Object posture recognition method

Abstract

The invention relates to an object posture recognition method, and aims to improve the recognition accuracy and robustness in a complex environment and meet the real-time requirement at the same time. According to the method, a binocular stereo camera is used for obtaining an image, through the steps of gray-scale map conversion, region growth, threshold segmentation, morphological operation and the like, target features are extracted, and the position of the target features in an image coordinate system is determined. According to the algorithm, a target area is separated by using the similarity between pixels, and the adaptive illumination change is set through an adaptive threshold, so that the detection accuracy is improved. The method is simple in operation, small in calculation amount and suitable for real-time processing, has self-adaptive parameters and a dynamic updating mechanism, and enhances the adaptability to various environment changes. Each frame of image is processed and corrected independently, error accumulation is avoided, and the overall precision is improved. The method can effectively recognize object postures, and is suitable for the field of industrial automation and intelligent manufacturing.

Description

Technical Field

[0001] The invention relates to an object posture recognition method, belonging to the technical field of posture correction. Background Art

[0002] With the development of industrial automation and intelligent manufacturing, object posture recognition technology has been widely used in production lines, robot navigation, intelligent sorting and other fields. Accurately identifying the posture and position of an object is the key to achieving high-precision operation and automated control. At present, object posture recognition methods based on machine vision mainly include template matching, feature extraction and 3D reconstruction.

[0003] Traditional template matching methods identify the position and posture of objects by pre-building a template of the target object and matching it with the acquired image. This method is sensitive to changes in the shape and size of the object, and the matching effect will be significantly reduced under lighting changes, partial occlusion and complex backgrounds, affecting the accuracy of recognition.

[0004] Feature extraction methods use feature points, edges, and texture information on the surface of an object to extract and match these features through algorithms to determine the object's posture. However, in practical applications, noise, complex backgrounds, and lighting changes can interfere with accurate feature extraction and reduce the reliability of recognition.

[0005] The 3D reconstruction method is based on the principle of stereoscopic vision. It reconstructs the 3D model of an object through multi-view images to obtain its posture information. Although this method can achieve high accuracy, it has high computational complexity and high hardware performance requirements, making it difficult to meet real-time requirements.

[0006] In the prior art, the object posture recognition method mainly has the following problems:

[0007] Lack of robustness: Weak ability to resist interference from lighting changes, complex backgrounds and noise, resulting in low recognition accuracy.

[0008] Poor real-time performance: Complex algorithms and high computational complexity make it difficult to meet the requirements of real-time detection and processing, limiting the scope of application.

[0009] Poor adaptability: Parameters need to be adjusted for specific objects or scenes, lack of adaptive capabilities, and difficulty in adapting to changing environments.

[0010] Therefore, there is an urgent need for a new object posture recognition method that can improve the accuracy and robustness of recognition in complex environments, while meeting real-time requirements and enhancing the adaptability to diverse application scenarios to promote the development of industrial automation and intelligent manufacturing. Summary of the invention

[0011] The technical problem to be solved by the present invention is to provide an object posture recognition method to overcome the deficiencies of the prior art.

[0012] The technical solution of the present invention is: a method for object posture recognition, the method comprising the following steps:

[0013] 1) Use a binocular stereo camera to obtain an image of the object;

[0014] 2) Convert the acquired color image into a grayscale image;

[0015] 3) In the first frame image, a feature point in the target area is selected as the seed point. According to the gray value of the seed point, the growth condition is set, and the pixel points with a gray value close to the seed point are added to the seed area. The average gray value Ga of the target feature area after growth is calculated as a reference for image segmentation. A preliminary circumscribed rectangle is generated according to the growth area to determine the position of the target in the image coordinate system;

[0016] 4) Using the calculated average gray value Ga to perform threshold segmentation on the entire image;

[0017] 5) Perform top hat operation, opening operation and closing operation on the segmented image;

[0018] 6) For the optimized image, extract the contours of each connected area, generate its circumscribed rectangle, calculate the area and aspect ratio of each circumscribed rectangle, and select the circumscribed rectangle with the largest area and the largest aspect ratio as the final target rectangle;

[0019] 7) Use the minimum bounding rectangle algorithm to obtain the center point coordinates, width, height and rotation angle of the target rectangle;

[0020] 8) Update the average gray value Ga calculated for the current frame as a reference for the next frame image processing;

[0021] 9) Repeat steps 1) to 8) to process the images of subsequent frames to achieve continuous detection and correction of the target object's posture.

