CN102324099B - Step edge detection method oriented to humanoid robot - Google Patents

Abstract

The invention is a step edge detection method for a humanoid robot, comprising the following steps: 1) preprocessing the image to be detected; 2) judging the image; 3) performing morphological processing on the image; 4) processing the image to obtain The edge of the light step; 5) Calculate the threshold and convert the image into a binary image to obtain the edge of the backlight step; 6) Use the edge line in step 5 to segment the image in step 2, and divide the part with high gray value Perform grayscale stretching to obtain a new image; 7) According to the new image obtained in step 6, undergo gradient transformation to obtain a gradient image; 8) According to the new image obtained in step 6, perform inverse color transformation and then perform gradient transformation to obtain another Gradient image; 9) Add the corresponding pixels of the two gradient images obtained in step 7 and step 8 to obtain the edge of the image; 10) merge the edge in step 5 and the edge in step 9 into one image . The edge detection method of the invention meets the requirements of timeliness and accuracy in robot vision.

Description

A Step Edge Detection Method for Humanoid Robot

technical field

The invention belongs to the technical field of image detection and segmentation, and in particular relates to a step edge detection method for humanoid robots.

Background technique

Robot vision technology mainly studies the use of computers to simulate human visual functions to extract information from images of objective things, process and understand them, and finally use them in actual detection, measurement and control. It is the intersection and combination of multiple disciplines, and the research goals of many disciplines are similar to or related to robot vision technology. These disciplines include image processing, pattern recognition or image recognition, scene analysis, image understanding, etc. Image processing is the basis of visual technology. The processed image is an objective reflection of natural scenery and an important way to perceive the surrounding environment.

The shape characteristic in the image is an important indicator of feature extraction. In many cases, only by knowing the shape characteristic of the image can the image be further analyzed quantitatively, and the determination of the boundary contour (ie edge) is the premise of studying the shape characteristic. It is also the premise of computer automatic recognition of images, and it has a significant impact on high-level processing such as feature description, recognition and understanding.

The edge of the image is the most basic feature of the image, and the so-called edge refers to the collection of those pixels whose gray level has a step change or a roof change. Edges widely exist between objects and backgrounds, between objects and objects, and between primitives. Because the edges of images usually contain a lot of important information, edge detection has become an important part of image processing, and its detection algorithms have also been extensively studied.

In an image, edges have two properties, direction and magnitude. The gray scale along the edge changes gently, while the gray scale perpendicular to the edge changes sharply. Edge detection is the measurement and positioning of grayscale changes. The significance of grayscale changes can be measured by derivatives, that is, the function derivative can reflect the significance of image grayscale changes. Therefore, a basic idea of edge detection is to obtain the first derivative The local maximum value of , and the zero-crossing point of the second derivative are reflected. The idea of using the maximum gradient to extract edge points has produced many classic edge detection methods such as: Sobel algorithm, Log algorithm, Canny algorithm, etc. However, the ordinary algorithm will be affected by the objective environment in robot vision and will not play a good role, and will be limited when dealing with specific problems, such as step images collected in machine vision in different environments and lighting conditions There will be a big difference under the circumstances, and if one method is used, the desired effect may not be achieved. Moreover, if a more complex algorithm is uniformly adopted, the processing time will be increased, which cannot meet the real-time requirements of robot vision.

Contents of the invention

In view of the problems existing in the above-mentioned robot vision technology, the present invention uses the idea of separate processing to propose a step edge detection method for humanoid robots, that is, first judge the environment of image acquisition, and then process it separately according to different environmental conditions The method adopts a simple processing method when the image is not complex and adopts a segmentation processing method when the image is complex, which not only ensures the processing speed, but also ensures the processing accuracy, thus conforming to the timeliness and accuracy of machine vision. sexual demands.

