3D sparse imaging in biosonar scene analysis

Bats can navigate through foliage thanks to their biosonar. To build an image of their environment they emit ultrasonic pulses and analyse the corresponding echo signals which consist of a complex superposition of a set of highly overlapping reflected pulses coming from the targets at their different positions. We propose to use a sparse representation of echoes in overcomplete dictionaries to perform 3D imaging using biosonar signals. We show that even if the echosignal is a superposition of highly overlapping echoes the algorithm can efficiently reconstruct the image only from a single measurement. Those results can be useful from a robot sensor perspective where the robot has to find a free path. It can also help to understand better how bats can efficiently navigate through dense foliage.