Abstract: This paper proposes signal processing technology and detection methods to reliably identify and grasp workpieces in robot assembly operations. It studies feature extraction methods that accurately describe object shapes and designs a robot automatic identification and grasping system based on vision and ultrasonic technologies. Experimental research on object recognition and grasping was conducted on a robot assembly platform. Keywords: robot; vision; ultrasonic technology; image recognition. Vision sensors can intuitively reflect the external information of objects, but a single camera can only obtain two-dimensional images. While stereo vision can provide three-dimensional information, it is difficult to identify objects with the same shape but different depths (such as objects with holes or stepped objects), and it has certain requirements regarding ambient light. Since ultrasonic sensors are insensitive to light and material, have a simple structure, and can directly obtain the distance from the test point to the sensor, this paper adopts a method combining vision and ultrasonic measurement. Two-dimensional image information is fused with depth information obtained by ultrasonic sensors to automatically identify and spatially locate the workpiece to be assembled, and to determine the spatial position and attitude of the robot's end effector, enabling it to accurately grasp the workpiece at the appropriate location. 1 System Principle and Structure The system consists of a robot arm, a CCD... The system consists of a vision sensor, an ultrasonic sensor, and corresponding signal processing units. A CCD is mounted on the robot's end effector, forming a hand-eye vision system. The receiving and transmitting probes of the ultrasonic sensor are also fixed to the robot's end effector. The CCD acquires a two-dimensional image of the object to be identified and grasped, guiding the ultrasonic sensor to acquire depth information. The system structure is shown in Figure 1. Image processing mainly completes the accurate description of the object's shape, including the following steps: a. Image edge extraction; b. Perimeter tracking; c. Feature point extraction; d. Curve segmentation and segment matching; e. Graphic description and recognition. After extracting the object image edges, perimeter tracking is used for edge refinement, removing false edge points and noise points. Freeman encoding is performed on the edge points forming the closed curve, recording the direction of each chain code and the XY coordinates of each point on the curve, facilitating further analysis of the object's geometric characteristics. This study improves the search direction and order of edge points in the traditional perimeter tracking algorithm and adopts a method to eliminate redundant points in a timely manner during the search process, reducing the amount of data and computation time, and achieving better noise reduction and smoothing effects. When extracting image feature points, the polygon approximation method is combined with the curvature calculation method to overcome the disadvantages of the polygon approximation method being prone to producing false feature points and the curvature calculation method having excessive computational load. After the object image obtained by CCD is processed, certain features of the object can be extracted, such as the centroid coordinates, area, curvature, edge, corner point and minor axis direction of the object. Based on these feature information, a basic description of the object shape can be obtained. On the basis of image processing, the depth of the test point is measured by the visual information to obtain the depth (height) information of the object, or it moves along the test surface of the workpiece. The ultrasonic sensor continuously collects distance information, scans to obtain the distance curve, and analyzes the edge or shape of the workpiece based on the distance curve [1]. After the computer fuses and infers the visual information and depth information, it performs image matching and recognition, and controls the robot arm to accurately grasp the object with a suitable posture. 2.1 Extraction of workpiece image edge Complex workpieces are often reflected in the image with more than one gray level. It is impossible to extract meaningful edges using only one gray level threshold. If a multi-threshold method is used, it will inevitably increase the computation time and the complexity of image processing. For the automatic thresholding method based on categorical variance, increasing the threshold value not only increases the complexity of data processing, but also affects the reliability of the algorithm when there are more than two thresholds. Therefore, a method of directly extracting edges from grayscale images is adopted. Image edges generally occur at discontinuities in grayscale function values ​​and can be obtained using the first or second derivative of the grayscale function. Classic methods for edge extraction using the first derivative include the Roberts operator and the Sobel operator, while methods using the second derivative include the Laplacian operator and the Marrs-Hilderth operator. Through analysis and comparison of several algorithms, the Sobel operator is considered to be not only easy to implement and fast in computation, but also provides the most accurate edge direction estimation. The Sobel operator consists of two 3 × 3 operators separated by 90°. Convolving these two operators with the image yields the image edges and their directions. For a digital image {f(i, j)}, the Sobel operator can be expressed as: Gx(i, j) = f(i - 1) Gy(i, j) = f(i-1, j-1) + 2f(i, j-1) + f(i-1, j+1) - f(i+1, j-1) - 2f(i+1, j) - f(i+1, j+1); Gy(i, j) = f(i-1, j-1) + 2f(i, j-1) + f(i+1, j-1) - f(i-1, j+1) - 2f(i, j+1) - f(i+1, j+1). After obtaining the gradient magnitude using G1 = |Gx| + |Gy|, to reduce the number of extracted edges, an amplitude threshold can be set, that is, only those edges with large corresponding gray-level changes are considered. Then, taking advantage of the characteristic that edge points have the largest local amplitude, the edges are refined. Using Sobel After the operator extracts the edge, in order to obtain the size information of the workpiece surface, it is also necessary to extract the corner points of the image [2] so as to calculate the feature information such as the side length of the workpiece. 