Obstacle Detection forLowFlying UnmannedAerial Vehicles UsingStereoscopic Imaging

This paperdescribes astereoscopic imaging algorithm thatismodified forobstacle detection inlowflying Unmanned Aerial Vehicles (UAVs). Inthis typeofflight, obstacle detection mustbecarried outquickly forthesystem tobeeffective inreal timea Additionally, since theaircraft isclose totheground, the horizon isusually atthetopofthefield ofviewandobstacles must bedistinguished fromtheclutter oftheterrain. Thesparse edge detection andreconstruction algorithm proposed, produces fast butpartial reconstructions oftheenvironment. Oneimageis passed through aseries ofedgedetectors togenerate averysparse outline oftheenvironment. Thisoutline isthencorrelated tothe second imageandtheresulting reconstruction isadded toamodel oftheenvironment. Although eachindividual reconstruction is incomplete, theoverall result after ashort initialization period isa modeloftheenvironment thatismorecomprehensive thana single stereoscopic correlation runwitha moredetailed edge detector. Simulation ofthealgorithm on testimagepatterns showedanincrease inperformance relative tothelength ofthe sequence ofstereo pairs. On average, thesignal tonoise ratio (SNR)forsparse edgereconstruction wassignificantly higher thanthat forsingle correlation withmoredetailed edgedetectors. Additionally itwasfoundthattheprocessing speedofthesparse edgedetection algorithm onapairofstereoscopic images isfaster thantheprocessing carried outbyamoredetailed edgedetector. A test flight wasalso carried outtotest thealgorithm inamore realistic scenario. Thetestconfirmed thatthesparseedge reconstruction algorithm resulted inamuchmoredetailed viewof theenvironment thanifasingle, moredetailed edgedetector had beenused