3D Pose Estimation with Rotation Data Using CNN and Sensors

3D pose estimation plays a vital role in many industries like, Medicine, Entertainment, Sports, etc. This usually costs a lot of money to efficiently capture the pose of the human, this includes the usage of high-precision motion capture suits and multiple cameras at different angles to calculate the position of each joint and process them. This involves a lot of computational resources and time to process the data. To minimize this work, we are proposing a CNN-based approach for 3D Pose estimation which can determine the Pose of the human with just a single RGB Image. For training this network, we are using our synthetic dataset made by rendering realistic 3D models and extracting the ground truth from it along with other Datasets. Unlike in other methods, we use rotation (X, Y, Z) of certain bones that are needed for movement instead of estimating the joint locations and we use Ultrasonic sensors for ground detection, which enables us to estimate the pose in the picture.