中文题名: | 基于RGB-D多维信息融合的果蔬目标检测研究 |
姓名: | |
学号: | 2013212007 |
保密级别: | 公开 |
论文语种: | chi |
学科代码: | 082801 |
学科名称: | 工学 - 农业工程 - 农业机械化工程 |
学生类型: | 博士 |
学位: | 工学博士 |
学校: | 南京农业大学 |
院系: | |
专业: | |
研究方向: | 农业机器人 |
第一导师姓名: | |
第一导师单位: | |
完成日期: | 2021-10-12 |
答辩日期: | 2021-12-05 |
外文题名: | Research on Fruits and Vegetables Detection Based on The Fusion of RGB-D Information |
中文关键词: | |
外文关键词: | Fruit and vegetable detection ; RGB-D ; Point cloud data processing ; Deep learning ; Feature fusion |
中文摘要: |
多年来,果蔬目标检测一直是农业机器人领域的研究热点,但成熟度、生长姿态、环境光照、硬件设备、算法模型等因素对目标检测准确率有不同程度影响。基于RGB图像的目标检测随着机器学习、深度学习技术的成熟,准确率和速度已达到较高水准,但由于缺少深度数据,单一RGB图像不能准确反映场景和目标的空间信息。深度图像和三维点云数据包含深度信息并能够反映出场景的空间结构,但对场景的语义信息表达能力较差。 综上,本文以苹果和甜椒为主要研究对象,开展在不同特征融合策略和目标检测算法下,基于RGB-D多维信息融合的果蔬目标检测研究。针对如何增强单一特征的表述能力、如何提升算法对环境因素的抗干扰能力、如何提升算法在特殊场景下的鲁棒性以及如何优化目标检测的精度等问题,提出了相应策略和解决方案,通过试验对方法进行验证。主要研究内容如下: (1) 在RGB-D场景中,三维几何特征对目标的表述能力不足。基于此,开展RGB信息与三维特征的融合研究。首先在传统区域生长分割算法的基础上,以点之间色差值为收敛条件,完成苹果树点云分割任务。然后,将三维特征FPFH与颜色特征进行点元级融合,生成新的三维点云描述子Color-FPFH,通过t-SNE算法对特征进行降维可视化,验证特征融合方法的可行性。最后,采用基于遗传算法的支持向量机GA-SVM作为分类器进行识别试验,苹果、树枝、树叶的识别正确率分别为:92.30%、88.03%、80.34%。通过Color-FPFH与其他特征在识别精确率、召回率、运算速度等方面的横向试验对比,进一步验证了Color-FPFH特征在RGB-D空间中的相比于其他特征的优势。 (2) 基于RGB图像的目标检测易受光照、果实生长状态(遮挡、重叠)等因素影响。试验表明,在光照较强或较弱场景、果实被遮挡或重叠情况较严重场景中,果实目标检测精确率下降超过10%,为了提高深度学习目标检测模型对上述因素的抗干扰性,提出一种基于DS证据理论的决策级信息融合策略。首先,将苹果果园采集到的RGB图像和深度图像分别输入目标检测模型进行多网络并行训练,从学习率、批尺寸、Dropout、优化器选择等方面对网络进行优化。然后用DS证据理论将多网络输出的目标概率值进行合并运算,得到最终目标识别结果后进行边界框回归,完成目标检测任务。试验结果表明,苹果果实检测精确率、召回率分别为0.931和0.950,优于试验中的对比方法。进一步,提出了一种RGB与深度图像配准的快速目标深度信息提取方法,在完成目标检测任务后,在相同尺寸、分辨率以及基坐标系下,将边界框平行映射到深度图像,提取目标的深度值。 (3) 三维点云数据密度散化问题在一定程度上会降低目标识别准确率,影响算法稳定性。基于此,提出一种基于八叉树Octree结构改进的PointNet++深度卷积网络。首先,通过八叉树构建点云数据分组网络,该网络特点为由父节点到子节点自上而下点密度逐步增大。然后将该网络并入到PointNet++深度卷积网络中,卷积尺寸随网络深度增加逐步减小,以自适应点云密度变化并有效提取不同密度下的特征。以大棚环境中的甜椒果实和果柄为识别对象,试验证明Octree-PointNet++网络识别准确率、精确率、召回率要优于方面试验中的对比算法,不仅在点云密度下降时的识别鲁棒性最强,而且相比较原网络,运算速度也有提升。 (4) 单视角下采集到的点云数据通常不完整,且经常会由于遮挡、腐蚀造成目标数据残缺,使得目标检测算法回归的三维边界框尺寸误差较大。针对这一问题,首先改进了Frustum PointNet网络的三维边界框回归损失函数,采用Smooth L1函数提升网络对数据异常样本的鲁棒性。然后,以精度较高的基于二维图像回归的目标边界框尺寸范围作为约束条件,对点云数据中的三维边界框参数估计进行优化。试验证明,本文提出改进方法能够更准确的估计出真实目标的边界框参数。对于场景中缺失数据的目标点云,能够帮助进行尺寸修复,进一步得到更接近完整目标的数据。 |
外文摘要: |
In recent years, fruit and vegetable target detection has been a research hotspot in the field of agricultural robotics, but factors such as maturity, growth pose, environmental lighting, hardware equipment, and algorithm models have different degrees of influence on detection accuracy. With the maturity of machine learning and deep learning technologies, the target detection accuracy and speed have reached a high level, but due to the lack of depth data, a single RGB image cannot accurately reflect the spatial information of the scene and the target. Depth images and 3D point cloud data contain depth information and can reflect the spatial structure of the scene, but the ability to express the semantic information of the scene is poor. In summary, this paper carries out the research of fruit and vegetable detection based on the fusion of RGB-D information under different feature fusion strategies and algorithms, with apples and bell peppers as the main research objects. The corresponding strategies and solutions are proposed for how to enhance the representation capability of single features, how to increase the immunity of the algorithm to environmental factors, how to improve the robustness of the algorithm in special scenes and how to optimize the accuracy of detection, and the methods are verified through experiments. The