A new robust technique for constructing intelligent environment with video image for mobile robots is proposed. The state of the art of the technique is an approach using autonomous extraction of feature from video images which captured in daily activities. Intelligent space embedded several sensors for supporting human activities are expected to expand future information infrastructure. However, in terms of practically, the robustness and computational cost are significant issues. In order to extract the feature robustly and to resolve an issue of tradeoff between accuracy and computational cost, support vector machine (SVM) is implemented for the classification and detection. Experiments in real world are conducted to evaluate the algorithm. The results show that the technique enables to detect image tag for constructing intelligent space.