高级检索

基于视觉实时引导的煤矸石精准跟踪方法

Vision-based real-time precise tracking method for gangue

  • 摘要: 现有煤矸分拣机器人在分拣煤矸石时存在误抓取、空抓、碰撞等问题,其主要原因是煤矸石随输送带运输过程中存在打滑、跑偏等现象,依靠带速的煤矸石跟踪方法难以实时获取其精准位姿信息,导致机械臂抓取时出现较大误差,影响机器人分拣效率。针对该问题,提出一种基于视觉实时引导的煤矸石跟踪方法,即通过相机获取煤矸石实时位姿信息,引导机械臂调整动作完成煤矸石跟踪抓取。首先,通过视觉识别模块获取待抓取目标初始位姿与跟踪模板,由控制系统进行策略分配,将煤矸石分配给对应机械臂进行抓取;当目标煤矸石进入机械臂抓取工作区后,由基于孪生网络构建的单目标跟踪模型获取煤矸石实时位姿信息,并实时调整机械臂动作,完成抓取。最后,对不同带速下的煤矸石进行视觉跟踪实验,并构建煤矸分拣机器人仿真系统完成不同程度打滑、跑偏工况的煤矸石跟踪轨迹规划仿真。仿真实验结果表明,构建的煤矸石跟踪模型跟踪准确率为96.9%,跟踪速度为39 FPS,满足实时引导的需求。当存在不同程度打滑、跑偏时,基于视觉实时引导的机械臂抓取误差均降低至1 mm以内。相较于基于带速的跟踪方法,可有效消除运输过程中由于输送带打滑、跑偏等带来的累积误差,提高系统实时响应能力,进一步提升煤矸石分拣效率。

     

    Abstract: Existing coal gangue sorting robots face challenges in mispicking, empty grabs, and collisions during the sorting process. These issues primarily stem from the phenomena of slippage and deviation of coal gangue during belt transportation, making it difficult to accurately acquire real-time positional information of the gangue using belt speed-based tracking methods. This leads to significant errors in the robotic arm’s grabbing actions, adversely affecting the sorting efficiency. To address this issue, a vision-based real-time guidance method for coal gangue tracking is proposed. This method involves capturing the real-time positional information of the coal gangue through cameras, guiding the robotic arm to adjust its actions for tracking and grabbing the gangue. Initially, the vision recognition module acquires the target's initial position and tracking template, with the control system allocating strategies to assign the gangue to the respective robotic arm for grabbing. Once the target gangue enters the robotic arm's grabbing work area, a monocular target tracking model, built on Siamese networks, captures the real-time positional information of the gangue, enabling the arm to adjust its movements dynamically to complete the grab. Subsequently, visual tracking experiments on coal gangue at different belt speeds were conducted, and a coal gangue sorting robot simulation system was developed to perform trajectory planning simulations under various conditions of slippage and deviation. The simulation experiment results demonstrate that the constructed coal gangue tracking model achieves a tracking accuracy of 96.9% and a tracking speed of 39FPS, meeting the needs for real-time guidance. When facing varying degrees of slippage and deviation, the grabbing error of the robotic arm, guided by real-time vision, is reduced to within 1 mm. Compared to the belt speed-based tracking method, this approach effectively eliminates cumulative errors caused by belt slippage and deviation during transportation, enhances the system’s real-time responsiveness, and further improves the efficiency of coal gangue sorting.

     

/

返回文章
返回