An improved obstacle separation method using deep learning for object detection and tracking in a hybrid visual control loop for fruit picking in clusters

• An improved active obstacle separation method for strawberry-harvesting robots. A faster vision system that used deep learning object detection and tracking. The combined push drag-push to precisely separate the obstacles. was controlled in a hybrid visual loop based on system. closed continuous ”look-and-move” made more precise. Selectively picking target fruit surrounded by obstacles remains challenge harvesting This paper presents improvements strawberry clusters. accurate developed two neural networks color thresholding real-time detection, tracking localization of strawberries. We propose an operation from three stages. drag vectors were simplified calculated exact locations Also, different many systems only “looked” once entire process, new vision-based control method. In stage 1, simple closed-loop at key points. stages 2 3, re-perceived environment update information before each round movements. Field evaluation showed proposed precise without reducing speed, increasing whole process success rate 62.4% clusters “Murano” cultivator 36.8% higher than previous work.