Research on variational optical flow method for rockfall monitoring

Rockfall is a global mountain disaster problem, which has the characteristics of wide distribution, strong randomness, and inconspicuousness. The contacted monitoring method is difficult to deploy. Simultaneously, the non-contact detection is easily affected by complex environments and generally has less robustness. This paper presents a rockfall monitoring technology based on machine vision. First, a high-resolution industrial camera continuously acquires images of the monitoring area. Then, we perform a series of operations on the landslide rockfall images, including image preprocessing, image optical flow algorithm preprocessing to identify the features of the rockfall. Finally, a dynamic target detection experiment is carried out to verify the feasibility of the image optical flow algorithm. Experiments of rockfall monitoring verify the effectiveness and generalization abilities of the algorithm. From the results: (1) The features of the image data in this paper are clear, and the recognition range covers the whole visual field. (2) The technology can capture the motion characteristics of landslide rockfall in different scenes. (3) The processing efficiency of the algorithm can reach 10 frames per second. Thus, this technology can be used in a monitoring system for rockfall.