Pose estimation is a computer vision technique that predicts and tracks different parts of a person or object. This is done by looking at a combination of the pose and the orientation of a given person/object.
In this project, we implemented a PoseNet Network on TensorFlow 2.1 and trained it on COCO dataset. We optimized the model using TensorFlow Lite and TensorRT to enable it to run on mobile phones and edge devices. The model has been used to detect a person’s different actions with surveillance cameras as well as cropping different parts of a body to be used in upstream tasks such as face recognition.