Technical scheme of the present invention comprises the following steps:

Step 1: Preprocessing the step grayscale image to be detected;

Step 2: Judging the preprocessed image, if the grayscale image of the step is facing the light or side facing the light, then go to step 3, if the grayscale image of the step is facing away from the light, then go to step 5;

Step 3: Perform morphological processing on the step grayscale image, that is, perform expansion and erosion operations on the step grayscale image respectively, and then subtract the two images obtained after expansion and erosion to obtain an image with enhanced edge information, go to Step 4;

Step 4: use the sobel edge detection operator to process the step grayscale image to obtain the edge of the light step;

Step 5: directly use the Otsu method to obtain the threshold value and convert the image into a binary image to obtain the edge of the backlight step, go to step 6;

Step 6: Segment the image in step 2 with the edge line in step 5, stretch the part with high gray value to get a new image and go to step 7;

Step 7: Obtain a gradient image through gradient transformation according to the new image obtained in step 6;

Step 8: Perform gradient transformation according to the new image obtained in step 6 through inverse color transformation to obtain another gradient image;

Step 9: Add the corresponding pixels of the two gradient images obtained in step 7 and step 8 to obtain a new gradient image, and then obtain the edge of the image from the new gradient image.

Step 10: Merge the edge in step 5 with the edge in step 9 in one image to get the edge of the backlit steps.

The preprocessing in the above step 1 adopts median filtering to remove the noise in the image and preserve the edge information in the image at the same time. The preprocessing uses a separable binary median filter to process the image. the

The method for judging the preprocessed image in said step 2 is to judge according to the variance and probability statistics of the pixel gray value in the image, first calculate the variance of the gray value of the image pixel, when the variance is greater than When it is 604-608, it is judged to be a backlit step, otherwise it is a light-oriented and photometric step.

In the step 3, select a 3x3 sub-image template, each position in the template takes the same gray value, and then use this template to traverse each pixel in the image from left to right and from top to bottom and perform morphological operations .

In the step 4, the threshold of the sobel algorithm adopts an iterative threshold selection method.

The preprocessing of the step grayscale image to be detected in the step 1 includes filtering and denoising processing.

The present invention has different step images under different lighting conditions, thereby adopting a better and more suitable edge detection method. Timeliness and accuracy requirements in robot vision.

Description of drawings

Fig. 1 is the flowchart of detection method of the present invention;

Fig. 2 is a diagram of structural elements of an embodiment of the present invention;

Figure 3 is a template diagram of the Roberts operator. the

Detailed ways

A step edge detection method for humanoid robots, comprising the following steps:

Step 1: Preprocessing the step grayscale image to be detected, the preprocessing uses median filtering to remove the noise in the image, and at the same time preserves the edge information in the image, the preprocessing uses separable binary median filtering to Process images.

Step 2: Judging the preprocessed image, if the step grayscale image is facing the light, go to step 3, if the step grayscale image is facing away from the light, go to step 5; for the preprocessed image The method of judgment is based on the variance and probability statistics of the gray value of the pixel in the image. First, calculate the variance of the gray value of the pixel in the image. When the variance is greater than 604-608, it is judged to be a backlight step , otherwise it is phototropic and photometric steps. This step is a key step of the present invention, and the methods for image processing of different properties are different, so the images must be discriminated and classified first. The specific method is to judge according to the variance and probability statistics of the gray value of the pixel in the image. First, calculate the variance of the gray value of the pixel in the image, because the steps of the backlight will produce shadows, resulting in the comparison of the difference in the gray value of the image pixel. When the variance is large, the variance will be relatively large. When the variance is greater than a certain value, it can be judged as a backlit step. Otherwise, it is a step toward the light.

Step 3: Perform morphological processing on the step grayscale image, that is, perform expansion and erosion operations on the step grayscale image respectively, and then subtract the two images obtained after expansion and erosion to obtain an image with enhanced edge information, go to Step 4; this step is specifically: select a structural element, as shown in Figure 1, that is, a 3x3 sub-image template, each position in the template takes the same gray value, and then use this template from left to right, from top to bottom Go down to traverse each pixel in the image and perform morphological operations. Specifically, for a pixel whose coordinates are (x, y) in the image, in the expansion process, it is necessary to add the gray value of the corresponding pixel in the structural element to each pixel within the template range, and then find the maximum gray value in the template And as the gray value of the pixel (x, y). In the corrosion process, the gray value of the corresponding pixel in the structural element is subtracted from each pixel in the template range, and then the minimum gray value in the template is found and used as the gray value of the pixel (x, y). In this way, two different images will be obtained from the original image, and then the pixels at the corresponding positions of the two images will be subtracted to obtain an image with enhanced edge information. the

Step 4: use the sobel edge detection operator to process the step grayscale image to obtain the edge of the light step; the threshold of the sobel algorithm adopts an iterative threshold selection method, specifically: select a threshold as the initial value at the beginning, and then use a The policy keeps improving this estimate until a given criterion is met. The strategy should have two characteristics: ① fast convergence; ② in each iteration process, the newly generated threshold is better than the last threshold.

