1,720,972 research outputs found
On-line real-time fruit size estimation using a depth-camera sensor
Full text linkFruit weight is one of the factors taken into account when performing yield estimations together with the trees density and orchard's area. Thus, having the possibility to collect data about the weight of a large number of fruits in the orchard gives the possibility to increase the reliability of the yield estimation. Over recent years, mathematical models able to convert the fruit size into fruit weight were evaluated as effective. Since then, manual data collection with calipers and automated/continuous fruit gauges were tested to collect fruit size data to perform yield predictions. Their main drawbacks are respectively the need for human-labour, repetitiveness, being time-requiring and the limited sample varying from 20 to 200 fruits per hectare. This research is trying to discover and deepen the use of AI in agriculture for doing a step further: sizing fruits after their detection with a YOLOv5 Neural network algorithm. To reach this goal, a system which takes as a input RGB-D depth-camera's color images and 16 bit depth maps was developed. After applying YOLOv5 detection, two different methodologies (by mean of squared bounding boxes and circular shapes) to extract from the depth map the distance data needed to size the target object were tested. Results from a preliminary data-set showed that the system could be a potential solution to increase the sample dimension and perform yield prediction. The main drawbacks of the developed vision-system are related to the errors in sizing the objects, which are ranging from an underestimation of about 9 mm to an overestimation of 24 mm. From the initial results was possible to identify the squared-bbox-mediated sizing process as a better pathway rather than the one performed with circular-bboxes, since the RMSE is always smaller with values of 7–9 m
Data-Driven Model Predictive Control for Skid-Steering Unmanned Ground Vehicles
No full textSkid steering vehicles rely on tracks slipping to perform turning maneuvers. In this context, the estimation of the right amount of slip turns out to be significant to correctly perform precise movements. In a typical agricultural scenario, with rough terrain and narrow navigating spaces, a reliable slip estimation is crucial to perform safe motions. In this work, we propose a novel Gaussian Process approach to slip estimation in a tracked wheel robots by showing experimental results obtained from our prototype robotic platform
OpenAcces_RGBD_apple_dataset
No full textIntel realsense d435i open access dataset of seasonal growth of fuji apple. The dataset contains images and reference caliper ground truth data. Data were collected during 2022 season in a 3 years old apple orchard trained as 'Planar Cordon' (bidimensional training system). 12 fruit on two trees (24 fruit in total) were monitored for their fruit size along the whole season. RGB-D pictures, manually labelled for the monitored fruit, were taken on 17 different dates from a fruit size of 40mm approx. to >80mm approx
For more detailed info check the 'data_exploration' Jupyter notebook in the notebook folde
Development of a consumer-grade scanning platform for fruit thermal and position data collection
Full text linkClimate change and more frequent heatwaves exacerbate the issue of fruit sunburn in orchards. To facilitate fruit temperature dynamics investigation, in relation to fruit sunburn damage occurrence, a low-cost thermal scanning platform, based on depth and thermal consumer-grade cameras, was developed to collect position and temperature fruit information. The platform exploits the Robotic Operating System (ROS) to synchronize data collection from the sensors, the YOLOv5 object detection algorithms to automatically detect fruits to be analyzed, and a Python based pipeline to align images and extract temperature and position information of the fruits (apple and grape cluster). Results referred to a first version of the system shown a high correlation between estimated and actual temperature (r>0.92) and an acceptable positional error (∼0.15 m). Many improvements of the system are currently on-going to reach the expected performance on a second version of the platform
A computer vision system for in-field quality evaluation: preliminary results on peach fruit
No full textIn Italy, peaches are paid according to size, color
and appearance. Real time fruit harvest quality information
could support growers and the whole fruit chain improving
segmented selection for consumers as well as to increase
growers’ income. In this study, a computer vision system was
tested aiming to quantifying and sizing peaches in bins at
harvest time. Two different depth cameras the Intel RealSense
D435i and D455, and two different light conditions, natural and
artificial, were tested, to assess potential issues and to achieve
the most suitable set-up for future developments. Automated
fruit detection appeared less difficult, while the system presents
generally overestimation in fruit size. The D435i camera in
artificial light condition obtained the best outcome with a RMSE
of 17.91 mm, compared to the reference diameter of measured
fruit. Although the results obtained are with low accuracy and
precision, the vision systems technique seems promising and
suggests solutions to further improvements. Future studies will
focus on improving the system for sizing and color estimation,
coupled to georeferenced data directly in the field with the aim
of mapping field quality variability. The idea is to develop a lowcost
tool that coupled to harvesting platforms connects fruit
quality at the time of harvest to post-harvest operations
Mixing supervised and unsupervised learning algorithms to solve human perception subjectivity in internal fruit quality assessment
No full textDispositivo di supporto per operazioni di ricerca e soccorso
No full textIl presente trovato riguarda un dispositivo di supporto per operazioni di ricerca e soccorso comprendente un drone ed almeno un sensore per il rilievo di parametri dell'area sorvolata. Il numero di persone che frequenta la montagna (ed in generale i luoghi non urbanizzati) è in crescita continua e tale crescita ha determinato un corrispondente incremento degli incidenti che possono verificarsi agli escursionisti e/o sciatori e/o alpinisti. Tra gli incidenti possibili, quelli che rendono più difficoltose le operazioni di ricerca e soccorso sono le valanghe. In tali scenari si è soliti ricorrere a strumentazione specifica con apparecchi trasmittenti attivi (ARTVA) o passivi (transponder di RECCO) che consentono ai soccorritori di localizzare i travolti anche se non visibili ad occhio nudo perché sepolti dalla neve. Il compito principale del presente trovato è quello di proporre un dispositivo che consenta di velocizzare e rendere più sicure le operazioni di ricerca e soccorso di persone disperse, integrando la strumentazione utilizzata nell'ambito con un drone. La descrizione di tale dispositivo è affiancata ad una forma di esecuzione preferita, ma non esclusiva, del dispositivo di ricerca e soccorso integrato nel drone
- …
