In　order　to　recognize　and　detect　tomatoes　for　providing　accurate　location　information　for　tomato　picking　robot　under　the　complex　environment　of　facility　greenhouse,　the　recognition　and　detection　method　based　on　YOLOV5　was　adopted　in　this　paper.　The　data　enhancement　method　was　used　to　improve　the　generalization　ability　of　network　model.　The　binocular　camera　was　also　used　to　collect　images　to　match　and　calculate　the　central　pixel　of　the　detected　tomatoes,　according　to　the　binocular　ranging　principle.　At　the　same　time,　the　parallax　value　of　the　detected　tomatoes　was　compared　with　the　real　value　in　different　environments.　It　is　proved　that　the　mAP　of　YOLOV5　method　is　96%,　the　absolute　value　of　stereo　matching　error　is　less　than　3　pixels,　and　the　matching　time　of　single　image　is　less　than　10ms,　which　effectively　improves　the　accuracy　and　efficiency　of　picking　robot.　　©　2022　IEEE.