In　the　field　of　machine　vision,　deep　learning　has　become　a　foremost　method　to　perform　target　detection.　Although　suitable　detection　results　have　been　published,　the　feasibility　of　using　deep　learning　in　practice　to　detect　complex　targets　requires　further　research.　This　paper　explores　the　use　of　the　Faster　R-CNN　model　for　pedestrian　detection.　We　design　different　network　structures　and　study　their　effects　to　determine　the　optimal　solution　for　pedestrian　detection.　We　propose　two　methods　to　improve　the　effectiveness　of　detection　for　this　application.　Our　optimization　achieves　a　5%-18%　increase　in　the　detection　accuracy,　reaching　a　rate　of　88%.　To　optimize　the　detection　area　for　a　driving　scenario,　we　include　a　speed　factor　for　the　target　region　of　interest　of　the　driver.　We　collected　the　areas　of　interest　for　30　drivers　and　verified　that　the　matching　rate　was　60%-70%.　We　propose　two　solutions　for　complex　detection　scenarios　(involving　rotation　factors)　and　conduct　comparative　experiments.　We　use　the　post-rotation　factor　method　to　improve　the　detection　accuracy　by　6%.　We　thus　explore　the　influence　of　different　network　structures　on　recognition　performance　and　propose　accuracy　optimization　methods　for　the　application　of　pedestrian　detection.　©　Indian　Pulp　and　Paper　Technical　Association　2018.　All　rights　reserved.