MGBM-YOLO: a Faster Light-Weight Object Detection Model for Robotic Grasping of Bolster Spring Based on Image-Based Visual Servoing

Liu, Huanlong; Li, Dafa; Jiang, Bin; Zhou, Jianyi; Tao, Wei; Yao, Xinliang

doi:10.1007/s10846-021-01564-2

Cited by 10 publications

(9 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It is necessary to be consistent with this feature when building a new backbone network of YOLOv3. In addition, when the input image dimension is 416 2 3 3, the YOLOv3 original backbone network will output three preliminary effective feature layers, whose dimensions are 52 2 3 256, 26 2 3 512, and 13 2 y¨C ¸1024 respectively. For the GhostNet backbone network, the three effective feature layer dimensions of its output are 52 2 3 40, 26 2 3 112, and 13 2 3 160.…”

Section: Network Architecture Of Ghostnet-yolov3mentioning

confidence: 99%

“…In addition, when the input image dimension is 416 2 3 3, the YOLOv3 original backbone network will output three preliminary effective feature layers, whose dimensions are 52 2 3 256, 26 2 3 512, and 13 2 y¨C ¸1024 respectively. For the GhostNet backbone network, the three effective feature layer dimensions of its output are 52 2 3 40, 26 2 3 112, and 13 2 3 160. Therefore, when building the new backbone network of YOLOv3, we should also consider correctly adjusting the number of input channels when accessing the next volume layer, to ensure the correctness of the network.…”

Section: Network Architecture Of Ghostnet-yolov3mentioning

confidence: 99%

“…With the rapid development of deep learning object detection technology and its in-depth application in the industrial field, 2–4 YOLO, 5 SSD 6 and their related improvement algorithms have relatively fast detection speeds and are suitable for practical applications such as real-time detection. RetinaNet 7 and Tiny RetinaNet 8 borrow the ideas of Faster R-CNN and multiscale object detection by reshaping the Focal Loss loss function to solve the problem of positive and negative sample class imbalance in the training samples of previous detection models during the training process.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A visual positioning method for wedge support robot based on Pruned-Ghost-YOLOv3

Tao

Liu

Tian

et al. 2023

Proceedings of the Institution of Mechanical Engineers, Part C:

Self Cite

View full text Add to dashboard Cite

The wedge support robot is the core module of the intelligent dismantling equipment for the bolster spring and wedge of railroad wagon. In this paper, to address the problem of accurate positioning of bolster hole with abnormal geometry and large error in the complex background, a light-weight object detection model is proposed based on GhostNet and model pruning algorithm: Pruned-Ghost-YOLOv3. The proposed models and the original YOLOv3 model are trained and evaluated on the bolster hole custom dataset. The results show that the improved Pruned- Ghost-YOLOv3 model is smaller, has fewer parameters and is faster compared to the original YOLOv3 model. To meet the requirements of operational efficiency and reliability, a precise positioning algorithm based on the bounding box of the bolster hole and an indirect positioning process are proposed in this paper. The comparative experiments of bolster hole positioning and wedge support are conducted on the wedge support robot. The experimental results show that the proposed Pruned-Ghost-YOLOv3 model can meet the requirements of rapid detection and positioning of the bolster holes, and the proposed precise positioning algorithm and robot positioning process have high accuracy and reliability. The research results are successfully applied to the development of a railroad wagon wedge support robot, and it has a good reference value for the development of the intelligent positioning system of heterogeneous workpiece operation robots in the complex background of related industries.

show abstract

Section: Network Architecture Of Ghostnet-yolov3mentioning

confidence: 99%

Section: Network Architecture Of Ghostnet-yolov3mentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A visual positioning method for wedge support robot based on Pruned-Ghost-YOLOv3

Tao

Liu

Tian

et al. 2023

Proceedings of the Institution of Mechanical Engineers, Part C:

Self Cite

View full text Add to dashboard Cite

show abstract

“…Recently, object detection NN based on the YOLO (You Only Look Once) algorithm have been applied for mobile robot visual navigation [25]- [28]. In [29] is introduced MGBM-YOLO, an application for visual servoing controller with object detection NN, where the authors propose two YOLOv3 models that are applied to the robotic grasping system of bolster spring based on image-based visual servoing. The MGBM-YOLO visual servoing controller presents a depth estimator depending on the actual area and the desired area of the object's bounding box, therefore, the system must know the object's dimensions beforehand.…”

