3D machine vision system for robotic weeding and plant phenotyping

摘要: The need for chemical free food is increasing and so the demand a larger supply to feed growing global population. An autonomous weeding system should be capable of differentiating crop plants weeds avoid contaminating crops with herbicide or damaging them mechanical tools. For plant genetics industry, automated high-throughput phenotyping technology critical profiling seedlings at large scale facilitate genomic research. This research applied 2D 3D imaging techniques develop an innovative recognition holographic system. A time-of-flight (ToF) camera was used broccoli soybean plants. developed overcame previously unsolved problems caused by occluded canopy illumination variation. Both features were extracted utilized task. Broccoli algorithms based on characteristics At field experiments, detection rates over 88.3% 91.2% achieved plants, respectively. algorithm also reached speed 30 frame per second (fps), making it applicable robotic operations. Apart from applying vision recognition, reconstruction physical trait parameter estimation corn In this application, precise alignment multiple views plant. Previously published highlighted high-throughput, high-accuracy, low-cost systems morphology related characterization. contributed realization such integrating camera, ToF chessboard-pattern beacon array track camera's position attitude, thus accomplishing point cloud registration views. Specifically, target detection, pose tracking, spatial relationship calibration between cameras developed. phenotypic data obtained novel validated experimental generated instrument manual measurements, showing that has measurement accuracy more than 90% most cases under average less five seconds processing time