Linear-Trajectory Filtering

Norm Bartley

We have employed simple 3-D inductance-resistance (LR) networks as the basis of 3-D recursive digital filtering algorithms for enhancing a fundamental class of 3-D spatiotemporal-domain signals, namely linear-trajectory signals. Such signals encompass objects of constant size, shape and intensity that move in a straight line at constant speed when viewed, say, on a television screen. Linear-trajectory filtering is then to selectively enhance such objects on the basis of their direction and speed.

Below are some input and output images from a couple of our linear-trajectory filtering examples.

Our first work in this area is described in this paper.

The Cement Truck

Shown here is a single frame of an image sequence filmed on a busy roadway near the University of Calgary. If you wish to view an MPEG-encoded segment of the moving image sequence, please play the input traffic sequence (330K, 160x120, 100 frames). The moving vehicles in this sequence all represent good examples of linear trajectory signals. We can choose the parameters of the LR prototype network to selectively enhance vehicles moving at a particular velocity in this sequence.

Shown at left is a single frame of the filtered sequenced obtained from a linear-trajectory filter designed to enhance vehicles having the velocity of the cement truck. Please play the MPEG-encoded output traffic sequence (142K, 160x120, 100 frames). You will notice that all objects moving at speeds other than that of the cement truck, including the static background and vehicles in the lane of oncoming traffic, have been attenuated. The cement truck, and also the car beginning to overtake it, are clearly visible. This seqeuence was presented by Len Bruton in his Keynote Address at the 1990 Midwest Symposium on Circuits and Systems here in Calgary and the still images at left are featured on the cover of the conference proceedings.

Taxiing Aircraft

Another example of linear-trajectory filtering is for the image shown at left. This is a single frame of a sequence obtained by recording a taxiing aircraft through a chain link fence. Please play the MPEG-encoded input moving sequence (215K, 160x120, 100 frames). There are two important linear trajectory signals here; one is the aircraft, which is appears stationary in the image, and the other is the moving fence.

At left is a single frame of the output sequence obtained from a linear-trajectory filter designed for enhancing zero-velocity signals. The aircraft is a zero-velocity signal, so it is transmitted by the filter without serious distortion. All objects that move, including the ground and particularly the fence, are rejected or severely smeared. Please play the MPEG-encoded output sequence (149K, 160x120, 100 frames).