Step 5: directly use the Otsu method to obtain the threshold value and convert the image into a binary image to obtain the edge of the backlight step, and then go to step 6; the Otsu method is a method for automatically determining the threshold value that maximizes the variance between classes, because This method is used because there is sufficient contrast between the shadows produced by the backlit steps and the steps.

Step 6: Segment the image in step 2 with the edge line in step 5, and perform gray scale stretching on the part with high gray value to obtain a new image Go to step 7; the gray scale stretching is to enhance the shadow Part of the contrast, the dynamic range of the gray value can be increased by improving the method used in histogram equalization, so as to achieve the effect of enhancing the overall contrast of the image.

Step 7: According to the new image obtained in step 6, the gradient image is obtained through gradient transformation;

Step 8: According to the new image obtained in step 6, undergo inverse color transformation and perform gradient transformation to obtain another gradient image;

The gradient operators used in Step 7 and Step 8 are both Roberts operators, as shown in Figure 2. The color inversion processing in step 8 is to invert the image whose gray scale range is [0, L].

Step 9: Add the corresponding pixels of the two gradient images obtained in step 7 and step 8 to obtain a new gradient image and then obtain the edge of the image from the new gradient image, and add the two gradient images in this step The essence is to intersect the edge of the original image with the edge detected after inversion, that is, the edge point is determined as the edge point when both detections are edge points, and some noise effects are eliminated in this process.

Step 10: Merge the edge in step 5 with the edge in step 9 in one image to get the edge of the backlit steps.

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

1. Image preprocessing. In order to preserve the edge information in the image, use the straight filter to remove the noise in the image. At the same time, in order to reduce the processing time, a separable two-dimensional median filter is used for operation. The specific steps are as follows:

Determine the size of a one-dimensional median filtering template, such as 1 x 3, and perform median filtering on each line in the image along the horizontal direction, that is, sort the pixels in the template according to the gray value, and select the gray value in the middle as a result. Perform median filtering on each column of the image along the vertical direction, the method is the same as above.

即可得到图像灰度值的方差，当

时，可判定台阶为背光的；否则为测光或正光的。其中，

为图像灰度值的期望，P为一先验值，在本次实验中若灰度值得平均值在180~200时，先验值P为24.6。 2. The pixel gray value of the preprocessed image is used as sample data, according to the following formula

The variance of the gray value of the image can be obtained, when

, it can be determined that the steps are backlit; otherwise, they are photometric or positive. in,

is the expectation of the gray value of the image, and P is a prior value. In this experiment, if the average value of the gray value is between 180 and 200, the prior value P is 24.6.

3. Morphological processing operations - dilation and erosion - are applied to the preprocessed image and the results are subtracted. First select the structural element—the 3x3 sub-image template. The gray scale of each position in the structural element is the same, which is recorded as b(x, y).

，定义为： The grayscale expansion of the image I with the structural element b(x, y) is expressed as:

,defined as:

，定义为： The grayscale erosion of the image I with the structuring element b(x, y) is expressed as:

,defined as:

和

分别是图像I和结构元素

的定义域。 in

and

are the image I and the structural element

domain of definition.

After the expansion and erosion operations are performed on the image according to the above formula, the two images are subtracted to obtain an image with enhanced edge information.

，如果

则认为该点是边缘点。 4. Use the sobel operator template to operate the image to calculate the gradient value of each pixel

,if

The point is considered to be an edge point.

The specific steps for determining the threshold T are as follows:

。 (1) Select the median of the grayscale of the gradient image as the initial threshold

.

和

，用下式计算区域

和

的灰度均值

和： (2) Use the threshold T to divide the gradient image into two regions——

and

, use the following formula to calculate the area

and

gray value of

and :

，

,

(3) Calculated and After that, the new threshold is calculated by the following formula :

 

 

和

的差小于某个给定的值。 (4) Repeat steps 2~3 until

and

The difference is less than a given value.

