Automatic abstraction of laparoscopic medical footage through visual activity analysis

In this paper we propose an automatic video highlighting approach which is based on the assessment of visual activity (i.e. temporal visual changes). The aim is to highlight major key-points in laparoscopic medical interventions. Visual activity is determined using a frame-based histogram difference approach. Preliminary tests show that during critical moments in laparoscopic intervention, camera tends to stand still, as it is focused on the local area of intervention (the only changes tend to be local, e.g. triggered by grabbing, cutting, stitching, etc.). Therefore, the proposed abstract will include the video segments with a reduced visual rhythm, i.e. the amount of visual changes occurring in a certain time window is below a predefined threshold. Experimental tests conducted on more than 13 hours of footage prove the potential of this approach to the automatic abstraction of medical videos.