Identification and manipulation of lily bulbs for an automated lily bulb planting system

Automation in agriculture is growing year by year. The goal of automating processes is to provide inexpensive and more effective solutions for everyday problems present in the industry. Automation in agriculture adds value to the product and in turn, to the farmer's infrastructure. This automation also aims to provide higher skill labour for workers that the automation processes substitute. Using machine vision as a means of automating processes is very common in factory environments and is being adapted for the external agriculture environments (i.e. automated detection for produce harvesting). Machine vision and manipulation techniques for a lily bulb plantation were presented. The techniques were investigated to determine the feasibility of using an autonomous, machine vision based approach to manipulate and plant lily bulbs from a provided source, to pre-augered holes produced by a pre-defined autonomous platform. The machine vision approach involved taking a top down image of the bulbs and identifying the head positions and what orientation they were facing relative to their root structures. This was achieved using various standard machine vision techniques like segmenting using global thresholding and identification of heads using the Hough circular transform. The investigated manipulation method involved applying the above mentioned vision system to a standard ABB IRB-120 universal manipulator with a three bellow suction gripper to pick up the detected bulbs and manipulate the bulbs in the orientation perceived by the vision system. It was found that the machine vision algorithm provided a 75 per cent success rate when providing an optimal region of interest within the bulbs head. The success rate is a considerably successful result as the detection algorithm not only needed to detect the location of the bulbs, but the centroid of its head and also determine the approximate orientation relative to each samples individual root structure. The manipulation results showed that the engagement of the suction gripper was a significant component of failure during testing. The observed success rate was at 41 per cent. This high failure rate means that further improvements should be made before a successful end effector and manipulation pair would be achieved. Improving suction rate or developing a specialized gripper for the specific amorphous bulbs would have to be investigated further before there is confirmation of a satisfactory solution for the Automated Lily Planter. Further work could be done to improve the algorithm and fine-tune the output provided. Improvements could be made to optimise the detection algorithm like improved lighting and better contrast between the bulbs colour gradient and that of the platform's background. Further development on the manipulators approach should also be conducted for validation