研究目的
To investigate the applicability of four different RGB-D sensors for close-range outdoor agricultural phenotyping tasks, including measuring plant attributes and object detection under varying conditions.
研究成果
The Intel D435 sensor outperforms others in fill rate, depth quality, and size estimation accuracy for outdoor agricultural phenotyping, making it suitable for low-weight, energy-efficient applications like drone-based systems. Microsoft Kinect II is competitive but has higher energy demands. Intel SR300 is limited to close ranges, and Orbbec Astra S performs poorly outdoors. Depth information can enhance object detection, and the developed application for maize stem width estimation shows promise with an RMSE of 1.58mm.
研究不足
The study is limited to close-range outdoor conditions (up to 1.5m) and specific objects (corn plants, tomatoes, balls). Sensors may have mutual interference (e.g., Kinect II and SR300). Depth information fill rate varies with object type and lighting, and statistical significance was not always achieved due to sample size. Applications may be hindered by occlusions in field conditions.
1：Experimental Design and Method Selection:
An outdoor experiment was conducted to evaluate four RGB-D sensors (Microsoft Kinect II, Orbbec Astra S, Intel SR300, Intel D435) for phenotyping tasks. The design involved measuring plant attributes at various distances and light conditions, using statistical analysis and deep learning models for object detection and size estimation.
2：Sample Selection and Data Sources:
Six young corn plants, two tomatoes (red and orange cherry), and two plastic balls (green 50mm diameter, yellow 7.8mm diameter) were used as measured objects. Data were collected in 12 cycles from sunrise to sunset, with ambient light measured using a Galaxy S8 smartphone.
3：8mm diameter) were used as measured objects. Data were collected in 12 cycles from sunrise to sunset, with ambient light measured using a Galaxy S8 smartphone. List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: RGB-D sensors (Microsoft Kinect II, Orbbec Astra S, Intel SR300, Intel D435), computers for data capture, Galaxy S8 for light measurement, plastic pots for plants, and reference balls.
4：Experimental Procedures and Operational Workflow:
Sensors were fixed in an assembly; objects were placed at distances from 0.2m to 1.5m and moved sequentially. Images and depth data were captured using official SDKs (Kinect V2 SDK, Astra SDK, librealsense SDK). Fill rate was calculated for regions of interest, and deep learning models (Mask R-CNN) were trained for object detection and segmentation.
5：2m to 5m and moved sequentially. Images and depth data were captured using official SDKs (Kinect V2 SDK, Astra SDK, librealsense SDK). Fill rate was calculated for regions of interest, and deep learning models (Mask R-CNN) were trained for object detection and segmentation. Data Analysis Methods:
5. Data Analysis Methods: Statistical analysis included Levene's test, Welch one-way ANOVA, Games-Howell post hoc test for fill rate and error estimation. Root mean square error (RMSE) and mean relative average error (MRAE) were computed for size estimation. Deep learning models were evaluated using mean average precision (mAP).
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容-
RGB-D Sensor
Microsoft Kinect II
Microsoft Corporation
Capturing RGB and depth images for phenotyping tasks, based on Time-of-Flight technology.
-
RGB-D Sensor
Orbbec Astra S
Orbbec 3D Tech. Intl. Inc
Capturing RGB and depth images for phenotyping tasks, based on structured-light technology.
-
RGB-D Sensor
Intel SR300
Intel Corporation
Capturing RGB and depth images for phenotyping tasks, based on structured-light technology.
-
RGB-D Sensor
Intel D435
Intel Corporation
Capturing RGB and depth images for phenotyping tasks, based on active stereoscopy technology.
-
Smartphone
Galaxy S8
Samsung
Measuring ambient light intensity using its ambient light sensor during the experiment.
-
Software Development Kit
Kinect V2 SDK
Microsoft
Used for capturing images and depth data from Microsoft Kinect II sensor.
-
Software Development Kit
Astra SDK
Orbbec
Used for capturing images and depth data from Orbbec Astra S sensor.
-
Software Development Kit
librealsense SDK
Intel
Used for capturing images and depth data from Intel SR300 and Intel D435 sensors.
-
Deep Learning Model
Mask R-CNN
Used for object detection and segmentation in the images captured by the sensors.
-
Deep Learning Model
Faster R-CNN
Used for stem detection in the maize stem width application.
-
登录查看剩余8件设备及参数对照表
查看全部
