Vehicle door frame positioning method for binocular vision robots based on improved YOLOv4

Abstract

In the work of using robots to grasp end-of-life cars, the position of the vehicle door frame needs to be grasped. Fast and accurate positioning of the vehicle door frame is the key to realize the automatic car grasping process. Traditional methods for locating and grasping scrap cars rely heavily on manual operation and suffer from low grasping efficiency and poor accuracy. Therefore, this paper proposes a binocular vision robot vehicle door frame spatial localization method based on the improved YOLOv4. A lightweight and efficient feature fusion target detection network in complex environments is proposed in the method, and the target detection results are combined with an improved SURF feature matching method to locate the vehicle door frame position. To simplify the network structure, MobileNetv3 is used instead of the backbone network CSPDarknet53, and deep separable convolution is used in the network. Its to improve the sensitivity of the network to vehicle door frame targets in complex environments, an improved convolutional block attention module is added to PANet, and adaptive spatial feature fusion (ASFF) is introduced into the network to make full use of features at different scales for more effective feature fusion. Compared with YOLOv4, the number of network parameters is reduced by 73.8%, the mAP is improved by 1.35%, and the detection speed is increased by 28.7%. The experimental results show that the positioning accuracy of the system is 0.745 mm, which meets the positioning measurement error of less than 1 cm required for the vehicle door frame. The paper also compares with other network models and the results show that the method achieves a good balance between detection speed and detection accuracy, satisfying the task of identifying vehicle door frames in complex environments with good detection results.

Enhanced curve-based segmentation method for point clouds of curved and irregular structures

“…The traditional three-frequency, six-step phase unfolding algorithm is slow, inaccurate, and relatively influenced by the environment due to ambient light and reflected light from the object's surface [22][23][24][25][26]. The industrial field is the intended application scenario for the algorithm proposed in this paper.…”

TAFFNet: Two-Stage Attention-Based Feature Fusion Network for Surface Defect Detection

“…To further validate the effectiveness of the YOLOv5-LW method proposed in this paper, network models with different target detection algorithms were trained on a homemade plank dataset. To ensure the reliability of the experimental data, the same hyperparameters were used to train the YOLOv5s [21], YOLOv4 [5], YOLOv3 [23], SSD [38] and Faster RCNN [39] networks. Figure 7 shows the PR (Precision Recall) curve for the YOLOv5-LW method, and the evaluation index results are shown in Table 6.…”

“…The experimental results in Table 6 show that the two-stage target detection algorithm Faster Rcnn [39], compared to the single-stage target detection algorithm YOLOv4 [5] with a larger number of parameters and computation, has a much higher detection accuracy, but the YOLO series has a great improvement in detection speed and a number of parameters, so it is more utilised for network deployment to embedded devices for real-time detection of wood panel defects. The YOLOv5-LW method proposed in this paper outperforms the other five models, which, in comparison, have weaker generalisation capabilities, slower detection speed, and a large number of parameters that are not easily deployed.…”

“…Wood panel defect detection [1][2][3][4][5] is very important in this day and age. Especially in the decoration industry [6], wood plays a direct role as a very important raw material and its quality is good or bad.…”

“…Wu et al [24] used YOLOv4 deep learning network in agriculture to detect apple blossoms and achieved good results. Song et al [5] improved YOLOv4 for spatial positioning of car door frames and achieved a positioning error of less than 1cm. In addition, it is also widely used in the field of wood panelling.…”