main research contents are as follows. (1) In RGB-D scenes, the 3D geometric features have insufficient ability to represent the target. Based on this, the research on the fusion of RGB and features is carried out. Firstly, based on the traditional region growth segmentation algorithm, the apple tree point cloud segmentation task is completed with the color difference value between points as the convergence condition. Secondly, the 3D features FPFH are fused with the color features at the point element level to generate a new 3D point cloud descriptor Color-FPFH, and the features are visualized by the t-SNE algorithm for dimensionality reduction to verify the feasibility of the feature fusion method. Finally, GA-SVM, a support vector machine based on genetic algorithm, is used as a classifier for recognition tests, and the recognition accuracy of apples, branches and leaves are 92.30%, 88.03% and 80.34%, respectively. Through the horizontal test comparison between Color-FPFH and other features in terms of recognition accuracy, recall rate, computing speed, the advantages of Color-FPFH feature in RGB-D space compared to other features are further verified. (2) The detection task based on RGB images is susceptible to factors such as illumination and fruit growth status (occlusion, overlap). Experiments show that the accuracy of fruit detection decreases more than 10% in scenes with strong or weak illumination and in scenes with serious fruit occlusion or overlapping conditions. To improve the anti-interference of deep learning detection models against the above factors, a decision-level information fusion strategy based on DS evidence theory is proposed. First, RGB images and depth images captured from apple orchards are input into the detection model separately for multi-network parallel training, and the networks are optimized in terms of learning rate, batch size, Dropout, and optimizer selection. Then the DS evidence theory is used to merge the target probability values from the multi-networks, and the final target recognition results are obtained and then the bounding box regression is performed to complete the detection task. The experiment results show that the apple fruit detection accuracy and recall are 0.931 and 0.950, respectively, which are better than the comparison methods in the experiment. Further, a fast target depth information extraction method with RGB and depth image alignment is proposed to extract the depth value of the target by mapping the bounding box to the depth image in parallel under the same size, resolution and base coordinate system after completing the target detection task. (3) The problem of density scattering of 3D point cloud data will reduce the recognition accuracy and affect the stability of the algorithm. Based on this, an improved PointNet++ deep convolutional network is proposed based on the Octree structure. First, a point cloud data grouping network is constructed by octree, which is characterized by increasing point density from parent nodes to children nodes from top to bottom. This network is then incorporated into the PointNet++ deep convolutional network in parallel, and the convolutional size decreases gradually with increasing depth of the network to adapt to the changes in point cloud density and extract the features at different densities effectively. Using sweet pepper in the greenhouse environment as recognition objects, it is demonstrated that the recognition accuracy, precision, and recall of the