In this paper we present our runs submitted to the automatic search tasks of TRECVID 2009. This year, we submitted six runs for the automatic search task. These search runs primarily focused on (1) adaptively ranking the relevance of low level visual features to a topic and fusing those results (2) Estimating topic distribution using the Latent Dirichlet Allocation (LDA) technique during the match in LDA topic space and (3) formation of bag of words for low level features. The automatic runs submitted include, two runs based on MPEG7 visual features, two runs based on visual and high level features and two runs based on bag-of-words derived from SIFT features. 1.