Multi-view stereo vision scene stream-based strenuous exercise detection method

Abstract

The invention provides a multi-view stereo vision scene stream-based strenuous exercise detection method which comprises the following steps: I: acquiring multiple groups of image sequences by using a calibrated multi-view camera; II: preprocessing the image sequences; III: designing a scene stream energy function data item; IV: designing a scene stream energy function smoothing item; V: optimally solving an energy function, and performing calculation with a calculation model from an image with the lowest resolution in an image pyramid obtained in the step II; VI: clustering a scene stream exercise region; VII: constructing an exercise direction discrete degree estimation model to judge whether an exercise is a strenuous exercise; VIII: constructing an exercise region kinetic energy size estimation model; and XI: setting a threshold value, and triggering an alarm when continuous n frames meet an estimation condition. By adoption of multi-view stereo vision scene stream-based estimation, the multiple groups of image sequences of the same scene are acquired by using the calibrated multi-view camera, and the strenuous exercise can be effectively detected by using a three-dimensional scene stream.

Description

technical field [0001] The invention relates to a violent motion detection method, in particular to a violent motion detection method based on multi-eye stereo vision scene flow. Background technique [0002] With the high development of today's scientific and technological information technology, especially human beings have made breakthroughs in computer vision and artificial intelligence, many tasks that should be done by humans can be done by computers. For example, video surveillance, the most common method of operation is for people to observe the monitoring display and then respond to abnormal events that occur. Since people cannot concentrate for a long time to monitor all the events happening in the video, the phenomenon of missed alarms inevitably occurs. Therefore, it is particularly important to use a computer to process video frames and determine whether an abnormal event occurs. [0003] Under normal circumstances, the camera of video surveillance is fixed in...

Images