Target　tracking　is　a　challenging　task　in　computer　vision.　It　aims　to　detect　and　track　particular　objects　in　sequences.　Illumination　variation,　motion　of　target,　occlusion　and　background　clutter　make　target　tracking　extremely　challenging.　We　propose　an　novel　online　target　tracking　method　which　based　on　extreme　learning　machine(ELM).　This　tracking　method　consists　of　three　modules:　training,　tracking　and　classifier　update.　The　training　stage　aims　to　train　ELM　by　using　the　training　set.　Extracting　histograms　of　oriented　gradients　(HOG)　features　in　the　first　frame　of　each　sequence　for　training　ELM.　Then　the　tracking　stage　will　make　predictions　about　the　object　position　and　detect　the　target　in　candidate　regions.　A　simple　object　motion　model　is　designed　to　predict　the　object　position.　Finally,　according　to　tracking　results,　the　classifier　can　be　updated　for　online　learning.　A　large　number　of　experimental　results　have　validated　this　proposed　method.　©　2016　IEEE.