A two-view high-speed synchronous vision system was developed based on high-speed cameras. It was used to monitor synchronously periodic mechanical motion in non-contact and long term modes from different views. It could automatically judge the abnormal motion and could record high-speed vision images at a moment. On the basis of extraction of motion characteristics of a mechanical system, an offline processing algorithm of multi-view high frame rate video image was proposed. Two high-speed cameras were used to accurately establish the sample image database and their relation. Then, a multi-angle synchronous real-time detection algorithm was proposed to quickly complete the real-time visual detection of the targets, meanwhile recording key images instantaneously. To confirm the validity and stability of algorithms, the two-angle-view high-speed vision monitor system was built in laboratory, and real mechanical parts were detected in experiments. The experimental results show that the vision monitor system achieves a high-speed synchronous detection of 500fps, the validity judgment of abnormal motion happened in 0.1 s, and the recording instantaneously for key images. The research working provide a higher speed visual detection method for the noncontact detection of mechanical systems.