Project DetailsThe goal of this work was to develop an algorithm for acoustic detection, localization, and tracking motorized targets, with the intention of implementing this algorithm on a drone in the future, which will be able to actively track a detected motorized target. For the benefit of the project, a small array of microphones is mounted on a drone, which picks up the sound coming from our target. The localization algorithm finds the most likely direction in which the target is located relative to the drone, based on the time differences between the microphones. After that, a Kalman filter-type tracking algorithm receives the output of the localization algorithm and performs smooth tracking of the target in real time, even when it moves on a completely random path. In addition, an algorithm was designed and implemented for classifying engine sounds, which determines whether it indeed hears a motorized target, and if so, whether it is necessary to continue tracking it. The localization algorithm runs in parallel with the engine classification algorithm, which verifies that the target has a continuous and consistent base frequency in the frequency ranges appropriate for the engines, and according to the existence of such frequencies, it enables or disables the tracking algorithm.
All this is done as a real-time operation to enable reliable tracking, while we have implemented several non-trivial techniques to accelerate performance, such as sampling and reconstruction of the cross-correlation signals between the microphones, and direction finding in a hierarchical grid.
