Any-time behavior for obstacle tracking
Karin Sobottka, Horst Bunke · 2003
Since any-time behavior allows a tradeoff between solution quality and search time, any-time algorithms are very useful for automotive applications in which robustness and real-time processing plays a significant role. In general, any-time algorithms have the property that a solution is ready whenever the algorithm is stopped and the solution improves with additional computing time. We make use of this for obstacle tracking. Our obstacle tracking scheme is based on finding feature correspondences. First, we apply a restricted matching strategy to obtain a first solution to the correspondence problem. Afterwards, while computing time remains, the current solution is improved using an optimal matching strategy. Following this strategy, a processing in real-time can be ensured at any time.