Design of and research on the robot arm recovery grasping system based on machine vision

With the development of urban modernization, the amount of generated waste has been constantly increasing, making waste classification necessary. In the process of waste bin recycling, the main challenge is improving recycling efficiency and reducing the workload of workers. To address the problems of waste bin positioning and retrieval in the waste bin recycling process, this study proposes an automatic retrieval system based on a combination of machine vision and robotic arm motion control. The main aim is to achieve accurate and efficient detection, recognition, and retrieval of different types of waste bins. First, the YOLOv5 deep learning recognition algorithm is improved using a channel pruning technique to reduce the complexity of the model while ensuring high recognition accuracy, thus facilitating the portability and deployment of the model on various mobile devices. Then, image preprocessing is conducted using the median filtering method and the Gamma brightness correction algorithm. The HSV color model is employed, and the H component distribution is used for classifying different types of waste bins under different lighting conditions. This allows for image segmentation for different-color waste bins, facilitating the classification and recognition of waste bin images. Finally, the waste bin localization algorithm and robotic arm motion algorithm are employed to accomplish the positioning and retrieval of waste bins. The experimental results indicate that compared to the original YOLOv5 model, the improved YOLOv5 algorithm can achieve a significant reduction in parameter number, decreasing it from 7,022,326 to 2,828,675, which represents an approximately 60 % decrease. Moreover, with a marginal 0.2 % decrease in accuracy, the FLOPs value decreases from 12.9G to 7.97G, demonstrating a reduction of nearly 70 %. The model size is also reduced by almost 60 %. The results indicate that the recognition rates of different-color waste bins exhibit a trend of initially increasing and then decreasing with the intensification of light. Among the four colors of waste bins, the recognition rate of red waste bins is the highest, with an average recognition rate of 95 %. In contrast, orange waste bins have the lowest average recognition rate, with an average value of 91 %. In the grasping experiments, the detection and grasping success rates for the red waste bins are the highest, reaching 95 % and 80 %, respectively. Those of the blue waste bins are the next highest, with detection and grasping success rates of 85 % and 80 %, respectively. Finally, the detection and grasping success of orange waste bins are 80 % and 75 %, respectively.