A generic and robust approach for the real-time detection of people and vehiclesfrom an Unmanned Aerial Vehicle(UAV) is an important goal within the frameworkof fully autonomous UAV deployment for aerial reconnaissance andsurveillance.Here we present an approach for the automatic detection of vehicles based onusing multiple trainedcascaded Haar classifiers with secondary confirmation inthermal imagery. Additionally we present a related approachfor people detectionin thermal imagery based on a similar cascaded classification techniquecombining additionalmultivariate Gaussian shape matching. The results presentedshow the successful detection of vehicle and people undervarying conditions inboth isolated rural and cluttered urban environments with minimal false positivedetection.Performance of the detector is optimized to reduce the overall falsepositive rate by aiming at the detection of each objectof interest (vehicle/person) at least once in the environment (i.e. per search patter flight path)rather than every object ineach image frame. Currently the detection rate forpeople is ~70% and cars ~80% although the overall episodic objectdetection ratefor each flight pattern exceeds 90%.
展开▼
