Biologically Inspired Scene Estimation (BIS-E)

Scene understanding is incredibly difficult, as it requires the recreation of a 3D environment from noisy 2D image data. Greedy approaches to estimating scene attributes are computationally expensive and are not sufficient for solving many real-world problems (e.g., detecting a suspect moving through a scene). Further complicating matters, is that different levels of detail/description are required for different tasks. Despite these challenges, biological systems have little difficulty efficiently describing scenes at multiple levels of detail. Therefore, our solution to this problem is a Biologically Inspired Scene Estimation (BIS-E) algorithm, which adaptively adjusts the features it uses through an optimal controller that balances the tradeoff between scene estimation uncertainty and computational cost. The result of this effort is a biologically-inspired algorithm that minimizes the computational cost associated with achieving the desired level of scene understanding.