When performing remote sensing, one often uses vehicular mounted sensors. This provides the flexibility of moving around and searching over a large area and can be done via airborne vehicles, ground vehicles or marine vehicles. For this type of sensing, one needs to know the position and orientation of the sensor platform in order to ground register the location of detected objects. The research reported herein is concerned with the use of a ground vehicle as a sensor platform. Digital camera-type sensors such as infra-red are considered. The focus is on requirements for accurate ground registration of detected objects of interest. A four-antenna GPS array has been chosen for vehicle attitude measurement. Relationships between positions of the array elements and vehicle attitude are derived. It is seen that the attitude computations depend on differences in the various measurements. Thus common-mode errors in the measured position of the array elements cancel, enabling quite accurate attitude measurement even when utilizing somewhat imprecise units in the GPS array. A more precise single differential GPS (DGPS) has been chosen for vehicle position measurements. Relationships between pixel coordinates in the image frame and the angle of the corresponding ray from the camera to the object of interest are derived. A series of transformations are used to convert this ray to ground coordinates. Finally the intersection of the ray with the ground is computed based on the assumption that the ground in the field-of-view is flat and at a known elevation. In this manner an object of interest in the image frame may be ground registered. Sensitivity of the ground registration with respect to vehicle attitude measurement errors is developed. It is seen that small errors in pitch, roll or yaw can cause quite large errors in the computed ground coordinates. In the case of multiple looks at the same object of interest, the geo-registration process involves target tracking and data association. This process is facilitated by combining the single-look measurements in an optimal fashion via a Kalman Filter. In fact the accuracy obtained via multiple looks can be significantly greater than for a single look. The results obtained indicate that accurate geo-registration of remotely sensed objects is possible when using vehicular mounted sensors in conjunction with DGPS and that such a scheme is feasible with commercially available GPS and IR cameras. Geo-registration accuracy within a fraction of a meter is attainable for near objects.
|