Octree-PointNet++ network are better than those of the comparison algorithms in the aspect experiments, and the comparison yields the most robust recognition when the point cloud density decreases, and the computational speed is improved compared with the original network. (4) The point cloud data collected in a single view is usually incomplete, and the target data is often incomplete due to occlusion and corrosion, which makes the 3D bounding box size error of the detection algorithm regression larger. To address this problem, the 3D bounding box regression loss function of the Frustum PointNet network is first improved, and the Smooth L1 function is used to enhance the robustness of the network to data anomaly samples. Then, the target bounding box size range based on 2D image regression with higher accuracy is used as a constraint to optimize the 3D bounding box parameter estimation in point cloud data. It is experimentally demonstrated that the proposed improved method can more accurately estimate the bounding box parameters of real targets. For the target point cloud with missing data in the scene, it can help to perform dimensional repair and further obtain data closer to the complete target. |
参考文献: |
[1] 李焕玲. 中国果品流通市场现状和发展方向[J]. 中国经贸导刊, 2019(19): 25–31. [2] 卢军, 王贤锋, 后德家. 水果采摘机器人视觉系统研究进展[J]. 湖北农业科学, 2012, 51(21): 4705–4708. [3] 王丹丹, 宋怀波, 何东健. 苹果采摘机器人视觉系统研究进展[J]. 农业工程学报, 2017, 3(10): 67–77. [4] 宋健, 张铁中, 徐丽明, 等. 果蔬采摘机器人研究进展与展望[J]. 农业机械学报, 2006(5): 158–162. [5] 国务院关于印发新一代人工智能发展规划[EB/OL]. [2017–07–20]. http://www.gov.cn/zhengce/content/2017–07/20/content_5211996.htm [6] 刘继展. 温室采摘机器人技术研究进展分析[J]. 农业机械学报, 2017(12): 1–19. [7] Clement J, Novas N, Gazquez A, et al. An active contour computer algorithm for the classification of cucumbers[J]. Computers & Electronics in Agriculture, 2013, 92(1): 75–81. [8] Henten E J V, Schenk E J, Willigenburg L G V, et al. Collision–free inverse kinematics of the redundant seven–link manipulator used in a cucumber picking robot[J]. Biosystems Engineering, 2010, 106(2):126–140. [9] 冯青春, 袁挺, 纪超, 等. 黄瓜采摘机器人远近景组合闭环定位方法[J]. 农业机械学报, 2011, 24(4): 111–121. [10] 叶敏, 邹湘军, 罗陆锋, 等. 荔枝采摘机器人双目视觉的动态定位误差分析[J]. 农业工程学报, 2016, 32(5): 50–56. [11] 熊俊涛, 邹湘军, 陈丽娟, 等. 基于机器视觉的自然环境中成熟荔枝识别[J]. 农业机械学报, 2011(10): 67–77. [12] Hayashi S, Shigematsu K, Yamamoto S, et al. Evaluation of a strawberry–harvesting robot in a field test[J]. Biosystems Engineering, 2010(9): 257–280. [13] 张凯良, 杨丽, 张铁中. 草莓采摘位置机器视觉与激光辅助定位方法[J]. 农业机械学报, 2010, 41(4): 151–156. [14] De–An Z, Jidong L, Wei J, et al. Design and control of an apple harvesting robot[J]. Biosystems Engineering, 2011, 110(2): 112–122. [15] 顾宝兴, 姬长英, 王海青, 等. 智能移动水果采摘机器人设计与试验[J]. 农业机械学报, 2012, 43(6): 153–160. [16] 吕继东, 赵德安, 伟 姬, 等. 苹果采摘机器人对振荡果实的快速定位采摘方法[J]. 农业工程学报, 2012, 28(13): 48–53. [17] Tanigaki K, Fujiura T, Akase A, et al. Cherry–harvesting robot[J]. Computers & Electronics in Agriculture, 2008, 63(1): 65–72. [18] Naoshi, Kondo, Koki, et al. Development of an End–Effector for a Tomato Cluster Harvesting Robot[J]. Engineering in Agriculture Environment & Food, 2010(2):57–71. [19] Kondo N . Robotization in fruit grading system[C] 第三届亚洲精细农业会议暨第五届智能化农业信息技术国际会议论文集, 2009: 606–615. [20] Bac C W, Hemming J, Van Henten E J. Robust pixel–based classification of obstacles for robotic harvesting of sweet–pepper[J]. Computers and Electronics in Agriculture, 2013, 96(1): 148–162. [21] Reed, J. N, Miles, et al. Automatic mushroom harvester development[J]. Journal of Agricultural Engineering Research, 2001(2):112–131. [22] 张铁中, 陈利兵, 宋健. 