DESIGN OF MULTI TARGET TRACKER FOR GROUND SURVEILLANCE RADAR

Abstract

The problem of multiple target tracking in ground surveillance radar is of much importance when high level information such as target trajectories are of interest to the users which is often the case. By gathering information originating from a target and using this information the trajectory of the target can be predicted with a certain degree of accuracy. This prediction can be useful for tasks such as target classi cation, source location and defensive actions etc. In this thesis, multiple-target tracker for ground surveillance radar is designed. To reduce the computational burden on the tracker and to eliminate redundancies induced by multiple target reports from a single target, multiple observations originating from the same target are fused together. This problem of clustering and fusion of measurements is studied and an algorithm is designed. The study of its computational complexity reveals that such algorithm is feasible and has a complexity of the order of Nlog(N). Data obtained over multiple scans of the radar is analyzed for the possibility of data forming tracks. This problem of curve tting is studied and an algorithm is developed which uses clustering technique to initiate the tracks for ltering and extrapolation. Data ltering techniques such as Kalman Filter is used to lter the data and Singer's model used for estimation of target states for extrapolation of target trajectory so that the new observations are associated with the currently initialized tracks. The problem of non-linearities induced by our observation scheme has also been addressed by the use of Extended Kalman Filter which uses Taylor's series expansion to linearize the observation scheme. Furthermore, the developed algorithms have been tested on the real life data gathered from radar developed at NUST. More than 50% of the observations were found to be redundant and were fused through clustering. 7 tracks were found to be present in the data through track initiation. Two of the more consistent targets were tracked by using Kalman Filter. The problem of multiple target tracking in ground surveillance radar is of much importance when high level information such as target trajectories are of interest to the users which is often the case. By gathering information originating from a target and using this information the trajectory of the target can be predicted with a certain degree of accuracy. This prediction can be useful for tasks such as target classi cation, source location and defensive actions etc. In this thesis, multiple-target tracker for ground surveillance radar is designed. To reduce the computational burden on the tracker and to eliminate redundancies induced by multiple target reports from a single target, multiple observations originating from the same target are fused together. This problem of clustering and fusion of measurements is studied and an algorithm is designed. The study of its computational complexity reveals that such algorithm is feasible and has a complexity of the order of Nlog(N). Data obtained over multiple scans of the radar is analyzed for the possibility of data forming tracks. This problem of curve tting is studied and an algorithm is developed which uses clustering technique to initiate the tracks for ltering and extrapolation. Data ltering techniques such as Kalman Filter is used to lter the data and Singer's model used for estimation of target states for extrapolation of target trajectory so that the new observations are associated with the currently initialized tracks. The problem of non-linearities induced by our observation scheme has also been addressed by the use of Extended Kalman Filter which uses Taylor's series expansion to linearize the observation scheme. Furthermore, the developed algorithms have been tested on the real life data gathered from radar developed at NUST. More than 50% of the observations were found to be redundant and were fused through clustering. 7 tracks were found to be present in the data through track initiation. Two of the more consistent targets were tracked by using Kalman Filter.