2.2 Determination of centroid coordinates The centroid point in the image can usually be obtained by two methods: one is to calculate the centroid coordinates by the method of calculating moments through region processing; the other is to calculate it by edge chain code integration. This algorithm is relatively simple and applicable to any image, but it needs to be combined with a pixel-based region segmentation algorithm. 2.3 Determining the Axis: To enable the robotic arm to accurately grasp the object with the correct posture, the object's axis must be precisely determined. In geometry, the major axis of an object is defined as a straight line passing through the object's centroid, and the object's second moment about this line is at its minimum. Let θ be the angle between the object's major axis and the positive X-axis of the image plane, and let |θ| ≤ π/2. Then, the object's second moment about this axis is... The second moment of the object about the axis is obvious. The method of determining the axis based on the second moment of inertia is to calculate the entire object region, and the membership region of each pixel must be determined first, so the computation is large. Figure 2 (a) is the workpiece axis determined by this algorithm. For some simple-shaped objects, the following simple axis estimation algorithm can be used: a. Determine the centroid coordinates of the object; b. Determine the point closest to the centroid of the object in the first half of the closed curve of the object's edge contour, and estimate the tangent direction of the point using the least squares method, and let the angle between it and the positive direction of the X-axis of the image plane be α1; c. Determine the corresponding tangent direction α2 in the second half of the curve using the same method; d. The object axis can be roughly estimated as θ = (α1 + α2) / 2. Figure 2 (b) is the workpiece axis diagram obtained by the simplified algorithm. This algorithm only processes the object's edge contour points, which greatly reduces the computation time. 3 Ultrasonic Depth Detection Due to CCD The image obtained by the camera cannot reflect the depth information of the workpiece. Therefore, for workpieces with the same two-dimensional graphics but only slightly different heights, visual information alone cannot be used to correctly identify them. This paper uses an ultrasonic ranging sensor to make up for this deficiency. After obtaining the edge, centroid and other features of the workpiece through image processing, the robot arm is guided to the test point to measure the depth of the workpiece. By fusing visual signals and ultrasonic signals, a more complete workpiece information can be obtained. The ultrasonic sensor installed on the robot end effector consists of transmitting and receiving probes. Based on the principle of sound wave reflection, the sound wave signal reflected back from the test point is detected and processed to obtain the depth information of the workpiece. In order to improve the detection accuracy, variable threshold detection, peak detection, temperature compensation and phase compensation are used in the receiving unit circuit [1] to obtain higher detection accuracy. For two cylindrical workpieces with the same shape and a height difference of 0.1 mm in the field of view, the method of fusing image and depth information proposed in this paper can be used to accurately identify and grasp them. 4 Experimental results and conclusions Based on the above method research, the object recognition and grasping experiment was completed on the MOVEMASTER2EX robot assembly platform. Under natural light and general lighting conditions, the robot assembly platform was used to automatically identify and grasp 3-5 typical workpieces of different shapes and sizes placed arbitrarily within its field of view. The results showed that the recognition time was less than 5 seconds (including the movement time of the robot arm during the recognition, positioning, and grasping process), the positioning error was less than ±2 mm, and it had good versatility and portability. Figures 3(a) to (d) are images of the workpiece recognition process. The experimental results show that the detection device proposed in this paper, which combines robot hand-eye vision with ultrasonic ranging, and the method of workpiece recognition and grasping by fusing two-dimensional image information and depth information, can accurately identify and locate objects. It has the characteristics of simple algorithm, low computational load, good real-time performance, and high reliability. It can provide information such as the shape, category, and size of objects for robot-environment interaction, enabling robot assembly operations to adapt to various complex environments and processes. It has good application prospects for realizing the automation, flexibility, and intelligence of industrial production processes. The experimental results show that the detection device proposed in this paper, which combines robot hand-eye vision with ultrasonic ranging, And the method of integrating two-dimensional image information and depth information for workpiece recognition and grasping can accurately identify and locate objects. It has the characteristics of simple algorithm, small amount of calculation, good real-time performance and high reliability. It can provide information such as object shape, category and size for robot interaction with the environment, so that robot assembly operation can adapt to various complex environments and processes. It has good application prospects for realizing the automation, flexibility and intelligence of industrial production process. References [1] Yang Jinsong, Wang Min, Huang Xinhan. Ultrasonic variable threshold ranging device. Electronic Technology Application, 1998, 24 (7): 7～9 [2] Zhang Xiaoli, Wang Min, Huang Xinhan. An effective corner detection method based on Freeman chain code. Journal of Electronic Measurement and Instrumentation, 1999, 13 (2): 14～19 [3] The C, Chin R. On Image Analysis by the Method of Moments. IEEE Trans. Pattern Anal. Mach. Intell., 1988 (10): 291～310