Performance　of　data-intensive　computing　is　one　of　the　kernel　problems　that　must　be　addressed　to　promote　the　development　of　embedded　high-resolution　object　detection　system.　In　this　study,　a　new　object　detection　framework　based　on　manycore　accelerator　was　established　to　improve　object　detection　performance　of　embedded　IoT　devices.　First,　the　fundamental　principle　of　object　detection　method　was　reviewed　as　the　basis　of　the　research.　Second,　some　key　designs　of　a　CPU-Accelerator　heterogeneous　architecture　based　parallel　object　detection　framework　including　data　splitting　strategy,　framework　architecture,　data　structure　design　and　parallel　cascade　classifier　design　were　proposed　to　improve　the　detection　speed　and　the　computational　resource　efficiency.　Third,　an　implementation　of　this　framework　on　a　Xilinx　Zynq　and　Adapteva　Epiphany　combined　hardware　platform　was　described.　Finally,　an　experiment　of　face　detection　application　was　conducted　to　evaluate　the　accuracy　and　performance　of　this　framework.　The　experimental　results　show　that　the　proposed　object　detection　system　provides　1.7　frame　per　second　process　speed　in　1920x1080　image　resolution,　about　7.8　times　speedup　than　the　cascade　classifier　algorithm　on　dual-core　ARM　CPU　which　was　integrated　in　Zynq　with　similar　accuracy.　The　results　demonstrate　the　promising　application　of　the　proposed　framework　in　the　field　of　object　detection　performance　improvement.