According to the position of the target, select samples to train the prior classifier and tracker in turn, and each training uses 5 samples, 5 samples including the target itself and 4 samples around the target;

Step S112: Divide the search area into a plurality of small blocks, wherein the divided small blocks are the same size as the target area, and there are overlapping parts between the divided small blocks; use the tracker trained in the previous stage The device evaluates all the small blocks that have been divided, calculates the credibility of each small block as the target, and selects the small block with the highest reliability as the position of the target at the next moment;

Step S113: According to the predicted new position of the target, use the tracker to reselect samples for training, and update the parameters of the classifier, where the new target position and 4 surrounding small blocks are used as samples, and the classifier is used to determine the position of the sample Label.

Specifically, the semi-supervised On-line Boosting algorithm considers that the labels of all samples in the update stage are unknown except for the labels of samples in the initial stage. The labels of samples in the update stage are determined by an offline prior classifier carrying only the original target information. This prior classifier is only trained following the target initialization phase and does not participate in updates thereafter. Therefore, it only contains information about the original target, that is, prior information.

There are two strong classifiers in the semi-supervised On-line Boosting algorithm: prior classifier and tracker. Among them, the prior classifier does not participate in the update, it is only used to evaluate the matching degree between the current sample and the original target sample. According to the degree of matching between samples, assign a label to the current sample. After the samples are labeled, these samples can be used to update the tracker in a supervised learning manner. The entire semi-supervised On-lineBoosting algorithm mainly includes an initialization training phase and a tracking phase, and the tracking phase is further divided into a prediction phase and an update phase.

Preferably, in step S12, the LS-SVR algorithm specifically includes the following steps:

选择适当的模型参数γ>0；其中，d是训练集大小，m是测试集大小；l是数据集大小；x_i代表输入样本的第i行，yⁱ表示相应输出值；Step S121: given data set

Select an appropriate model parameter γ>0; where, d is the size of the training set, m is the size of the test set; l is the size of the data set; x _i represents the i-th row of the input sample, and y ⁱ represents the corresponding output value;

作为核函数；Step S122: Select radial basis function

as a kernel function;

Step S123: Calculate p=H ^-1 y, q=H ^-1 L and s=L ^T q; where p is the wheelchair pose, H is a positive definite matrix, q is a transformation matrix, L=(1,..., 1) ^T ∈ R ^l ;

Step S124: Calculate b ^* = ^ηT y/s and a ^* =p-bq; wherein b is an offset vector; n is a parameter matrix; a ^* is a vector composed of Lagrangian multipliers;

Step S125: Constructing a regression function

When the smart wheelchair tracks the target, there is a common problem of "short-sightedness". It only considers the current position and does not consider the future position of the target. Therefore, in order to better track the dynamic target, it is necessary to understand the movement rules of the dynamic target in the environment, that is, to predict the dynamic target. In this embodiment, the least squares support vector regression (LS-SVR) method adopted considers dynamic target trajectory prediction as a small sample, non-linear time series data analysis and prediction problem. In the problem, the input is time, and the output is the dynamic target trajectory corresponding to this period of time. According to the trajectory of the past period of time to learn the motion model of the target, that is, the nonlinear relationship between time and dynamic targets, so as to predict the trajectory of the target for a period of time in the future.

Specifically, the nonlinear relationship can be well approximated by LS-SVR for trajectory prediction of the target. When using SVR for trajectory prediction, the problem is regarded as a time series data prediction problem, that is, the mapping relationship between time and dynamic target trajectory is established. The input variable X=(t ₁ ,t ₂ ,…,t _n ) in the training data is a time series, and the corresponding output variable Y=(p ₁ ,p ₂ ,…,p _n ), where p _k represents the pose.

Selecting different kernel functions has a great influence on the regression effect of SVR, and the trajectory of the dynamic target is generally nonlinear. In this embodiment, the radial basis function (RBF) kernel function is used to simply process the data. It can produce good results for both linear regression and nonlinear regression.

Preferably, in step S13, the final desired position of the wheelchair can be obtained by formula (1):

Step S13: Let the predicted target task space pose be P _p (x _p , y _p ), set the following orientation θ _f , and obtain the expected position P _D (x _D , y _D ) of the smart wheelchair according to formula (1) ):

where d is the relative distance expected to be maintained during the follow-up.