草莓采摘机器人的研究:Ⅱ.基于图像的草莓重心位置和采摘点的确定[J]. 中国农业大学学报, 2005, 10(1): 48–51. [23] 王粮局, 张立博, 段运红, 等. 基于视觉伺服的草莓采摘机器人果实定位方法[J]. 农业工程学报, 2015, 31(22): 25–31. [24] 李伟, 王库, 谭豫之, 等. 温室环境下黄瓜采摘机器人技术研究[C]. 全国先进制造装备与机器人技术高峰论坛, 2008:25–36. [25] 纪超, 冯青春, 袁挺, 等. 温室黄瓜采摘机器人系统研制及性能分析[J]. 机器人, 2011, (06): 726–730. [26] 吕继东, 赵德安, 姬伟. 苹果采摘机器人目标果实快速跟踪识别方法[J]. 农业机械学报, 2014, 45(1): 65–72. [27] 项荣, 应义斌, 蒋焕煜. 田间环境下果蔬采摘快速识别与定位方法研究进展[J]. 农业机械学报, 2013(11): 218–233. [28] 赵源深, 贡亮, 周斌, 等. 西红柿采摘机器人非颜色编码化目标识别[J]. 农业机械学报, 2016(7): 35–46. [29] 宋怀波, 张传栋, 潘景朋, 等. 基于凸壳的重叠苹果目标分割与重建算法[J]. 农业工程学报, 2013, 29(3): 163–168. [30] 汤修映, 张铁中. 果蔬收获机器人研究综述[J]. 机器人, 2005, 27(1): 90–96. [31] 戚利勇. 黄瓜采摘机器人视觉关键技术及系统研究[D]. 浙江工业大学, 2012. [32] Li J B, Huang W Q, Zhao C J. Machine vision technology for detecting the external defects of fruits a review[J]. Journal of Photographic ence, 2014, 63(5): 241–251. [33] Apple crop–load estimation with over–the–row machine vision system[J]. Computers & Electronics in Agriculture, 2015, 120(1): 26–35. [34] Liu, Chengliang, Zhao, et al. A review of key techniques of vision–based control for harvesting robot[J]. Computers and Electronics in Agriculture, 2016, 9(1):66–81. [35] Zhang Z, Heinemann P H, Liu J, et al. The Development of Mechanical Apple Harvesting Technology: A Review[J]. Transactions of the Asae American Society of Agricultural Engineers, 2016, 59(5): 1165–1180. [36] Jing Z, Qin Z. Multi–class object detection using faster R–CNN and estimation of shaking locations for automated shake–and–catch apple harvesting[J]. Computers & Electronics in Agriculture, 2020(1):173-182. [37] Piatti D, Rinaudo F. Sr–4000 and camcube3.0 time of flight (tof) cameras: Tests and comparison[J]. Remote Sensing, 2012, 4(4): 1069–1089. [38] Yang L, Zhang L, Dong H, et al. Evaluating and improving the depth accuracy of Kinect for Windows v2[J]. IEEE Sensors Journal, 2015, 15(8): 4275–4285. [39] Al-Mallahi A . A novel image processing algorithm to separate linearly clustered kiwifruits[J]. Biosystems Engineering, 2019, 183:184-195. [40] Zhenglin W, Kerry W, Brijesh V. On–Tree Mango Fruit Size Estimation Using RGB–D Images[J]. Sensors, 2017, 17(12): 2738–2749. [41] Tu S, Xue Y, Zheng C, et al. Detection of passion fruits and maturity classification using Red–Green–Blue Depth images[J]. Biosystems Engineering, 2018, 175(1): 156–167. [42] Tu S, Pang J, Liu H, et al. Passion fruit detection and counting based on multiple scale faster R–CNN using RGB–D images[J]. Precision Agriculture, 2020(1): 172–189. [43] Gené–Mola J, Vilaplana V, Rosell–Polo J R, et al. Multi–modal deep learning for Fuji apple detection using RGB–D cameras and their radiometric capabilities[J]. Computers and Electronics in Agriculture, 2019, 162(1): 689–698. [44] Liu Z, Wu J, Fu L, et al. Improved kiwifruit detection using pre–trained VGG16 with RGB and NIR information fusion[J]. IEEE Access, 2020, 8(1): 2327–2336. [45] Inkyu S, Zongyuan G, Feras D, et al. DeepFruits: A Fruit Detection System Using Deep Neural Networks[J]. Sensors, 2016, 16(8): 1222–1235. [46] Milella A , Marani R , Petitti A , et al. In-field high throughput grapevine phenotyping with a consumer-grade depth camera[J]. Computers and Electronics in Agriculture, 2019, 156:293-306. [47] Wang H, Mao W, Liu G, et al. Identification and location system of multi–operation apple robot based on vision combination[J]. Nongye Jixie Xuebao/Transactions of the Chinese Society of Agricultural Machinery, 2012, 43(12): 165–170. [48] Arad B, Balendonck J, Barth R, et al. Development of a sweet pepper harvesting robot[J]. Journal of Field Robotics, 2020, 19(2): 235–351. [49] 麦春艳, 郑立华, 孙红, 等. 