Pose invariant people detection in point clouds for mobile robots

To be able to navigate in socially complaint fashion and safely, people detection is a very important ability for robots deployed in our social environments. However, it is a challenging task since humans exhibit various poses in daily life as they bend, sit down, touch or interact with each other. A robust people detector should detect humans also in these arbitrary poses. In addition, mobile robots should be able to carry out detection in a real-time manner because our environment is highly dynamic. In this study we developed a fast head and people detector which can, pose invariantly, detect people. Method depends only on depth information of point clouds taken from RGB-D sensors. As a result, it is robust against sudden light and contrast changes. The algorithm runs relying only on CPU, which makes it applicable to mobile robots with low computational resources