This paper suggests a new technique for 3D reconstruction that makes use of two cameras. The setup environment was limited by the need for many cameras in conventional 3D reconstruction techniques. Alternatively, decreased accuracy and problems with invisible areas resulted from utilizing fewer cameras. We apply domain knowledge of the human skeleton structure to our technique, using BlazePose as matching points for skeletal keypoint recognition. The suggested technique comprises of triangulation to reconstruct 3D coordinates and applying RANSAC for extrinsic parameter estimation. Despite severe parallax, this method produces 3D reconstruction with good accuracy. We employed motion capture data from athletes’ agility measurement moves for evaluation. In the studies, two cameras were recording at 960 Hz and ten cameras were recording at 250 Hz while the subjects wore markers. The average distance error of the evaluation, which was 42 mm on average, indicated accuracy appropriate for sports video analysis. Furthermore, examining the effect of frame rates revealed no appreciable variations in accuracy between 120 Hz and 960 Hz. This suggests that common electronics could be used for high-accuracy 3D reconstruction. With fewer cameras, this research achieves high-precision 3D reconstruction, indicating possible uses in sports science.