Motor　coordination　is　crucial　for　preschoolers＇　development　and　is　a　key　factor　in　assessing　childhood　development.　Current　diagnostic　methods　often　rely　on　subjective　manual　assessments.　This　paper　presents　a　machine　vision-based　approach　aimed　at　improving　the　objectivity　and　adaptability　of　assessments.　The　method　proposed　involves　the　extraction　of　key　points　from　the　human　skeleton　through　the　utilization　of　a　lightweight　pose　estimation　network,　thereby　transforming　video　assessments　into　evaluations　of　keypoint　sequences.　The　study　uses　different　methods　to　handle　static　and　dynamic　actions,　including　regularization　and　Dynamic　Time　Warping　(DTW)　for　spatial　alignment　and　temporal　discrepancies.　A　penalty-adjusted　single-frame　pose　similarity　method　is　used　to　evaluate　actions.　The　lightweight　pose　estimation　model　reduces　parameters　by　85%,　uses　only　6.6%　of　the　original　computational　load,　and　has　an　average　detection　missing　rate　of　less　than　1%.　The　average　error　for　static　actions　is　0.071　with　a　correlation　coefficient　of　0.766,　and　for　dynamic　actions　it　is　0.145　with　a　correlation　coefficient　of　0.653.　These　results　confirm　the　proposed　method＇s　effectiveness,　which　includes　customized　visual　components　like　motion　waveform　graphs　to　improve　accuracy　in　pediatric　healthcare　diagnoses.