In　order　to　measure　the　heart　rate　and　breath　rate　of　human-body　and　evaluate　the　fatigue　level　in　real-time,　a　noncontact　human　fatigue　assessment　system　based　on　millimeter　wave　radar　AWR1642　is　proposed　in　this　paper　to　assess　the　health　condition　of　human　conveniently.　It　has　the　characteristics　of　high　classification　accuracy　and　high　accuracy.　The　function　of　system　mainly　includes　getting　accurate　heart　rate　and　breath　rate　by　AWR1642,　establishing　data　set,　extracting　features,　annotating　data　and　predicting　fatigue　level　by　Particle　Swarm　optimization　Back　Propagation　(PSO-BP)　neural　network　model.　The　experiment　results　show　that,　heart　rate　got　by　AWR1642　had　an　error　less　than　5%　compared　with　heart　rate　got　by　standard　medical　oximeter.　The　system　can　be　used　for　the　parameters　measurement　of　vital　sign.　On　the　other　hand,　the　PSO-BP　neural　network　model　built　with　the　variance　of　heart　rate　and　breath　rate　is　difficult　to　cause　over　fitting.　The　system　has　a　good　practicability　can　be　used　for　the　prediction　of　fatigue　level　and　it＇s　accuracy　can　reach　93.74%.　©　2021　IEEE.