[0022] The beneficial effects of the present invention are: compared with the prior art,

[0023] 1) The region growing algorithm of the present invention utilizes the similarity between pixels and can effectively separate the target area from the noise and complex background. By calculating the average gray value Ga and setting the adaptive threshold, it can adapt to a certain degree of illumination change and improve the detection accuracy of the target object. The average gray value Ga is calculated based on the image features of the current frame and can adapt to different illumination conditions and grayscale changes of the target object. Through threshold segmentation, the background and interfering objects with large differences in grayscale values ​​from the target can be effectively filtered out. By screening the area and aspect ratio, non-target areas can be effectively excluded and the target object can be accurately located. Even in the presence of multiple interfering objects, the algorithm can still identify the area that best meets the target characteristics. Therefore, the method has enhanced anti-interference capabilities against noise, complex backgrounds and illumination changes, improved recognition accuracy, and improved robustness.

[0024] 2) The method of the present invention mainly utilizes grayscale image processing, region growing, threshold segmentation and morphological operations. Compared with complex three-dimensional reconstruction and feature matching algorithms, the amount of calculation is greatly reduced. Ga is updated in each frame, which avoids complex parameter adjustment and reduces calculation time. The algorithm steps are executed in sequence, and each step provides necessary information for the next step, reducing redundant data processing. By selecting seed points in the first frame and updating Ga in subsequent frames, the algorithm utilizes the information of the previous frame and reduces the amount of calculation. Therefore, the algorithm design of this method is efficient, the amount of calculation is small, and it is suitable for real-time processing;

[0025] 3) The present invention uses the average gray value Ga calculated in the current frame for threshold segmentation, which can adapt to changes in illumination and grayscale of objects. Ga is updated in each frame, so that the algorithm can dynamically adapt to environmental changes without manual intervention. The adaptive parameters and dynamic update mechanism of the present invention enable the algorithm to adapt to a variety of environmental changes, thereby enhancing adaptability;

[0026] 4) Since each frame of the present invention is processed and corrected independently, the accumulation of errors is avoided and the overall accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a flow chart of the algorithm for judging whether a cigarette rod is centered according to the present invention. DETAILED DESCRIPTION

[0028] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0029] Implementation Example 1:

[0030] refer to Figure 1 , a method for object posture recognition, the method comprising the following steps:

[0031] 1) Use a binocular stereo camera to obtain an image of the object;

[0032] 2) Convert the acquired color image into a grayscale image;

[0033] 3) In the first frame image, a feature point in the target area is selected as the seed point. According to the gray value of the seed point, the growth condition is set, and the pixel points with a gray value close to the seed point are added to the seed area. The average gray value Ga of the target feature area after growth is calculated as a reference for image segmentation. A preliminary circumscribed rectangle is generated according to the growth area to determine the position of the target in the image coordinate system;

[0034] 4) Using the calculated average gray value Ga to perform threshold segmentation on the entire image;

[0035] 5) Perform top hat operation, opening operation and closing operation on the segmented image;

[0036] 6) For the optimized image, extract the contours of each connected area, generate its circumscribed rectangle, calculate the area and aspect ratio of each circumscribed rectangle, and select the circumscribed rectangle with the largest area and the largest aspect ratio as the final target rectangle;

[0037] 7) Use the minimum bounding rectangle algorithm to obtain the center point coordinates, width, height and rotation angle of the target rectangle;

[0038] 8) Update the average gray value Ga calculated for the current frame as a reference for the next frame image processing;

[0039] 9) Repeat steps 1) to 8) to process the images of subsequent frames to achieve continuous detection and correction of the target object's posture.

[0040] The image information of the cigarette strips is obtained by using a binocular stereo camera. In order to extract the target, a region growing algorithm is designed to realize the feature recognition of the attached target. The basic idea of ​​the region growing algorithm is to start from a single pixel point and merge pixels with similar properties to form the same region according to the algorithm growth rule. The specific process is to first select a target point in the image as a seed point, then add the seed region according to the pixel points in the image that are similar to the seed region and meet the growth rule conditions, and finally generate a seed region with similar properties to the seed point.