Specifically, the calculation of the expected position P _D (x _D , y _D ) of the smart wheelchair enables the smart wheelchair to better plan the walking path, and further realizes that the smart wheelchair can autonomously follow the dynamic target object in the case of dynamic obstacles.

Preferably, step S2 specifically includes the following steps:

和当前时刻该障碍物的位置

计算得到下一时刻该障碍物的位置

Step S21: pass the position of the i-th obstacle at the previous moment

and the position of the obstacle at the current moment

Calculate the position of the obstacle at the next moment

Step S22: Record the displacement of the obstacle at two adjacent moments as (o _x,i ,o _y,i ), conduct a probabilistic analysis on the obstacle, and obtain the expected value μ _x and μ _y of the obstacle, and the variance σ _x and σ _y , covariance σ _xy , and correlation coefficient ρ _xy :

Among them, N is the amount of historical data involved in the prediction, and the most recent N historical sequences are intercepted as predictions, which can avoid data expansion in the later stage of planning;

Step S23: divide the coordinate system into several grids, and the probability density U _ob (m, n) of each grid (m, n) can be obtained by the following formula:

Among them, U _ob (m,n) is the probability density of each grid (m,n).

Specifically, when the smart wheelchair tracks the target point in a dynamic environment, effective navigation and obstacle avoidance is an essential part, and the smart wheelchair needs to plan a safe path without collision. Some traditional obstacle avoidance methods do not consider the dynamic characteristics of obstacles. This application adopts a navigation method that adapts to dynamic obstacles, that is, the random potential field method (Probability potential fields.PPF). The environmental information around the smart wheelchair is collected and transmitted to the industrial computer, which communicates through the Ethernet connection. The application program processes the collected data, predicts the probability distribution of the position of the obstacle at the next moment, and generates a collision-free trajectory.

Preferably, step S3 specifically includes the following steps:

Step S31: According to the calculation principle of step S13, the final expected position P _D (x _D , y _D ) of the smart wheelchair is obtained, and the gravitational potential field is formed according to the final expected position of the smart wheelchair as the target point, wherein the gravitational potential field is in the grid The potential energy U _att (m,n) generated by the lattice position (m,n) is expressed by formula (8):

Among them, S represents the size of the actual motion environment, (x _D , y _D ) represents the target position;

Step S32: Obtain the resultant potential field U _t (m, n) of the current point (x, y) according to the superposition calculation of the gravitational potential field and the repulsive potential field; the specific calculation is:

U _t (m,n)=U _att (m,n)+U _ob (m,n) (9)

Step S33: Perform a partial derivative operation on the resultant potential field to obtain the following potential field gradient:

According to the potential field gradient, the expected position of the smart wheelchair at the next moment is calculated, and the specific calculation is as follows:

Among them, x _d and y _d represent the desired X-axis coordinates and Y-axis coordinates of the wheelchair at the next moment; θ _d represents the expected posture of the wheelchair at the next moment, and x and y represent the X-axis coordinates and Y-axis coordinates of the wheelchair at the current moment; D Indicates the moving distance of the wheelchair in one cycle, which can be regarded as a reference speed value. The smaller the value, the safer the movement of the smart wheelchair; R is the reference value of the moving distance of the smart wheelchair in one cycle.

R is expressed as follows:

Specifically, the collision-free trajectory planned for the smart wheelchair through the random potential field method can make the smart wheelchair safe and effective when tracking the dynamic target object.

Preferably, in step S4, the generation of the motion instruction of the intelligent wheelchair is based on the trajectory planned by the potential field method tracked by the inversion controller of the kinematic model, specifically comprising the following steps:

Step S41: Introduce virtual input α, according to the equation of robot kinematics, take

Where v is the linear velocity of the robot, and the Lyapunov function is used to judge the stability of the nonlinear system; let the Lyapunov function V ₁ be

Where e _x represents the position error of the wheelchair in the X direction, and e _y represents the position error of the wheelchair in the Y direction;

e _x = x _d -x (15)

e _y =y _d -y

From formula (16) can get

By designing the virtual quantity α, so that

Among them, x _d and y _d represent the desired X-axis coordinates and Y-axis coordinates of the wheelchair at the next moment, then

Among them, c ₁ and c ₂ are adjustable parameters;

将线速度v和虚拟控制律α设计为：make

The linear velocity v and virtual control law α are designed as:

Then guarantee formula (20) is established;

Step S42: let e=α-θ, define the Lyapunov function V ₂ as:

but

The angular velocity control law ω is designed as:

where c ₃ is an adjustable parameter, then

Where C _m is a constant, C _m ≤ min(c ₁ ,c ₂ ,c ₃ );

即V₂(t)以指数形式收敛于零，从而t→∞时，e_x→0，e_y→0，θ→θ_d且以指数形式收敛。but

That is, V ₂ (t) converges to zero exponentially, so when t→∞, _{ex→0, e y → 0} , θ→θd converge exponentially.

