3D　object　detection　based　on　point　cloud　data　is　an　essential　part　of　L4　automatic　driving.　This　paper　proposes　a　novel　end-to-end　trainable　deep　learning　network　structure,　PPMGNet,　which　can　quickly　encode　point　clouds,　obtain　the　spatial　feature　of　point　clouds,　predict　multiple　categories,　and　perform　3D　object　detection　in　real　time.　A　large　number　of　experiments　show　that　in　terms　of　speed　and　accuracy,　PPMGNet＇s　detection　performance　is　very　suitable　for　direct　deployment　in　autonomous　driving　applications.