[0041] The cigarette feature detection process is designed based on the region growing algorithm. First, a feature point in the target area is selected in the first acquired target image, and region growth is performed based on the image features of the point. The average grayscale value of the target feature area is calculated as a reference value for image segmentation and a preliminary bounding rectangle is generated to determine the coordinate position of the target in the image coordinates. In order to improve the detection accuracy of the target, the entire image is feature screened according to the average grayscale value to complete the image segmentation. Optimization processing is performed based on the image features, and top hat operations, open operations, and closed operations are used to improve features such as the contour and connected areas of the processed area. A bounding rectangle is then generated based on the processed features. The area and aspect ratio of all bounding rectangles are screened out to select the rectangle with the largest area and aspect ratio as the target rectangle. Finally, the center coordinate position of the bounding rectangle in the image coordinate system is fed back to the onboard computer for control processing, and the average grayscale value of the target feature area is updated at the same time as the reference value for the next frame of image processing. The overall process of target detection is as follows: Figure 1 shown.

[0042] In the process, the image is first converted from a color image to a grayscale image, and then region growing and bounding rectangle operations are performed to obtain information such as the average pixel grayscale and the bounding rectangle position. The target position information is obtained by performing operations in the pixel coordinate system, satisfying the following formula:

[0043]

[0044] Where M1, M2, N1, and N2 represent the extreme values ​​of the horizontal and vertical coordinates of the coordinate system, roi.x represents the horizontal coordinate of the upper left corner of the target circumscribed rectangle in the pixel coordinate system, roi.y represents the vertical coordinate, roi.w represents the width of the circumscribed rectangle, roi.h represents the height of the circumscribed rectangle, Cw and Ch represent the set width and height thresholds, g(x,y) represents the pixel value after segmentation, Ga represents the target average pixel grayscale, T1 represents the adaptive adjustment threshold, and f(x,y) represents the grayscale value corresponding to the pixel point (x,y) coordinates.

[0045] In the optimization process, the interference features can be filtered by morphological top-hat operation. After the top-hat operation, the image features are basically horizontal features. The opening operation is used to erode first and then expand. The rectangles with feature width far greater than the height are retained, and the slender horizontal elements in the image features are filtered. Then the image is closed, expanded first and then eroded, and the target area features are filled to make the target area smooth and complete. Finally, the area and aspect ratio of the circumscribed rectangle are screened through the area contour after optimization. The target area P is selected as:

[0046] P = max(rectg i ×ratio i )

[0047] In the formula, rectgi represents the area of ​​the circumscribed rectangle, and ratioi represents the ratio of the width to the height of the circumscribed rectangle.

[0048] Then use the minimum enclosing rectangle method, which will return the left side of the center point of the minimum enclosing rectangle, the rectangle width, height, and rotation angle, to calculate the rotation angle of the cigarette.

[0049] The target area and coordinate position are obtained through visual processing, and the target coordinate information is transmitted to the control system through the communication system, providing a data basis for the next step of posture correction.

[0050] Knowing the rotation angle and center position of the cigarette strip under the coordinate axis, the offset between the cigarette strip and the conveyor belt can be calculated in the same coordinate system. After the offset is calculated, it is fed back to the control system. The control system adjusts the rotation angle of the servo to control the moving distance of the center baffle, and the pressure sensor on the center baffle is used as feedback to determine whether the centering is accurate. Prepare for the next step of attitude correction.

[0051] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for object posture recognition, characterized in that: The method comprises the following steps: 1) Use a binocular stereo camera to obtain an image of the object; 2) Convert the acquired color image into a grayscale image; 3) In the first frame image, a feature point in the target area is selected as the seed point. According to the gray value of the seed point, the growth condition is set, and the pixel points with a gray value close to the seed point are added to the seed area. The average gray value Ga of the target feature area after growth is calculated as a reference for image segmentation. A preliminary circumscribed rectangle is generated according to the growth area to determine the position of the target in the image coordinate system; 4) Using the calculated average gray value Ga to perform threshold segmentation on the entire image; 5) Perform top hat operation, opening operation and closing operation on the segmented image; 6) For the optimized image, extract the contours of each connected area, generate its circumscribed rectangle, calculate the area and aspect ratio of each circumscribed rectangle, and select the circumscribed rectangle with the largest area and the largest aspect ratio as the final target rectangle; 7) Use the minimum bounding rectangle algorithm to obtain the center point coordinates, width, height and rotation angle of the target rectangle; 8) Update the average gray value Ga calculated for the current frame as a reference for the next frame image processing; 9) Repeat steps 1) to 8) to process the images of subsequent frames to achieve continuous detection and correction of the target object's posture.

Images