This application uses an inversion controller based on a kinematic model to track the trajectory planned by the potential field method. During the tracking process of the controller, the odometer based on the incremental encoder disc locates the smart wheelchair in real time and provides motion feedback to the controller.

为了实现θ跟踪θ_d，步骤S42要保证θ跟踪α。In one embodiment, in step S41, set x _e =0, y _e =0, then

In order to realize that θ tracks θ _d , step S42 should ensure that θ tracks α.

Another aspect of the present application provides an intelligent wheelchair target tracking control system adapted to a dynamic environment, the system comprising a target detection module, an obstacle detection module, a motion planning module and a motion control module;

The target detection module is used to identify the target and predict the position of the target according to historical information;

The obstacle detection module is used to identify obstacles and predict the probability distribution of obstacle positions;

The motion planning module is used to construct the gravitational potential field and the repulsive force potential field respectively according to the probability distribution of the predicted target position and the predicted obstacle position, and the gravitational potential field and the repulsive force potential field work together to generate a safe path for the intelligent wheelchair;

The motion control module is used to track the safe path generated by the motion planning module, and execute the motion command of the intelligent wheelchair.

An intelligent wheelchair target tracking control system adapted to a dynamic environment in this scheme, as shown in Figure 1-2, the intelligent wheelchair includes a power supply component, a main control component, a motion component and a sensing component; The power supply module provides DC power for the entire intelligent wheelchair system. As the main control component of the intelligent wheelchair system, the industrial computer is equipped with a perception component composed of a visual sensor and a laser sensor. The motion components of the smart wheelchair are composed of a DC motor and a servo drive with a code disc installed on the wheels. When the smart wheelchair follows and controls the target, the perception component acquires target information and environmental information, sends them to the industrial computer for data processing and analysis, and the main control component sends instructions to the motion component to control the wheel speed and direction.

To further explain, in the power supply components, the industrial computer, servo driver and laser sensor are directly powered by 24V DC power supply, and the display screen is powered by stepping down the 24V DC power supply to 12V through a transformer. In the main control component, on the one hand, the industrial computer uses the written application program to receive and analyze the data of external sensors; on the other hand, it sends control instructions to the smart wheelchair through the application program to perform different tasks. In the perception component, the laser sensor and the industrial computer are connected and communicated through the Ethernet port. The laser sensor collects the environmental information around the smart wheelchair and transmits it to the industrial computer; the vision sensor and the industrial computer are connected through USB to transmit the collected image information to the industrial computer. In the motion component, the servo driver and the industrial computer are connected through the CAN bus to receive control commands and send relevant data about the movement of the smart wheelchair. The driver receives the control commands and processes and converts them through the built-in integrated chip to generate the motor speed transmission. The motor drives the smart wheelchair to move; at the same time, the built-in data interface of the driver can obtain the precise position of the motor in real time through the incremental encoder, and reversely transmit the data to the industrial computer, thereby realizing the feedback control of the status information.

In this scheme, the target detection module is used to identify the target and predict the target position according to the historical information. At the same time, the obstacle detection module identifies the obstacle and predicts the probability distribution of the obstacle position according to the historical information, and generates a gravitational potential field based on the predicted target position. A repulsive potential field is generated based on the probability distribution of the predicted obstacle position, and a safe following path is planned through the joint action of the two potential fields. This solution is different from the traditional target following method. Based on the historical data of the target detection module and obstacle detection module, the system has the ability to predict the movement of targets and obstacles, so that the planned trajectory has better adaptability to the dynamic environment.

In addition, each functional unit in each embodiment of the present invention may be integrated into one processing module, each unit may exist separately physically, or two or more units may be integrated into one module. The above-mentioned integrated modules can be implemented in the form of hardware or in the form of software function modules. If the integrated modules are realized in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium.

Although the embodiment of the present invention has been shown and described above, it can be understood that the above embodiment is exemplary and should not be construed as a limitation of the present invention, and those skilled in the art can make the above-mentioned Implementing implements changes, modifications, substitutions and variations.

