The goal of many vision systems for traffic monitoring is to measure the speed of vehicles and the occupancy of the road. Based on these figures, existing models can, in principle, detect traffic incidents and predict traffic congestion. One of the basic problems of employing vision for solving this problem comes from the perspective of the images. Because of the geometry of the placement of the cameras with respect to the road, the sizes of the vehicles change as they move in the field of view of the camera. Moreover, different image velocities may correspond to the same physical vehicle velocity and vice versa. Thus, the camera should be calibrated before being actually used for this type of measurements. It is also obvious that a top-down view of a road could greatly simplify the problem: in that case the problem becomes a 2D one.. Unfortunately, real top down views are impractical because they would either limit dramatically the field of observation or they would require mounting the camera at an impractically high altitude. The goal of this work is to provide means for transforming an ordinary view to a top down one by exploiting techniques in the field of projective geometry.
We would like to thank Ajuntament De Barcelona (AB), Barcelona, Spain for providing the original traffic video data.
Fig 4(a): Input traffic image. The goal is to provide a top-down view of the three central lanes (traffic approaching the camera). Click on the image to see a short video.
Fig 4(b): Close, top-down view, traffic top to bottom.This view has been synthetically created from the imagery data of Fig. 4(a). Click on the image to see a short video.
Fig 4(c): Distant, top-down view, traffic top to bottom. This view has been synthetically created from the imagery data of Fig. 4(a). Click on the image to see a short video.
Fig 4(d): Close, top-down view, traffic left to right. This view has been synthetically created from the imagery data of Fig. 4(a). Click on the image to see a short video.
Fig 4(e): Distant, top-down view, rtaffic left to right. This view has been synthetically created from the imagery data of Fig. 4(a). Click on the image to see a short video.
*** You may click on the above images to view a short video from the corresponding viewpoint ***
Technical report under preparation.
