@InProceedings{OgataCVPRW2019,
                            author    = {Ryoji Ogata and Edgar Simo-Serra and Satoshi Iizuka and Hiroshi Ishikawa},
                            title     = {Temporal Distance Matrices for Squat Classification},
                            booktitle = "Proceedings of the Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)",
                            year      = 2019,
                        }
                      

                          When working out, it is necessary to perform the same action many times for it to have effect. If the
                        action, such as squats or bench pressing, is performed with poor form, it can lead to serious injuries in the
                        long term. With the prevention of such harm in mind, we present an action dataset of videos where different
                        types of poor form are annotated for a diversity of subjects and backgrounds, and propose a model for the
                        form-classification task based on temporal distance matrices, both in the case of squats. We first run a 3D
                        pose detector, then normalize the pose and compute the distance matrix, in which each element represents
                        the normalized distance between two joints. This representation is invariant under global translation and
                        rotation, as well as robust to individual differences, allowing for better generalization to real world data.
                        Our classification model consists of a CNN with 1D convolutions. Results show that our method significantly
                        outperforms existing approaches for the task.