5. The specific steps of the Otsu method are:

1): Calculate the probability of each gray level pixel appearing:

，      i=0，1，2，…，L-1

, i=0,1,2,…,L-1

Where N is the number of pixels in the image, and [0, L-1] is the grayscale range.

2): according to the following formula:

 ，

，

，

。 in

,

,

,

.

最大的T值就是最佳的阈值。 Let T take values sequentially in the range [0, L-1], so that

The largest T value is the best threshold.

6. Specific steps of grayscale stretching:

其中

为灰度级为k的像素的个数，n像素的总个数。 1): First calculate the probability of each gray level:

in

is the number of pixels with a gray level of k, and the total number of n pixels.

2): Then use to calculate the transformation function value:

, k=0, 1, 2,...,L-1

将

扩展到[0，L-1]范围内并取整得到不同灰度级的变换函数然后将相同值归并起来，就得到修正后的灰度级变换函数 3) Using the formula

Will

Extend to the range of [0, L-1] and round to obtain the transformation functions of different gray levels, and then merge the same values to get the modified gray level transformation function

4) Add the number of pixels of the corresponding original gray level to get the number of pixels of the new gray level, that is, get a new image with higher contrast.

7. Use the 2x2 Roberts operator template to operate on the image to obtain a gradient map.

。s为原图像的灰度值，t为变换后图像的灰度值。 8. Inverting the image is to transform the grayscale from [0, L-1] to [L-1, 0]. The formula can be used

. s is the gray value of the original image, and t is the gray value of the transformed image.

9. The specific method is to perform a bit-AND operation on the detected edge pixels.

10. The edge of the entire image can be obtained by displaying the pixels of the edge points of the two images in one image. the

Claims (7)

1. a step edge detection method for humanoid robot, it is characterized in that comprising the following steps: Step 1: Preprocessing the step grayscale image to be detected; Step 2: Judging the preprocessed image, if the grayscale image of the step is facing the light or side facing the light, then go to step 3, if the grayscale image of the step is facing away from the light, then go to step 5; Step 3: Perform morphological processing on the step grayscale image, that is, perform expansion and erosion operations on the step grayscale image respectively, and then subtract the two images obtained after expansion and erosion to obtain an image with enhanced edge information, go to Step 4; Step 4: Process the step grayscale image with the sobel edge detection operator to obtain the edge of the light step, and the method ends; Step 5: directly use the Otsu method to obtain the threshold value and convert the preprocessed image into a binary image to obtain the edge of the backlight step, go to step 6; Step 6: Segment the image in step 2 with the edge line in step 5, stretch the part with high gray value to get a new image and go to step 7; Step 7: Obtain a gradient image through gradient transformation according to the new image obtained in step 6; Step 8: According to the new image obtained in step 6, undergo inverse color transformation and then perform gradient transformation to obtain another gradient image; Step 9: Add the corresponding pixels of the two gradient images obtained in step 7 and step 8 to obtain a new gradient image, and then obtain the edge of the image from the new gradient image; Step 10: Merge the edge in step 5 with the edge in step 9 in one image to get the edge of the backlit steps. 2. The step edge detection method for humanoid robots according to claim 1, characterized in that: the preprocessing in the step 1 adopts median filtering to remove the noise in the image, while preserving the edge information in the image. 3. The step edge detection method oriented to humanoid robot according to claim 2, characterized in that: the preprocessing in the step 1 adopts separable binary median filter to process the image. 4. the step edge detection method facing humanoid robot according to claim 1, it is characterized in that: the method for judging the preprocessed image in the said step 2 is according to the variance of pixel point gray value in the image To judge with probability and statistics, first calculate the variance of the gray value of the image pixel. When the variance is greater than the variance threshold, it is judged to be a backlight step, otherwise it is a step towards the light and side light, where the variance threshold is taken as 604-608. 5. the step edge detection method facing humanoid robot according to claim 1, is characterized in that: in described step 3, choose the sub-image template of a 3x3, each position gets identical gray scale value in the template, then Use this template to traverse each pixel in the image from left to right and top to bottom and perform morphological operations. 6. The step edge detection method for humanoid robots according to claim 1, characterized in that: in the step 4, the threshold of the sobel algorithm adopts an iterative threshold selection method. 7. The step edge detection method for humanoid robots according to claim 1, characterized in that: in step 1, the preprocessing of the step grayscale image to be detected includes filtering and denoising processing.

Images