Section: Introductionmentioning

confidence: 99%

ViKi-HyCo: A Hybrid-Control Approach for Complex Car-Like Maneuvers

Velasco-Sánchez,

Ángel Muñoz Bañón,

Candelas

et al. 2024

IEEE Access

View full text Add to dashboard Cite

While Visual Servoing is deeply studied to perform simple maneuvers, the complex cases where the target is far out of the camera field of view (FOV) during the maneuver are not common in the literature. For this reason, in this paper, we present ViKi-HyCo (Visual Servoing and Kinematic Hybrid-Controller). This approach generates the necessary maneuvers for the complex positioning of a non-holonomic mobile robot in outdoor environments. In this method, we use LiDAR-camera fusion for automatic target calculation. The multi-modal nature of our target representation allows us to hybridize the visual servoing with a kinematic controller. In this way, we can perform complex maneuvers even when the target is far away from the camera's FOV. The automatic target calculation is performed through object localization for outdoor environments, which estimates the spatial location of a target point for the kinematic controller and allows the dynamic calculation of a desired bounding box of the detected object for the visual servoing controller. The proposed approach does not require an object-tracking algorithm and can be applied to any visually tracking robotic task where its kinematic model is known. ViKi-HyCo has an error of 0.0428 ± 0.0467 m in the X-axis and 0.0515 ± 0.0323 m in the Y-axis at the end of a complete positioning task.INDEX TERMS autonomous robots, domestic waste localization, hybrid-control, outdoor environment, sensor fusion, visual servoing.

show abstract

“…To address the issue of accurate position and classification, the use of fusion methods is required. Liu et al [15] proposed an improved YOLO algorithm for the scratch of spring in the industrial servo system. By changing the lightweight structure of the main network to the MobileNetv3 and GhostNet, the feature extraction efficiency of the main network is improved, the visual servo system in a single phase is further optimized.…”

Section: Introductionmentioning

confidence: 99%

RP-YOLOX-DL: a deep learning hybrid method for parallel robots target positioning

Zhang

Wang

et al. 2023

Meas. Sci. Technol.

View full text Add to dashboard Cite

To address the issues of low classification pickup efficiency and slow response time of parallel robots positioning methods based on machine vision, this paper proposes a deep learning hybrid method RP-YOLOX-DL (YOLOX-Deeplabv3+ method for robots positioning) for parallel robots to achieve accurate picking. Firstly, the RP-YOLOX lightweight network is used to complete target recognition classification and rough positioning. A new feature enhanced network called DS-PANet is proposed to optimize the original up-and-down sampling structure, and the computational efficiency is improved through the attention mechanism and deep convolution. The loss function in network evaluation is enhanced, and an emphasizing the target Binary CrossEntropy loss function (ETBCE) is proposed for the objective loss strategy. Secondly, the Deeplabv3+(DL) network is used and the pooling structure is improved to obtain rich multi-scale information by using different types of Atrous convolution. Then, the extracted semantic segmentation center coordinates are finely positioned, and by using a hybrid positioning strategy, the RP-YOLOX and DL network modules are combined to obtain the best positioning coordinates of the target. Finally, a hand-eye calibration operation was performed to convert the robot, the camera, and the conveyor belt to eye-to-hand coordinate computation. The experimental results indicate that the hybrid method achieves a pick-up rate of 92.56% and a response time of 2.357s, showing better performance than the traditional algorithms Faster-RCNN, YOLOv3, and YOLOv5. Meanwhile, the identification efficiency is increased by 2.41% compared to the YOLOX algorithm. These results verify the efficiency and robust adaptability of the mixed method. This study has a certain reference value for applying the deep learning method to robots positioning pick-up.

show abstract

MGBM-YOLO: a Faster Light-Weight Object Detection Model for Robotic Grasping of Bolster Spring Based on Image-Based Visual Servoing

Cited by 10 publications

References 22 publications

A visual positioning method for wedge support robot based on Pruned-Ghost-YOLOv3

A visual positioning method for wedge support robot based on Pruned-Ghost-YOLOv3

ViKi-HyCo: A Hybrid-Control Approach for Complex Car-Like Maneuvers

RP-YOLOX-DL: a deep learning hybrid method for parallel robots target positioning

Contact Info

Product

Resources

About