From 2003 to 2006 I developed a highly successful tracker for tracking moving objects from a moving vehicle using a single-axis laser scanner. As described in a Navlab publication this code formed the basis of an ongoing research program, including successful application to off-road robots with two-axis scanners.

This implementation is based primarily on fitting a rectangle to objects (vehicles) then fusing the motion estimates from the corner features by using a measurement-noise adaptive Kalman filter. When an apparently moving track detected, it is further refined by applying a more computationally expensive process of seeing how well the predicted trajectory matches the recent observations, and whether the motion hypothesis is significantly better than the not-moving hypothesis. This two-step process was necessary to get usefully low levels of false motion detection for collision warning and avoidance applications.

My project page from that time has pointers to documents and videos of performance results.