Coupling　effects　of　various　loading　conditions　can　cause　deflections,　settlements　and　even　failure　of　in-service　bridges.　Although　it　is　one　of　the　most　critical　loads,　unfortunately,　loading　conditions　of　moving　vehicles　are　difficult　to　capture　in　real　time　by　bridge　monitoring　systems　currently　in　place　for　sustainable　operation.　To　fully　understand　the　status　of　a　bridge,　it　is　essential　to　obtain　instantaneous　vehicle　load　distributions　in　a　dynamic　traffic　environment.　Although　there　are　some　methods　that　can　identify　overweight　vehicles,　the　captured　vehicle-related　information　is　scattered　and　incomplete　and　thus　cannot　support　effective　bridge　structural　health　monitoring　(BSHM).　This　study　proposes　a　noncontact,　vision-based　approach　to　identification　of　vehicle　loads　for　real-time　monitoring　of　bridge　structural　health.　The　proposed　method　consists　of　four　major　steps:　(1)　establish　a　dual-object　detection　model　for　vehicles　using　YOLOv7,　(2)　develop　a　hybrid　coordinate　transformation　model　on　a　bridge　desk,　(3)　develop　a　multiobject　tracking　model　for　real-time　trajectory　monitoring　of　moving　vehicles,　and　(4)　establish　a　decision-level　fusion　model　for　fusing　data　on　vehicle　loads　and　positions.　The　proposed　method　effectively　visualizes　the　3D　spatiotemporal　vehicular-load　distribution　with　low　delay　at　a　speed　of　over　30FPS.　The　results　show　that　the　hybrid　coordinate　transformation　ensures　that　the　vehicle　position　error　is　within　1　m,　a　5-fold　reduction　compared　with　the　traditional　method.　Wheelbase　is　calculated　through　dual-object　detection　and　transformation　and　is　as　the　primary　reference　for　vehicle　position　correction.　The　trajectory　and　real-time　speed　of　vehicles　are　preserved,　and　the　smoothed　speed　error　is　under　5.7%,　compared　with　the　speed　measured　by　sensors.　The　authors　envision　that　the　proposed　method　could　constitute　a　new　approach　for　conducting　real-time　SHM　of　in-service　bridges.