基于RGB–D相机的果树三维重构与果实识别定位[J]. 农业机械学报, 2015(1): 35–40. [50] Nguyen T, Vandevoorde K, Wouters N, et al. Detection of red and bicoloured apples on tree with an RGB–D camera[J]. Biosystems Engineering, 2016(2): 33–44. [51] Lehnert C, English A, Mccool C, et al. Autonomous Sweet Pepper Harvesting for Protected Cropping Systems[J]. IEEE Robotics & Automation Letters, 2017, 12(2): 1–15. [52] Milella A, Marani R, Petitti A, et al. In–field high throughput grapevine phenotyping with a consumer–grade depth camera[J]. Computers and Electronics in Agriculture, 2019, 156(1): 293–306. [53] 王丹丹, 宋怀波, 何东健. 苹果采摘机器人视觉系统研究进展[J]. 农业工程学报, 2017, 33(10): 59–69. [54] Xia C, Wang L, Chung K, et al. In Situ 3D Segmentation of Individual Plant Leaves Using a RGB–D Camera for Agricultural Automation[J]. Sensors, 2015, 15(8): 20463–20479. [55] Ilea D E, Whelan P F. Image segmentation based on the integration of colour–texture descriptors—A review[J]. Pattern Recognition, 2011, 44(10): 2479–2501. [56] 付鹏. 基于机器视觉的苹果检测与识别关键技术研究[D]. 西北农林科技大学, 2012. [57] 李宏利. 图像中被遮挡苹果的还原及定位技术研究[D]. 西北农林科技大学, 2014. [58] 谢忠红, 姬长英, 郭小清, 等. 基于改进Hough变换的类圆果实目标检测[J]. 农业工程学报, 2010, 26(7): 157–162. [59] 崔永杰, 苏帅, 吕志海, 等. 基于Hough变换的猕猴桃毗邻果实的分离方法[J]. 农机化研究, 2012(12): 166–169. [60] 张铁中, 周天娟. 草莓采摘机器人的研究:基于BP神经网络的草莓图像分割[J]. 中国农业大学学报, 2004(10): 67–77. [61] 徐丽. 基于机器视觉的菠萝果实识别与定位[D]. 华南农业大学, 2012. [62] 阳江平. 基于计算机视觉的果蔬识别方法研究[D]. 大连理工大学, 2012. [63] 朱枫. 基于计算机视觉的果蔬识别技术的研究[D]. 东南大学, 2014. [64] 包晓敏, 汪亚明. 基于最小错误率贝叶斯决策的苹果图像分割[J]. 农业工程学报, 2006, 22(5): 122–124. [65] 程玉柱, 陈勇, 车军, 等. 基于Bayes与SVM的玉米彩色图像分割新算法[J]. 江苏农业科学, 2012(7): 355–358. [66] Kuan W, Ee O, Wei L S. Comparative Study of Intel R200, Kinect v2, and Primesense RGB–D Sensors Performance Outdoors[J]. IEEE Sensors Journal, 2019, 19(99): 8741–8750. [67] 王金亮, 陈联君. 激光雷达点云数据的滤波算法述评[J]. 遥感技术与应用, 2010(5): 40–46. [68] 杨必胜, 梁福逊, 黄荣刚. 三维激光扫描点云数据处理研究进展、挑战与趋势[J]. 测绘学报, 2017(10): 1509–1516. [69] 庞晓红, 邾继贵. 视觉传感器应用中三维扫描点云数据处理的研究[J]. 传感器与微系统, 2009(3):259–268. [70] Rusu R B, Cousins S. 3d is here: Point cloud library (pcl)[C]. Robotics and Automation (ICRA), 2011 IEEE International Conference on, 2011: 1–4. [71] Meng N, Zhang J Z, Dai B Q, et al. Point Cloud Process of Laser Scanning with a Mathematical Noise Model[J]. Applied Mechanics & Materials, 2013, 275–277(1): 2537–2542. [72] Ilamchezhian J, Raj V C. An Efficient Parallel Approach for 3D Point Cloud Image Segmentation using OpenMP[J]. Digital Image Processing, 2013, 5(5):115–129. [73] Sun H, Li H. 3D Point Cloud Model Segmentation Based on K–means Cluster Analysis[J]. Computer Engineering & Applications, 2006, 42(10): 42–45. [74] YanXiong W, Feng L, Liu F, et al. Point Cloud Segmentation Using Euclidean Cluster Extraction Algorithm With the Smoothness[J]. Measurement & Control Technology, 2016(3): 795–778. [75] 杨玲香, 王田田, 何旭. 基于随机抽样一致性算法(RANSAC)的农作物行提取[J]. 江苏农业科学, 2017, 45(2): 195–197. [76] 杜廷伟, 刘波. 基于高斯混合模型聚类的Kinect深度数据分割[J]. 计算机应用与软件, 2014(12): 245–248. [77] Schnabel R, Wahl R, Klein R. Efficient RANSAC for Point–Cloud Shape Detection[J]. Computer Graphics Forum, 2007, 26(2): 214–226.. [78] 周炳南, 闵华松, 康雅文. PCL环境下的3D点云分割算法研究[J]. 微电子学与计算机, 2018, 35(6): 101–105. [79] Bertolotto, Michela, Anh–Vu, et al. Octree–based region growing for point cloud segmentation[J]. Isprs Journal of Photogrammetry & Remote Sensing, 2015, 21(4): 298–311. [80] Wang X, Zou L, Shen X, et al. A region–growing approach for automatic outcrop fracture extraction from a three–dimensional point cloud[J]. Computers & Geoences, 2017, 99(2): 100–106. [81] 谢忠红, 姬长英, 郭小清, 等. 