Mesh addition based on the depth image (MABDI)

Many robotic applications, especially those whose goal is to aid or assist through human-robot interaction, utilize a rich map of the world for reasoning tasks such as collision detection, path planning, or object recognition. Such map, and the method used to produce it, must take into consideration real-world constraints. Most mesh-based mapping algorithms resemble a “black box” and do not provide a mechanism to close the loop and make decisions about the incoming information. MABDI leverages the global mesh by finding the difference between what we expect to see and what we are actually seeing, and using this to classify the incoming measurements as novel or not. This allows the surface reconstruction method to be run only on data that has not yet been represented in the global mesh. The result is an algorithm that becomes computationally inexpensive once the environment is known, but can also react to new objects.