Vision-Based Pedestrian Tracking And Facial Active Perception For Unmanned Aerial Vehicles

In this paper,we study the problem of pedestrian tracking and facial active perception using an UAV(unmanned aerial vehicle)with pattern recognition and computer vision skills.Pedestrian tracking means that in addition to tracking the specified walking pedestrian,the UAV will fly circularly around the target with a fixed distance for grasp whole body information.It has broad application prospects in criminal arrest,security monitoring and other fields.Facial active perception means on the basis of pedestrian circular tracking,the UAV will fly close to the target and locate his face,and flying to the front of the face to capture a proper and clear face photo on its own.In terms of pedestrian circular tracking,focusing on UAV motion platform,we present a new method in order to solve the problem of the circular tracking of a specific pedestrian under multi-disturbance conditions.Pedestrian circular tracking algorithm consists of four modules: illumination processing,pedestrian detection,feature extraction and target matching.In the Illumination processing part,we use the classical Retinex method to deal with the problem of the fast variation of the light field in the circling flight.In the pedestrian detection part,we adopt the deep learning object detection method SSD(Single Shot Multi-Box Detector)to detect pedestrian,and we use blocked image color feature when extracting features.In the target matching part,two different periods feature queues are constructed to store the features of the target pedestrian,and we use LLC(Locality-constrained Linear Coding)to reconstruct any pedestrian feature based on the feature queue,and use the results of the reconstruction error to determine whether it is the specific pedestrian.In terms of the facial active perception,we apply MTCNN(Multi-task Cascaded Convolutional Neural Networks)to detect face and locate key points,and design face scoring mechanism and flight control algorithm.The UAV can fly front the face of the target pedestrian under multi-disturbance conditions,and take a proper and clear face photo.Our UAV is Parrot Bebop2 which only has a fixed front camera without depth information.The pedestrian tracking and facial active perception are only vision-based,and we test a lot of experiments outdoors.The Experimental results verify the effectiveness and practicability of our method.