Marker-less motion capture systems can provide online recordings of human biomechanics during rapid dynamic exercises such as countermovement jump (CMJ) which could indicate an athlete's risk of injury to the anterior cruciate ligament (ACL). However, without additional post-processing the localisation accuracy of the joints can be insufficient. Subsequently, biomechanics measurements, e.g. knee flexion angles, can be severely corrupted. We propose a calibration algorithm to correct for deviations in the bone length during CMJ as recorded by a low cost marker-less motion capture system (i.e. Kinect, version 2). Results were compared to gold standard VICON measurements. In this single subject study of three CMJs the accuracy of the measured knee flexion angle during stabilisation (post jump) was significantly improved from -9.6 degrees to -3.8 degrees (p<0.05) for the left knee, and from -5.0 degrees to 1.7 degrees (p<0.05) for the right knee. In conclusion, bone-length calibration and correction may enhance the joint localisation accuracy for low cost marker-less motion capture to the extend where clinically-relevant decisions can be facilitated.