**6. Conclusion**

A construction method of automatic human tracking system with mobile agent technology is proposed using neighbor node determination algorithm. The construction method consists of the neighbor node determination algorithm and tracking methods. The neighbor node determination algorithm can compute neighbor nodes. This algorithm can be efficient to compute neighbor node and can make the system robust even if view distance of camera is changed. The tracking methods consist of follower method and detection method. The follower method can identify feature of a target locally. The detection method can search a lost target but a search cycle has to be within *walking speed × distance between cameras*. The detection method can be efficient to detect a target if the search cycle is near the walking speed. A mobile agent can keep tracking a target by using these detection methods if the agent lost the target. In addition, from the experiment results, the Stationary net detection method can detect a target faster than the Ripple detection method. And the Stationary net detection method can use smaller number of agents than the Ripple detection method. Because the Ripple detection method searches a target by widening a search gradually but the Stationary net detection method can widen a search efficiently by the Neighbor node determination algorithm.

The effectiveness of proposed tracking method was experimented using simulator and real environment. In the experiment using simulator, the tracking methods are experimented by the walking speed of a target. In the detection methods, consideration is added about the propriety of the parameter *n* which gives the number of non-camera nodes. Aimed to confirming behavior of the automatic human tracking system, the system in a real environment uses the simple image processing which can identify the color information of a target except the influence by the accuracy of image processing. And the follower method and the detection method are confirmed to be effective by a toy instead of a targeted person and to be able to construct the automatic human tracking system.

We will research more efficient detection to improve the automatic human tracking system. In addition, the accuracy of image processing has to be improved more to track a target more accurately. We are considering to improve our tracking system by combining effective studies and to improve image processing program by using PCA-SIFT (Ke et al. 2004) or SURF (Bay et al. 2008) algorithm.
