Objective Line-structured light three-dimensional (3D) measurement system is widely used in object measurement, target detection, and other fields. The laser stripe center extraction algorithm is the key technology in the system. The center of robust and uniform laser stripe is easily extracted. However, in many industrial environments, there are some uncertain poor reflection factors, such as dirt and rust on the surface of the measured object and the bad laser emitting quality, leading to the existence of low and uneven brightness areas in the laser stripe images. Compared with the robust region, the pixel value of the robust region is close to 255 and differs greatly from the background pixel value. However, the pixel value of these low and uneven brightness areas is random and uncertain. Thus, it is not easy to select the appropriate pixel threshold accurately using the fixed threshold method. If the threshold is not selected properly, it can easily lead to a bad extraction result. Many existing algorithms cannot easily determine the appropriate pixel threshold in these areas. They tend to cause a larger calculation error and disconnection of the laser stripe centers in these areas. Thus, it is essential to improve the self-adaptability of the center extraction algorithm to solve this problem and improve robustness. At the same time, the complexity of many existing algorithms is high, and redundant scanning for the image region without laser stripe decreases the speed of the algorithms. The laser stripe center calculation is only related to the laser stripe itself and its surrounding pixels.
Methods This study proposes an internal propulsion algorithm for laser stripe center extraction. First, according to the distribution characteristics of the laser stripe in the image, the algorithm uses an internal propulsion strategy to plan the search path. The search path moves forward or backward along the center of the stripe to reduce processing image regions without laser stripes; thereby, improving the computational speed of the laser stripe center. Besides, the 8-connectivity search is used to eliminate noise points when searching the starting point of internal propulsion. Second, the proposed internal propulsion algorithm is inspired by the mean shift algorithm, which is an unsupervised iterative clustering algorithm in the field of machine learning. This algorithm draws lessons from the idea of moving, updating, and ending of the mean shift algorithm. The internal propulsion algorithm also belongs to a clustering algorithm. The core process is as follows: calculate the initial center first, move forward or backward a pixel as prediction center, calculate the new threshold using the proposed adaptive threshold method, calculate the new center point, update the center point, and repeat until the end. Among them, the calculation of the center uses the geometric center method. The geometric center method extracts the laser stripe center using the geometric centroid of the upper and lower boundaries of the laser stripe cross-section. The stripe boundary is determined by the pixel threshold. The proposed algorithm uses the improved Otsu adaptive threshold method to update the threshold during internal propulsion continuously. The traditional Otsu method is a global threshold method. However, the global threshold method is not easy to consider the detailed segmentation, and it requires high calculation. In this study, the traditional Otsu method is improved to extract the laser stripe center. Let the global threshold become multi-threshold, and the maximum between-class variance is computed for each column pixel of the laser stripe within a limited length to determine the threshold so that each column of the laser stripe has an optimal center calculation threshold.
Results and Discussions The proposed algorithm uses an internal propulsion strategy to plan the search path and reduce processing image regions without laser stripes. Thus, it improves the computational speed of the laser stripe center (Table 3). The improved Otsu adaptive threshold method overcomes the inappropriateness of the global threshold to detailed segmentation and reduces computational effort. Besides, it makes the laser stripe center extraction more robust. Using the improved Otsu adaptive threshold method, the proposed algorithm significantly improves the robustness and accuracy of center extraction. It solves the problem of disconnection in the low and uneven brightness stripe areas (Fig.13). It also reduces center extraction error, especially in low and uneven brightness areas (Tables 1 and 2). Finally, the proposed algorithm has good anti-noise properties when adding noise to the experiment images (Table 4).
Conclusions This study proposes an internal propulsion algorithm for laser stripe center extraction. The experimental results show that the proposed algorithm has low complexity, fast running speed, good robustness, and high accuracy. The proposed algorithm achieves excellent anti-noise effect after adding noise points. In particular, the internal propulsion algorithm can reduce processing image regions without laser stripes and has a good extraction effect on the non-robust laser stripe with low and uneven brightness areas. Owing to these advantages, the proposed algorithm will be of great significance in many industrial applications, such as online product inspection and welding seam tracing, especially when fast speed and good robustness are required in real working conditions where the reflectivity of the object surface is complicated and bad.