基于改进Hough变换的类圆果实目标检测[J]. 农业工程学报, 2010, (7): 157–162. [82] Zhao Y, Gong L, Huang Y, et al. A review of key techniques of vision–based control for harvesting robot[J]. Computers and Electronics in Agriculture, 2016, 127: 311–323. [83] 马晓丹, 刘刚, 周薇,等. 基于量子遗传模糊神经网络的苹果果实识别[J]. 农业机械学报, 2013, 44(12): 227–232.,. [84] Wachs J P, Stern H I, Burks T, et al. Low and high–level visual feature–based apple detection from multi–modal images[J]. Precision Agriculture, 2010, 11(6): 717–735. [85] 赵德安, 沈甜, 陈玉, 等. 苹果采摘机器人快速跟踪识别重叠果实[J]. 农业工程学报, 2015(2): 22–28. [86] Wahabzada M, Paulus S, Kersting K, et al. Automated interpretation of 3D laserscanned point clouds for plant organ segmentation[J]. BMC bioinformatics, 2015, 16(1): 248-264. [87] Sa I, Lehnert C, English A, et al. Peduncle Detection of Sweet Pepper for Autonomous Crop Harvesting—Combined Color and 3–D Information[J]. IEEE Robotics and Automation Letters, 2017, 2(2): 765–772. [88] Rusu R B, Blodow N, Marton Z C, et al. Aligning point cloud views using persistent feature histograms[C]. IEEE/RSJ International Conference on Intelligent Robots & Systems, 2008:3384–3391. [89] 黄源, 达飞鹏, 陶海跻. An Automatic Registration Algorithm for Point Cloud Based on Feature Extraction[J]. Chinese Journal of Lasers, 2015, 42(3): 242–248. [90] An Y, Li Z, Shao C. Feature Extraction from 3D Point Cloud Data Based on Discrete Curves[J]. Mathematical Problems in Engineering, 2013(3): 87–118. [91] Rusu R B, Blodow N, Beetz M. Fast point feature histograms (FPFH) for 3D registration[C]. 2009 IEEE international conference on robotics and automation, 2009: 3212–3217. [92] Arbeiter G, Fuchs S, Bormann R, et al. Evaluation of 3D feature descriptors for classification of surface geometries in point clouds[C]. 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2012: 1644–1650. [93] Tombari F, Salti S, Di Stefano L. Unique shape context for 3D data description[C]. Proceedings of the ACM workshop on 3D object retrieval, 2010: 57–62. [94] Tombari F, Salti S, Di Stefano L. Unique signatures of histograms for local surface description[C]. European conference on computer vision, 2010: 356–369. [95] Chen G, Bui T D, Krzy?ak A. Rotation invariant feature extraction using Ridgelet and Fourier transforms[J]. Pattern Analysis and Applications, 2006, 9(1): 83–90. [96] Steder B, Rusu R B, Konolige K, et al. NARF: 3D range image features for object recognition[C]. Workshop on Defining and Solving Realistic Perception Problems in Personal Robotics at the IEEE/RSJ Int. Conf. on Intelligent Robots and Systems (IROS), 2010: 2655–2669. [97] Guo Y, Sohel F, Bennamoun M, et al. Rotational projection statistics for 3D local surface description and object recognition[J]. International journal of computer vision, 2013, 105(1): 63–86. [98] Rusu R B, Bradski G R, Thibaux R, et al. Fast 3D recognition and pose using the Viewpoint Feature Histogram[C]. 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2014:1748–1795 [99] Aldoma A, Tombari F, Rusu R B, et al. OUR–CVFH–oriented, unique and repeatable clustered viewpoint feature histogram for object recognition and 6DOF pose estimation[C]. Joint DAGM (German Association for Pattern Recognition) and OAGM Symposium, 2012: 113–122. [100] Wohlkinger W, Vincze M. Ensemble of shape functions for 3d object classification[C]. 