Yoshio Kosuge

Target tracking under dense environments using range rate measurements

Summarizes a multiple target tracking algorithm using both position and range rate measurements. This filtering algorithm is discussed in terms of the fundamental issues of tracking performance, tracking success rate and tracking errors as compared with other conventional methodologies. Through several simulations, validity of this algorithm has been confirmed.

Sensor-to-sensor target association in a network of passive sensors

This paper address the problem of sensor-to sensor target association in a distributed passive sensor network. If two or more targets are coplanar with two sensors, ghosts occur among the intersections of line-of-sights. To cancel these ghosts, the authors propose a new association method in which they track angles and angular rates of targets being viewed by each sensor and use not only angle but also angular rate information in the association process. The performance of the proposed method is investigated through Monte Carlo simulation.

Tracking a three-dimensional moving target with two-dimensional angular measurements from multiple passive sensors

This paper addresses the problem of target tracking with angular measurements from distributed passive sensors. The usual triangulation method requires that the sensors observe a moving target synchronously to determine the target range. This paper proposes a new method which is applicable for the situation where the sensors observe the target asynchronously. The method assumes that each sensor observes two-dimensional angle, i.e., azimuth and elevation, and the target position and velocity in a three-dimensional space are to be estimated.

Target tracking with time-delayed data in multiple radar system

The problem of target tracking in a distributed radar system is considered. Measurements about target trajectory obtained by each radar are transmitted to a fusion center. One of the difficult problems encountered when we apply the well-known Kalman filter technique to these systems is a possibility of data out of sequence caused by time delay. We propose a filtering algorithm which can perform a negative-time update to resolve this problem. Our algorithm is an extension of the Kalman filter and is derived using a minimum mean-square error criterion. The performance of the proposed algorithm is investigated through Monte Carlo simulations.

Kalman Filter and ?-? filters for radar tracking

This paper discusses the tracking filter that estimates the true values of the states of the target such as position and velocity by Cartesian coordinates with the target position used as the observation value of the radar. A typical example is a Kalman filter or an α-β filter. A Kalman filter has a good tracking accuracy but a heavy computational load, whereas the α-β filter has a light computational load and is practical, although it has a problem in tracking accuracy. If the α-β filter with a light computational load has a tracking accuracy similar to the one for a Kalman filter, this filter becomes even more practical. Therefore, in this paper, the conditions are discussed where the α-β filter can approximate a Kalman filter. In this paper, the radar coordinates, where the target position vector is one axis, is used for calculation of the smoothing values (estimated values of the states for target motion at the present sampling time). It is proven that the α-β filter can approximate a Kalman filter if the angular velocity of the target and the rotation of the radar coordinates with the coordinate axis rotated with the target motion are infinitesimal and the process noise (a parameter indicating the ambiguity of the target motion model) is independent and the same between the coordinates. This result indicates that the α-β filter can approximate a Kalman filter if the target is at a low speed, the target range is large, or the target is proceeding toward the radar in a straight line.

Multisensor vehicle tracking method for intelligent highway system

Presents a vehicle tracking method which fuses data from an image sensor and radar installed on the roadside. The image sensor is widely used for road monitoring systems, but it is difficult to detect a vehicle in poor visibility. On the other hand, millimeter waves are less attenuated by fog, rain and snow. Consequently, it is possible to detect vehicles in all weather by using radar together with the image sensor. As for observation accuracy, the image sensor is superior in the accuracy of angle measuring. On the other hand, radar is superior in accuracy for range measurement. By utilizing these various features, therefore, the tracking performance is improved. This method adopts a correlation technique that uses a likelihood of range rate observed by radar, in addition to a likelihood of position, so that this method is generally able to track the vehicles from observation vectors even if false detection occurs. The performance of this method is evaluated by simulations.

Multi-target and multi-sensor data fusion by rule-based tracking methodology

The multi-target and multi-sensor tracking problem is a major topic of interest for air surveillance systems employing one or more sensors, to form and confirm tracks of targets amid a background of noise sources such as radar clutter. A practical and feasible technique, named here RBT (rule-based tracking) method, is proposed for this tracking problem. We define and implement several rules which field operators are supposed to apply to track initiation, track maintenance, track deletion and track integration. Simulation examples are given to illustrate how this RBT implementation leads to remarkable tracking performance.

Minimum eigenvalue analysis using observation matrix for bias estimation of two 3-dimensional radars

Abstract : Statistical analysis techniques in a two-radar system are presented for estimation of three dimensional (range, elevation and azimuth) radar biases, using a six-order Kalman filter. The minimum eigenvalue analysis using observation matrix is proposed in order to examine whether we could obtain satisfactory estimation accuracy. The performance of this method is evaluated by computer simulation. Computer simulation results show the close relationship between the minimum eigenvalue and the accuracy of estimated radar biases.

Computation-time reduction of track oriented multiple hypothesis tracking

In recent years MHT (multiple hypothesis tracking) attracts much attention because of its tracking performance. MHT has heavy computation load problem and an approach for computation time reduction by applying N-best algorithm was reported. We have improved the function of MHT and named the improved algorithm as track oriented MHT. It has different data structure from original Reids MHT d(1984), so N-best algorithm can not be applied to it straightforwardly. We show the applying method of N-best algorithm to track oriented MHT. And also we show the efficiency of the method by comparing the computation time in a simulation.

A study of target tracking using track-oriented multiple hypothesis tracking

MHT (multiple hypothesis tracking) is a multiple target tracking methodology in dense environments. While conventional tracking filters have only the function of track maintenance, MHT has the function of track initiation and mis-track removal in addition to the above function. In the conventional MHT, the extraction of target track is very complicated, because the track information is referred to by using combination of hypotheses and observation vectors. However, track-oriented MHT, our proposed methodology, generates hypotheses based on the correlation of tracks and observation vectors and, therefore, extraction of the tracks is facilitated. The paper presents the study of target tracking using track-oriented MHT.