Claims (9)

1. An intelligent wheelchair target tracking control method suitable for a dynamic environment is characterized by comprising the following steps: the method comprises the following steps:

step S1: identifying a target and predicting a target location from historical information;

step S2: identifying an obstacle and predicting a probability distribution of the obstacle position from the historical information;

and step S3: respectively constructing an attraction force potential field and a repulsion force potential field according to the predicted target position and the predicted probability distribution of the position of the obstacle, and generating a safe path of the intelligent wheelchair under the combined action of the attraction force potential field and the repulsion force potential field;

and step S4: and generating an intelligent wheelchair motion instruction based on the intelligent wheelchair safety path controller, and executing the intelligent wheelchair motion instruction.

2. The intelligent wheelchair target tracking control method adapted to the dynamic environment as claimed in claim 1, characterized in that: in step S1, the method specifically includes the following steps:

step S11: acquiring image data of a target, selecting a Haar-like feature as a target gray feature, processing the image data by using a semi-supervised On-line Boosting algorithm, calculating to obtain an image space position coordinate of the target, and further obtaining a task space position coordinate of the target based On depth image information;

step S12: predicting the task space pose of the target by using a Least square support vector regression (LS-SVR) algorithm based on the historical data of the target pose coordinate in the task space to obtain predicted dynamic target pose information P _p (x _p ,y _p )；

Step S13: from predicted dynamic object pose information P _p (x _p ,y _p ) And a set following orientation theta _f Calculating the final desired position P of the wheelchair _D (x _D ,y _D )。

3. The intelligent wheelchair target tracking control method adapted to the dynamic environment as claimed in claim 2, characterized in that: in step S11, the On-lineBoosting algorithm specifically includes the following steps:

step S111: selecting a target area, and amplifying the target area according to a certain proportion to form a search area; the search area comprises a possible position where the target is located after the target moves next time; after the target area and the search area are determined, selecting samples to train a prior classifier and a tracker in sequence according to the position of the target, wherein 5 samples are used for each training, and the 5 samples comprise the target and 4 samples around the target;

step S112: dividing the search area into a plurality of small blocks, wherein the size of the divided small blocks is the same as that of the target area, and the divided small blocks have overlapping parts; evaluating all the divided small blocks by using the tracker trained in the previous stage, calculating the credibility of each small block as a target, and selecting the small block with the highest credibility as the position of the target at the next moment;

step S113: and according to the predicted new target position, reselecting a sample by using a tracker for training, and updating the parameters of the classifier, wherein the label of the sample is determined by using the new target position and 4 peripheral small blocks as samples through a prior classifier.

4. The intelligent wheelchair target tracking control method adapted to the dynamic environment as claimed in claim 2, characterized in that: in step S12, the LS-SVR algorithm specifically includes the following steps:

step S121: given data set

Selecting the appropriate model parameter gamma>0; wherein d is the training set size and m is the test set size; l is the data set size; x is the number of _i Line i, y representing the input sample ⁱ Representing the respective output value;

step S122: selecting radial basis functions

As a kernel function;

step S123: calculation of p = H ^-1 y，q＝H ^-1 L and s = L ^T q; wherein p is the pose of the wheelchair, H is the positive definite matrix, q is the transformation matrix, L = (1, · ·, 1) ^T ∈R ^l ；

Step S124: calculation of b ^* ＝η ^T y/s and a ^* = p-bq; wherein b is an offset vector; eta is a parameter matrix; a is ^* A vector composed of Lagrange multipliers;

step S125: constructing a regression function

5. The intelligent wheelchair target tracking control method adapted to dynamic environment as claimed in claim 2, characterized in that: in step S13, the final desired position of the wheelchair can be obtained from equation (1):

step S13: the predicted target task space pose is P _p (x _p ,y _p ) Set following azimuth θ _f According to the formula (1), the expected position P of the intelligent wheelchair can be obtained _D (x _D ,y _D )：

Where d is the relative distance desired to be maintained in following.