2011 IEEE international conference on robotics and biomimetics, 2011: 2987–2992. [101] Warcho? D. Hand posture recognition using modified ensemble of shape functions and global radius–based surface descriptor[J]. Computer Science, 2018, 19(2): 115–131. [102] 陈凤东, 洪炳镕, 朱莹. 基于HSI颜色空间的多机器人识别研究[J]. 哈尔滨工业大学学报, 2004(7): 928–930. [103] Van Der Maaten L. Accelerating t–SNE using tree–based algorithms[J]. The Journal of Machine Learning Research, 2014, 15(1): 3221–3245. [104] Vishwanathan S, Murty M N. SSVM: a simple SVM algorithm[C]. Proceedings of the 2002 International Joint Conference on Neural Networks, 2002: 2393–2398. [105] Cherkassky V, Ma Y. Practical selection of SVM parameters and noise estimation for SVM regression[J]. Neural networks, 2004, 17(1): 113–126. [106] Wang J, Krudy G, Xie X Q, et al. Genetic algorithm–optimized QSPR models for bioavailability, protein binding, and urinary excretion[J]. ChemInform, 2006, 38(6): 2674–2683. [107] Kim D S, Nguyen H N, Park J S. Genetic Algorithm to Improve SVM Based Network Intrusion Detection System[C], 2005:1124–1131. [108] Ji W, Zhao D, Cheng F, et al. Automatic recognition vision system guided for apple harvesting robot[J]. Computers & Electrical Engineering, 2012, 38(5): 1186–1195. [109] Zhao C, Wu H. CNN intelligent early warning for apple skin lesion image acquired by infrared video sensors[J]. High Technology Letters, 2016, 22(1): 67–74. [110] Sabouri P, Gholamhosseini H. Lesion border detection using deep learning[C]. Evolutionary Computation, 2016:489–501. [111] Bargoti S, Underwood J. Deep Fruit Detection in Orchards[J]. Robotics and Automation (ICRA), 2016(1): 3626–3633. [112] Girshick R. Fast r–cnn[C]. Proceedings of the IEEE International Conference on Computer Vision, 2015: 1440–1448. [113] Ren S, He K, Girshick R, et al. Faster R–CNN: Towards real–time object detection with region proposal networks[C]. Advances in neural information processing systems, 2015: 91–99. [114] Dai J, Li Y, He K, et al. R–FCN: object detection via region–based fully convolutional networks[C], 2016:379–387. [115] Liu W, Anguelov D, Erhan D, et al. SSD: Single Shot MultiBox Detector[C]. European Conference on Computer Vision, 2016: 21–37. [116] Redmon J, Divvala S K, Girshick R B, et al. You Only Look Once: Unified, Real–Time Object Detection[J]. computer vision and pattern recognition, 2016(9): 779–788. [117] Redmon J, Farhadi A. YOLOv3: An Incremental Improvement[J]. Computer Vision and Pattern Recognition, 2018(1): 178–192. [118] Redmon J, Farhadi A. YOLO9000: Better, Faster, Stronger[J]. computer vision and pattern recognition, 2017(1): 6517–6525. [119] Tan W, Zhao C, Wu H. Intelligent alerting for fruit–melon lesion image based on momentum deep learning[J]. Multimedia Tools and Applications, 2016, 75(24): 16741–16761. [120] Bargoti S, Underwood J. Image Segmentation for Fruit Detection and Yield Estimation in Apple Orchards[J]. Journal of Field Robotics, 2017(1): 1039–1060. [121] 张哲远. 果园机器人自主导航关键技术研究[D], 2016. [122] Shorten C, Khoshgoftaar T M. A survey on image data augmentation for deep learning[J]. Journal of Big Data, 2019, 6(1): 60–71. [123] Alippi C, Disabato S, Roveri M. Moving convolutional neural networks to embedded systems: the alexnet and VGG–16 case[C]. 