Introduction Video-assisted thoracoscopic surgery (VATS) is a minimally invasive technique that uses camera guidance to visualize the lungs in real time through small incisions, thereby reducing risks and enabling accurate diagnosis and intervention. High-quality imaging is essential for early detection of lung cancer and effective treatment. In order to better perceive the different surfaces of organs and surgical tools, this visualization can be performed stereoscopically using a 3D endoscope. However, surface reconstruction requires extremely accurate camera calibration. Camera calibration is a widely studied field in computer vision, with reference techniques such as Zhang Zhengyou's method [1] based on images of a calibration pattern (usually a grid or chessboard), now implemented in well-established libraries such as OpenCV. Endoscopic images are quite specific due to their wide angle and generally narrow and restricted fields of view. The choice of a calibration pattern adapted to these optical specificities, and giving good results in terms of precision, ease of use and robustness is essential for a future good reconstruction. In this work, we have studied and evaluated several calibration patterns in order to find the calibration object best suited to our problem. Method In [2], a camera calibration module with OpenCV is proposed based on Zhang's method, using a chessboard pattern as a reference object. Although reliable, chessboard patterns have limitations in angle detection, impacting accuracy. Various approaches such as combining chessboard patterns with additional processing techniques (distortion models, sub-pixel corner detection, etc.) have been proposed. However, the use of particular patterns such as ChAruco grids, which combine the characteristics of Chessboard and ArUco markers, has not yet been proposed for endoscopic vision for medical purposes. ArUco markers are binary square markers with an inner binary matrix that determines its identifier. By combining the two, the ArUco part is used to interpolate the position of chessboard corners, giving it the versatility of markers, since it allows occlusions or partial views. Moreover, as the interpolated corners belong to a chessboard, they are highly accurate in terms of subpixel precision. Figure 1 illustrates three ChArUco boards of the same size but with different numbers of elements. Figure 1: 8x8 ; 4x4 and 3x3 ChArUco grids. Therefore, to achieve accurate recognition of the ChAruco board, the algorithm was divided into three distinct stages: (1) Image capture, (2) Recognition of the ChAruco and (3) stereo camera calibration using OpenCV. Once the cameras were calibrated, we were able to use the classic feature extraction and reconstruction methods. Results To validate the algorithm, two Logitech C270 HD WEBCAM cameras were used to simulate endoscopic views with specific calibration baselines. Three grid configurations (8x8 ; 4x4 and 3x3) were tested by recording one-minute videos and analyzing them with the algorithm for recognition and calibration. To estimate the accuracy of calibration and reconstruction, we placed a grid at a distance z of 50 cm from the cameras. The 3D location of the grid points was obtained by triangulation using DLT (Direct Linear Transformation). The accuracy was measured by comparing the z-distance estimated by reconstruction of the various grid points with the actual value. The evaluation showed that ChAruco grids outperformed the classical grids and, in our case, the 4x4 and 3x3 ChAruco grids gave the best performance in terms of calibration parameter estimation accuracy and 3D reconstruction acuracy. This result can be attributed to external conditions such as ligthing, brightness, and grid positioning that affect detection accuracy. Initially, we thought that the 8x8 grid would perform better because it had more features than the other two. One of the hypotheses to explain this underperformance is that the features of the 8x8 grid are too small to ensure good precision and that the 4x4 or 3x3 grids have the right size characteristics to ensure accurate and robust calibration for our cameras