In　the　task　of　Human-Robot　Interaction,　human　beings　are　at　the　center　of　the　whole　cooperation　process.　Robots　always　pay　attention　to　the　cooperator’s　operation　behavior　through　multi-modal　information,　constantly　speculate　on　human’s　operation　intention,　and　then　plan　their　interaction　trajectory　in　advance.　Therefore,　understanding　human’s　operation　intention　and　predicting　human’s　trajectory　play　an　important　role　in　the　interaction　between　robot　and　human.　This　paper　presents　a　framework　of　human　intention　reasoning　and　trajectory　prediction　based　on　multi-modal　information.　It　mainly　consists　of　two　parts:　Firstly,　the　robot　accurately　infers　the　human　operation　intention　by　combining　four　kinds　of　information:　the　human　behavior　identified　by　GCN-LSTM　model,　the　object　type　obtained　by　YOLO,　the　relationship　between　verbs　in　sentences　and　interested　objects　and　targets　extracted　by　dependency　parsing　language,　and　the　gesture　of　handheld　objects　obtained　by　GGCNN.　Then,　after　understanding　the　intention　of　human　beings,　combined　with　some　tracks　that　human　beings　are　walking　out　of　at　present,　KMP　is　used　to　predict　the　next　tracks　of　human　beings,　which　is　the　preparation　for　the　trajectory　analysis　of　robots　in　future　human-computer　interaction　tasks.　We　verify　the　framework　on　several　data　sets,　and　the　test　results　show　the　effectiveness　of　the　proposed　framework.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.