2018 17th ACM/IEEE International Conference on Information Processing in Sensor Networks (IPSN), 2018: 212–223. [124] Wu Z, Shen C, Van Den Hengel A. Wider or deeper: Revisiting the resnet model for visual recognition[J]. Pattern Recognition, 2019, 90(3): 119–133. [125] Szegedy C, Ioffe S, Vanhoucke V, et al. Inception–v4, Inception–ResNet and the Impact of Residual Connections on Learning[J]. arXiv preprint arXiv:1602.07261, 2016. [126] Deng Z, Jan Latecki L. Amodal detection of 3d objects: Inferring 3d bounding boxes from 2d ones in rgb-depth images[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2017: 5762-5770. [127] Li Y, Chen J, Ye F, et al. The improvement of DS evidence theory and its application in IR/MMW target recognition[J]. Journal of Sensors, 2016(4): 897–892. [128] Zeng D, Xu J, Xu G. Data fusion for traffic incident detection using DS evidence theory with probabilistic SVMs[J]. Journal of computers, 2008, 3(10): 36–43. [129] Hoseini F, Shahbahrami A, Bayat P. AdaptAhead optimization algorithm for learning deep CNN applied to MRI segmentation[J]. Journal of digital imaging, 2019, 32(1): 105–115. [130] Poojary R, Pai A. Comparative Study of Model Optimization Techniques in Fine–Tuned CNN Models[C]. 2019 International Conference on Electrical and Computing Technologies and Applications (ICECTA), 2019: 1–4. [131] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]. Proceedings of the IEEE conference on computer vision and pattern recognition, 2014: 580–587. [132] Guo Y , Wang H , Hu Q ,et al. Deep Learning for 3D Point Clouds: A Survey[J] IEEE TPAMI 2020, 2019: 795-806. [133] Qi C R , Su H , Mo K , et al. PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017:77-85. [134] Qi C, Yi L, Su H, et al. Pointnet++: Deep hierarchical feature learning on point sets in a metric space. Proc. Adv. Neural Inf. Process. Syst., 2017: 5099–5108. [135] Lun Z , Gadelha M , Kalogerakis E , et al. 3D Shape Reconstruction from Sketches via Multi-view Convolutional Networks[J]. IEEE, 2017:945-953. [136] Wang P S , Liu Y , Guo Y X , et al. O-CNN: Octree-based Convolutional Neural Networks for 3D Shape Analysis[J]. Acm Transactions on Graphics, 2017, 36(4):72. [137] Riegler G , Ulusoy A O , Geiger A.OctNet: Learning Deep 3D Representations at High Resolutions[C]. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2017:3577-3586. [138] 郭军. 多传感器融合目标识别关键技术研究[D]. 国防科学技术大学. [139] Chen X, Ma H, Wan J, et al. Multi-view 3d object detection network for autonomous driving [C]. In IEEE CVPR. 2017: 3. [140] Ku J, Mozifian M, Lee J, et al. Joint 3d proposal generation and object detection from view aggregation [C]. In 2018 IEEE/RSJ international Conference on IntelligentRobots and Systems (IROS). 2018: 1–8. [141] Liang M, Yang B, Wang S, et al. Deep continuous fusion for multi-sensor 3dobject detection [C]. In Proceedings of the European Conference on Computer Vision(ECCV). 2018: 641–656. [142] Zeng Y, Hu Y, Liu S, et al. RT3D: Real-time 3-d vehicle detection in lidar point cloud for autonomous driving [J]. IEEE Robotics and Automation Letters. 2018, 3 (4):3434–3440. [143] Qi C R, Liu W, Wu C, et al. Frustum pointnets for 3D object detection from rgb-d data [C]. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 918–927. [144] Xu D, Anguelov D, Jain A. Pointfusion: Deep sensor fusion for 3d bounding box estimation [C]. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 244–253. [145] Richard. Computer Vision: Algorithms and Applications [M]. Springer, 2010. [146] Zhou Y, Tuzel O. Voxelnet: End-to-end learning for point cloud based 3d object detection [C]. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2018: 4490–4499. [147] Simon M , Milz S , Amende K, et al. Complex-YOLO: An Euler-Region-Proposal for Real-Time 3D Object Detection on Point Clouds[C], Springer. 2018:2106-2113. |
中图分类号: | S23 |
开放日期: | 2022-01-11 |