The　process　of　motion　segmentation　and　measurement　of　mechanical　swing　based　on　traditional　block　shape　optical　flow　trajectory　group　clustering　exhibits　limitations　in　terms　of　over-segmentation　and　fragmentation　due　to　the　partial　occlusion,　interruption,　and　uneven　velocity　distribution　of　the　optical　flow　trajectory.　To　overcome　these　limitations,　we　herein　propose　an　arc-shaped　trajectory　clustering　algorithm　that　uses　curvature　as　a　similarity　metric　and　combines　it　with　point　cloud　registration　to　perform　mechanical　swing　measurement.　The　algorithm　first　performs　sparse　Gaussian　regression　of　the　active　subset　to　learn　the　average　trajectory　of　the　arc-shaped　trajectory　group.　Subsequently,　the　average　trajectory　is　used　as　the　seed　sample　of　the　sparse　subspace　clustering　to　complete　the　motion　segmentation　at　one　time.　Finally,　the　non-seed　sample　is　reclassified　into　its　surrogate　seed　sample　cluster　to　obtain　the　point　set　of　each　frame.　Through　conditional　expectation　point　cloud　registration,　the　rotation　component　is　extracted　to　complete　the　swing　angle　measurement.　The　proposed　algorithm　is　used　for　a　vehicle　windshield　wiper　under　the　four-link　wiper　assembly　model　and　six　different　environment　illuminances,　as　part　of　a　visual　automation　system　project　targeting　the　daily　safety　inspection　of　passenger　station　vehicles,　and　compared　with　other　algorithms.　The　experimental　results　show　that　the　proposed　algorithm　can　fully　learn　the　blocked　trajectory,　and　the　mean　square　error　with　an　artificially　calibrated　value　is　less　than　10%.　Furthermore,　the　computational　complexity　is　only　equivalent　to　that　in　the　case　of　a　single　iteration　of　the　alternating　direction　method　of　multipliers　(ADMMs),　Therefore,　the　proposed　algorithm　can　be　used　for　mechanical　vision　motion　measurement　in　industrial　intelligent　manufacturing　and　automatic　control　systems.