Human-robot　interaction　torque　(HT)　plays　an　important　role　in　the　application　of　lower　limb　rehabilitation　robots.　To　accurately　and　quickly　obtain　the　HT　applied　by　the　subject　during　the　rehabilitation　training,　an　estimation　approach　of　HT　is　proposed　on　the　basis　of　the　strong　tracking　Kalman　filter　(STKF)　and　modified　moving　average　method　in　this　paper.　First,　the　dynamics　and　its　corresponding　state　space　model　of　the　human-robot　hybrid　system　are　established.　Second,　the　STKF　is　designed　to　track　the　sudden　change　of　the　HT　rapidly.　Based　on　the　hull　moving　average　(HMA)　and　the　weighted　moving　average　(WMA),　the　hull-weighted　moving　average　with　a　variable　time　window　(H-WMA)　is　introduced　to　improve　the　estimation　accuracy　of　the　HT.　Finally,　experiments　were　conducted　using　a　lower　limb　rehabilitation　robot.　The　experimental　results　demonstrate　its　accurate　estimation　capabilities　for　assessing　the　HT.　In　addition,　the　proposed　method　has　an　effective　tracking　ability　even　when　the　HT　changes　suddenly.　　©　2023　IEEE.