Research On Vision-based Tracking Approaches For Ground Targets By A Small-scale Autonomous Helicopter

A small-scale helicopter is a highly maneuverable and versatile UAV plat-form presenting the following characteristics:it can take off and land vertically, perform longitudinal and lateral flight as well as drop and retrieve objects from otherwise inaccessible places. Owing to its flexibility, a helicopter is very suitable to accomplish such tasks as aerial photography, search-and-rescue, terrain explo-ration, military reconnaissance, and so on. An autonomous helicopter equipped with a visual sensor has the abilities of air surveillance and tracking moving tar-gets, thus it has become an ideal choice to implement such tasks as terrorism anti-riot, traffic surveillance, disaster relief. and so on. Therefore, it has recently received more and more attention and gradually becomes a topic of considerable interests over the past few years.The considered system is an autonomous helicopter equipped with a pan-tilt camera to track a ground dynamic target with a fixed relative position and ori-entation. To address the tracking problem, the pan-tilt movement is controlled to keep the target in the center of the image plane, at the same time, the states of the moving target are retrieved from its image coordinates, the attitude of the helicopter and the camera through geometry analysis, and this information is used to compute a control law which drives the helicopter to track the ground target. In the tracking system, the visual sensor is used to estimate the states of the ground target. However, due to the translation and rotation of the camera, the background image varies in the tracking process, which makes it a very dif-ficult problem to detect and to track the ground dynamic target. A small-scale helicopter is a typical nonlinear system with multiple variables and strong states coupling, it is also susceptible to interference from external disturbance such as wind. Based on these reasons, how to make an autonomous helicopter fly reliably is still a difficult problem, and the task to make it track a target with complex dynamic behavior is even more challenging. Besides, owing to the narrow field of view(FOV), it is necessary to control the camera's line-of-sight (LOS) to achieve accurate pointing. However, the problem of keeping the camera facing the dy-namic target is particularly challenging when the camera is carried on a mobile vehicle or when the target is highly dynamic.This dissertation makes a further study on the task of vision-based ground target tracking by using an autonomous helicopter. In general, the research con-tents in this dissertation can be summarized as the following three aspects:(1) Target states estimation by using a visual sensor. For a ground target tracking system using an unmanned helicopter, an estimator is proposed in this dissertation to estimate the states of the target in the world coordinate. With the goal to reduce the estimation errors caused by the noises from on-board sensors, the static features in two successive images are first utilized to calculate the image variation induced by the camera movement, which is then fused with the on-board sensors to estimate the states of the dynamic target.(2) Control of the autonomous helicopter to track a ground target. A novel control scheme is presented to drive an autonomous helicopter to track a ground target, which contains a two-level structure. On the higher level, the states of the target and the helicopter are used to compute a desired trajectory for the helicopter. On the lower level,the helicopter is controlled by the flight controller so as to follow the desired trajectory.(3) Control of the on-board pan-tilt movement. An adaptive on-board pan-tilt movement controller is proposed in this dissertation to adjust the camera's pose. so that the tracked ground moving target can be maintained in the center of the image, even in the presence of unknown camera extrinsic parameters. In addition, to address the problem of occlusion, an on-board pan-tilt movement controller is proposed to adjust the attitude of the camera, so that the target can be maintained in the center of the field of view(FOV) as close as possible even if the target image can not be obtained in a short period of time due to various reasons.Besides, to analyze the characteristics of the tracking system, a simulation testbed and an experiment platform are constructed to demonstrate the superior performances of the proposed control techniques.