Plane sweep method for optimal line fitting in track-before-detect

The Hough transform line detection method has been established as a viable technique for track-before-detect (TBD). However, its basic operation of binning and accumulating votes in the parameter space is computationally expensive. A more critical weakness of Hough transform is its dependence on parameter tuning (e.g., bin size and various thresholds), which can be non-intuitive and data-dependent. This leads to low detection rates in data with low signal-to-noise ratio and significant clutter. In this paper we propose a line detection algorithm with guaranteed global optimality for TBD. Our algorithm is based on the plane sweep algorithm for robust linear regression, with novel modifications to ensure its applicability under the TBD setting. Unlike the Hough transform, our algorithm has only one parameter to set (essentially the sensor false alarm rate) and can deterministically find the best solution according to a well-defined criterion. Simulation results on multi-dimensional TBD problems validate the accuracy and efficiency of our method.