The ASSET-2 system is built around feature-based image motion estimation. The features used primarily are two dimensional (often referred to as ``corners''), and edges can be used to refine the results obtained by using two dimensional feature motion. The main part of ASSET-2 has solely two dimensional calculations and not three. This means that feature matching is performed in the image plane, as is object segmentation and tracking. The only area in the main body of ASSET-2 where three dimensional behaviour is acknowledged is in predicting which one of two touching objects will occlude which. The reason for staying in the image plane is that no more information about the world processes can be obtained by working in three dimensions than by staying in the original two. Unless interaction with outside processes and measurements is needed at an early stage in a vision system, there is no need to move to three dimensions until the final interpretation of the processed data is made.
A brief summary of ASSET-2 is now given, and can be seen graphically in Figure 1.
Figure 1: Overview of the ASSET-2 system.
ASSET-2 is fed a stream of digitized video pictures, usually taken from a moving platform. For an example sequence see Figure 2, taken from ROVA (ROad Vehicle Autonomous, the autonomous vehicle at DRA Chertsey). In this sequence ROVA is travelling forwards and turning slowly to the left with the curve of the road. The ambulance is overtaking ROVA and the Landrover is travelling with almost identical speed to ROVA. At about frame 103 the ambulance passes behind the Landrover, and becomes occluded.
Figure 2: 15 frames from an example video sequence which ASSET-2 is
expected to process. Every ninth frame is shown.
Each frame taken by the video camera is initially processed to find two dimensional features and edges. The two dimensional feature list is passed to a feature tracker which uses a two dimensional motion model to match and track features over as many frames as possible. A two dimensional vector field can then be created by taking either feature velocities or displacements over a fixed number of frames. The resulting vector field is passed to a flow segmenter which splits the list of flow vectors into clusters which have similar flow within them and are different to each other, using the assumption of first order flow variation.
Next, this cluster list is compared with a temporally filtered list of clusters, and the filtered list is updated using the newly found clusters. This gives good results as it stands, but the boundaries of the clusters in the filtered list can be further refined by using image edges, if available.
This list of clusters which are spatially and temporally significant is finally used to provide information about the motions of objects viewed by the video camera. The three dimensional positions and motions of these objects can be estimated by making simple assumptions about the world. In the case of an autonomous vehicle, this information is of practical value; for example it can enable the avoidance or the following of other vehicles.