6. The intelligent wheelchair target tracking control method adapted to the dynamic environment as claimed in claim 1, characterized in that: in step S2, the method specifically includes the following steps:

step S21: passing the position of the ith obstacle at the last moment

And the position of the obstacle at the current time

Calculating to obtain the position of the obstacle at the next moment

Step S22: the displacement of the obstacle at two adjacent times is recorded as (o) _x,i ,o _y,i ) Performing probabilistic analysis on the obstacle to obtain the obstacleExpected value mu _x And mu _y Variance σ _x And σ _y Covariance σ _xy And correlation coefficient ρ _xy ：

N is the historical data volume participating in prediction, and the latest N historical sequences are intercepted as predictions, so that data expansion in the later planning stage can be avoided;

step S23: dividing the coordinate system into a number of grids, the probability density U of each grid (m, n) _ob (m, n) can be obtained by the following formula:

wherein, U _ob (m, n) is the probability density of each grid (m, n).

7. The intelligent wheelchair target tracking control method adapted to dynamic environment as claimed in claim 2, characterized in that: in step S3, the method specifically includes the following steps:

step S31: according to the calculation principle of the step S13, the final expected position P of the intelligent wheelchair is obtained _D (x _D ,y _D ) And according to the final expectation of the intelligent wheelchairThe position as a target point forms an attractive potential field, wherein the potential energy U generated by the attractive potential field at the grid position (m, n) _att (m, n) is represented by formula (8):

where S represents the size of the actual motion environment, (x) _D ,y _D ) Representing a target location;

step S32: calculating the resultant force field U of the current point (x, y) according to the superposition of the attractive force potential field and the repulsive force potential field _t (m, n); the specific calculation is as follows:

U _t (m,n)＝U _att (m,n)+U _ob (m,n) (9)

step S33: and carrying out deviation calculation on the generated potential field to obtain the following potential field gradient:

and (3) calculating the expected position of the intelligent wheelchair at the next moment according to the potential field gradient, wherein the specific calculation is as follows:

wherein x is _d And y _d Representing the expected X-axis coordinate and Y-axis coordinate of the wheelchair at the next moment; theta _d Representing the expected posture of the wheelchair at the next moment, and X and Y representing the X-axis coordinate and the Y-axis coordinate of the wheelchair at the current moment; d represents the moving distance of the wheelchair in one period and can be regarded as a reference speed value, and the smaller the value is, the safer the movement of the intelligent wheelchair is; and R is a reference norm value of the movement distance of the intelligent wheelchair in one period.

R represents the following:

8. the intelligent wheelchair target tracking control method adapted to dynamic environment of claim 1, characterized in that: in step S4, the generation of the motion instruction of the intelligent wheelchair is a potential field method planned track tracked by an inversion controller based on a kinematic model, and specifically includes the following steps:

step S41: introducing virtual input alpha, and taking the alpha according to the kinematic equation of the robot

Wherein v is the linear velocity of the robot, and the Lyapunov function is used for judging the stability of the nonlinear system; let Lyapunov function V ₁ Is composed of

Wherein e _x Indicating the position error of the wheelchair in the X direction, e _y Indicating a position error of the wheelchair in the Y direction;

is obtained by the formula (16)

By designing the virtual quantity alpha such that

Wherein x is _d And y _d Representing the desired X-axis and Y-axis coordinates of the wheelchair at the next moment, then

Wherein, c ₁ 、c ₂ Is an adjustable parameter;

order to

The linear velocity v and the virtual control law α are designed as follows:

ensuring that equation (20) holds;

step S42: let e = α - θ, define the Lyapunov function V ₂ Comprises the following steps:

then

Wherein, ω is the angular velocity of the robot, and the angular velocity control law ω is designed as:

wherein c is ₃ To adjust the parameters, then

Wherein C is _m Is a constant number, C _m ≤min(c ₁ ,c ₂ ,c ₃ )；

Then

I.e. V ₂ (t) converges exponentially to zero, so that t → ∞ time, e _x →0，e _y →0，θ→θ _d And converges exponentially.

9. The utility model provides an intelligence wheelchair target tracking control system who is adapted to dynamic environment which characterized in that: the intelligent wheelchair target tracking control method adapted to a dynamic environment of any one of claims 1-8, the system comprising a target detection module, an obstacle detection module, a motion planning module and a motion control module;

the target detection module is used for identifying a target and predicting the position of the target according to historical information;

the obstacle detection module is used for identifying obstacles and predicting probability distribution of obstacle positions according to historical information;

the motion planning module is used for respectively constructing an attraction potential field and a repulsion potential field according to the predicted target position and the probability distribution of the predicted obstacle position, and the attraction potential field and the repulsion potential field jointly act to generate a safety path of the intelligent wheelchair;

the motion control module is used for tracking the safety path generated by the motion planning module and executing the motion instruction of the intelligent wheelchair.

Images