STEREO VISION FOR AUTONOMOUS GROUND-BASED TRACKING OF MIGRATING RAPTORS
Open Access
- Author:
- Sarfraz, Sana
- Graduate Program:
- Aerospace Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- None
- Committee Members:
- Jacob Willem Langelaan, Thesis Advisor/Co-Advisor
Jacob Willem Langelaan, Thesis Advisor/Co-Advisor - Keywords:
- computer vision
systems
distributed
tracking
estimation
bird migration - Abstract:
- Migration places extreme demands on birds as they travel long distances. Factors influencing when and where migrating birds stop to feed are not well understood, and may include internal conditions such as amount of body fat and external conditions such as local weather. By collecting data on individual behavior in response to internal and external conditions researchers will be able to build models of bird behavior and identify critical habitats along migration routes. The following thesis presents the design and simulation of a system of stereo cameras and processors, to enable the distributed, autonomous tracking of migratory raptors in order to facilitate the study of their flight patterns. The motivation behind this study is to map the migratory routes of raptors and other endangered species of birds. The knowledge of migratory routes will aid biological research, conservation efforts and shed light on whether man-made structures like windmills along these routes pose a threat to migrating birds or if raptors are able to alter their flight patterns to avoid collision. A cost effective method that can continuously monitor the flight path of birds around the observation site with minimal or no human input, can provide valuable information about the migratory patterns of the birds without incurring the cost of employing more people or buying specialized equipment. The proposed system, consisting of a set of off-the-shelf cameras and processors, would not require a great deal of fiscal or labor input while providing an accurate estimate of bird migration patterns. Observation points may be set up along the ridge to obtain bearings to birds that come into view in order to compute their position and velocity at every time interval, using Kalman Filter based tracking algorithms. The accuracy of the estimates is lowered due to the Dilution of Precision, DOP, introduced into the measurements because of the large distances between the birds and the camera systems. The tracking system proposed in this thesis consists of a number of stations each composed of a camera pair and a processor, set up in a user specified geometry, along the ridge under observation. The estimation process is further complicated by the fact that the relations between the bird position and the bearings to it, utilize trigonometric properties, making the measurement model non-linear. A non-linear estimation method is therefore required. Each station computes an initial estimate of the position and velocity of the birds viewed by its cameras using a Particle Filter and further tracking is carried out by the Unscented Kalman Filter. Data association between measurements is performed from camera to camera in each stereo set and from frame to frame between one time step and the next. The estimates from each station are transmitted to a master computer that computes the association of local estimates to each other before fusing all the independent estimates to any particular bird and transmitting the resulting final estimate back to all the stations. Results of Monte Carlo simulations show the convergence of the estimated error to the true error for estimates from one or more stations. The tracking system provides fairly accurate estimates under realistic constraints and may be implemented with readily available hardware.