Mechanism Action Detection Based on YOLO

The field of mechanism motion detection has significant research value in the fields of mechanism motion characteristic analysis, information collection, and mechanism fault prevention and diagnosis. The traditional method of detecting the motion of mechanical mechanisms relies on contact sensors to transmit the execution of the motion and reflect the motion status of the mechanism. The traditional contact type mechanism still faces the following challenges in motion detection: (1) under certain operating conditions, the mechanism does not allow the installation of contact type sensors for motion detection; (2) The mechanism action has a measurement dead angle and cannot obtain the execution status of the specified position action; (3) Built in contact sensors often require a pre designed structure to accommodate the sensor, which can make it difficult to temporarily change the measurement method. This article proposes a binocular vision mechanism motion detection technology based on YOLO v5, which utilizes the YOLO v5 algorithm combined with binocular vision technology to achieve motion detection. YOLO v5 obtains feature pixel plane position information by prefabricating feature information. This article utilizes information such as internal parameters and focal length of binocular cameras to obtain spatial three-dimensional coordinates through the similarity triangle method. Ultimately, this method can obtain the motion of the mechanism. The experimental results of YOLO v5 are based on a training method from scratch, without using pre trained weights. This detection method was tested on Nvidia's Jetson Xavier NX system and achieved a detection speed of 30 frames per second.