A Model-Based Pose Estimation Framework : From Single Insects to Multiple Hands

Abstract

Robust pose estimation for objects such as human hands and insects plays a central role in human-computer interaction, virtual reality, augmented reality and behavioral studies of animals such as bees. Estimating poses of insects and hands are challenging because they suffer from (self-)similarities, (self-)occlusions, a varying number of targets, etc. Moreover, multi-instance pose estimation is more difficult because it requires the correct detection of all instances while also precisely localizing the corresponding target landmarks.