Container keyhole positioning based on deep neural network

A large-scale container dataset and a baseline method for container hole localization

“…However, regardless of the one-stage or two-stage object detection model, the model's output is a bounding box containing the object but cannot represent the object contour. Accordingly, simply adaption of detection techniques to container detection will suffer from coarse container keyhole localization [10].…”

“…Benefiting from the ability of deep neural networks, Lee et al [9] apply the LSTM model [31] to improve the detection efficiency. [10] also gets a similar real-time performance through using YOLO algorithm [16]. All existing deep learning-based methods focus on detecting corner casting.…”

“…Convolutional Neural Networks (CNN), as one of the most popular deep learning architectures, have been proven effective in many computer vision applications [6]. Automatic container hole detection using computer vision and CNNs has also recently drawn attention [7][8][9][10]. Benefiting from advances in object detection [11], detection-based methods [9,10] can detect the bounding box of the corner casting or keyholes with high precision, but it is actually hard to apply such methods in the container terminal.…”

“…Automatic container hole detection using computer vision and CNNs has also recently drawn attention [7][8][9][10]. Benefiting from advances in object detection [11], detection-based methods [9,10] can detect the bounding box of the corner casting or keyholes with high precision, but it is actually hard to apply such methods in the container terminal. This is because the container can only be grabbed when the lock on the spreader falls accurately into the lock hole of the container.…”

A vision-based container position measuring system for ARMG

“…12 In this reported work, the detection rate of the neural network reached 94.57%, while the Intersection-Over-Union (IOU) reached 87.79%. Li et al 13 proposed another container corner detection method based on YOLO, 14 while the IOU reached 80.43%. The accuracy of CNN based positioning method is insufficient.…”