 Hi everyone, I am Megh from video team. So this is a part of the diving and come gymnastics course that's on EDX, IIT Bombay X. So here I have some videos processed of diving. The left half is from the original diving video. We have extracted the diver based on HSU filtration on skin color and so but as you can see here the camera is moving. We need to compensate for the motion of the camera. So we have utilized optical flow Lucas Canade algorithm for finding the motion in the camera and that's the water disturbance you can see on the right hand side because water disturbance is one of the major criteria for judging the performance of a diving. Right now we have attempted assessment on the basis of water disturbance but that's not the sole criteria. Work needs to be done but due to lack of time and business knowledge we could not attempt the whole. You have used the color. Suppose the background color matches with the body color what will happen? Sir in some cases yes it creates a problem but when we need to get both the motion and the color. No suppose the background color and the foreground color the body color is same then you will not get the image itself. There will be no motion available because your basic is color skin color. Sir whole background while diving cannot be maybe the wall is painted into the same color then. Sir then yes there will be a problem. So sir what we can do is in future there is also if we will analyze two consecutive frames we can get an idea of how much of pixel is moving between consecutive frames and the background pixels will move only a little but the pixels of that person will move more so we can estimate from that by using color plus motion we can get better. We can get an idea as to where the diver is going. Yes sir. Yes sir. Why he is still and So that's why we need to combine both the color as well as the motion. Sir I am also calculating the amount of rotation the number of somersaults by the diver. Myself Patel Basudev from SV90 I was trying assessment in field of gymnastics because what we want is something which is very technical. We were not able to find some applications in which highly it is technical it mostly aesthetics and grace and all that is involved so because it is means it is very difficult to handle all that so we somehow thought that in gymnastics there is some technical stuff so we can assess something so first of all I tried with pommel horse so first step was to remove the background and do some segmentation to get the person so we can do further processing on that so I was able to get these results but the problem now is the for next step what you have done here image cropping or something no it beyond the basis of motion it is estimating and how you get the read of the background how you get the read of the background I have taken difference between the frames okay you didn't do the image thresholding thresholding and all are the intermediate steps so if you do the image thresholding automatically the background will be wiped out no but sir the background is not exactly still it is moving so if we'll directly do thresholding it will not work I have tried a lot of things in doing this so but the next step here is to extract the characteristics based on how judging is done in this and we don't exactly know how judging is done and if we know then also it is very much difficult to extract the parameters because image processing technologies are not that advanced to detect the posture of person from video so we have to left this before the assessment thing in rings we thought that we can do some assessment because in it the grading right idea is based on how much longer and how much steady you can be in some positions which require a lot of strength it is just for men so we thought we can do something by pixel difference and also we tried assessment in this so this is just I am taking a difference I did some filtering to remove the frames from the video which contain the intermediate flips and all because they are not used in assessments what is used in assessments is how much he can hold for in some positions like this so I tried some filtering and only captured those frames in which the athlete is still and then by calculating the difference in the position between the frames we can estimate the score but I tried the scoring but it was not giving good results and we have to add more heuristic and better go with them in order to grade we use now algorithm but it was not giving proper results rings