1 00:00:00,330 --> 00:00:07,140 Welcome back to the course, so the next chapter we take a look at is it involves object tracking by 2 00:00:07,140 --> 00:00:07,570 color. 3 00:00:07,950 --> 00:00:10,620 So let's open this notebook already have it open here. 4 00:00:11,220 --> 00:00:14,160 So this for this, I'll give you a warning. 5 00:00:14,400 --> 00:00:21,870 This lesson isn't it isn't well-done, mainly because I couldn't find a test video to use in club that 6 00:00:21,870 --> 00:00:23,010 actually worked fairly well. 7 00:00:23,400 --> 00:00:27,420 I should have just recorded a video with my webcam to demonstrate this, which I might do later on. 8 00:00:27,960 --> 00:00:33,840 However, for now, I tried to find a video where we can filter on the color yellow and track a yellow 9 00:00:33,840 --> 00:00:38,700 BMW in the see in a scene and demonstrate the track in a color. 10 00:00:39,690 --> 00:00:45,360 However, it doesn't look entirely perfectly to well, it does with briefly for a moment so we can demonstrate 11 00:00:45,360 --> 00:00:45,690 that. 12 00:00:46,290 --> 00:00:47,690 So let's luto images. 13 00:00:47,700 --> 00:00:50,400 Actually, this is the image for the video. 14 00:00:50,820 --> 00:00:52,170 You'll probably not see this. 15 00:00:53,730 --> 00:00:55,410 We can delete this box here. 16 00:00:56,040 --> 00:00:57,360 So this is the code here. 17 00:00:57,360 --> 00:00:59,100 So all this stuff is good. 18 00:00:59,340 --> 00:01:01,410 Well, let me show you the output of this goodness. 19 00:01:02,010 --> 00:01:05,340 This is a fairly long involve code, but it's actually not that complicated. 20 00:01:06,060 --> 00:01:12,120 And this is the output video here where truly we track BMW hopes to go back to this. 21 00:01:13,170 --> 00:01:18,030 So the video resolution is a bit high, so it doesn't fit the whole thing on screen. 22 00:01:18,630 --> 00:01:24,270 So initially, we initialize a random spot here and it's tracking some objects, something hidden in 23 00:01:24,270 --> 00:01:25,170 the film's colorful, too. 24 00:01:25,620 --> 00:01:31,020 However, luckily it does grab on to the yellow car here, and you can see it follows this yellow call. 25 00:01:31,200 --> 00:01:32,070 That's right here. 26 00:01:32,490 --> 00:01:33,990 But then it loses it. 27 00:01:35,160 --> 00:01:36,450 I'm not even sure where here it is. 28 00:01:36,960 --> 00:01:40,380 It loses it for a bit and then tracks some random objects here. 29 00:01:40,930 --> 00:01:44,280 Now, this can probably be improved by improving the HSV color filter. 30 00:01:44,730 --> 00:01:50,280 But what we're actually doing here is we're filtering on a color. 31 00:01:50,280 --> 00:01:55,770 The color filtering as yellow sits between 20 and 40 and HSV each. 32 00:01:55,920 --> 00:02:00,340 I should say the hue and what we're doing, we're creating an empty area for points. 33 00:02:00,360 --> 00:02:04,430 Now this is the points that are being tracked so that you can see that line. 34 00:02:04,440 --> 00:02:06,390 There's a historical line of tracking points. 35 00:02:06,900 --> 00:02:09,060 We just keep appending that to the disarray. 36 00:02:09,930 --> 00:02:15,210 We create our video writer and then all of let's move into the core of the code here. 37 00:02:15,690 --> 00:02:18,240 So like I said, we are tracking one color. 38 00:02:18,360 --> 00:02:20,850 So remember the HSV method to filter and color? 39 00:02:21,300 --> 00:02:22,170 That's what we're doing. 40 00:02:22,170 --> 00:02:25,620 And then after that, we're running fine contours on that output. 41 00:02:26,040 --> 00:02:31,950 So fine contours should only be able to, in theory, fill a get a box. 42 00:02:31,950 --> 00:02:35,700 The largest box around one of the yellow objects on screen. 43 00:02:35,850 --> 00:02:37,590 That's how we have it set here. 44 00:02:39,000 --> 00:02:42,180 So we are creating them to read a story to centroid. 45 00:02:42,180 --> 00:02:43,700 That's the center of the ground control. 46 00:02:44,340 --> 00:02:49,560 So that's how we get contours for filter contours at the center of the control. 47 00:02:50,040 --> 00:02:52,020 If we have contours, this is what we do. 48 00:02:52,410 --> 00:02:55,640 Get the largest contour, get the area, the radius. 49 00:02:55,650 --> 00:02:57,720 That's a minimum closing circle for the contour. 50 00:02:57,750 --> 00:03:04,170 It's a function we haven't discussed before, but basically it takes an input of a cognitive control 51 00:03:04,770 --> 00:03:08,280 and gives you the ex-wife position of the center and the radius. 52 00:03:08,880 --> 00:03:10,800 Then we get the moments here with him. 53 00:03:11,310 --> 00:03:18,300 So this is how we actually get the and using movements, we get the center of the circle as well. 54 00:03:18,660 --> 00:03:22,530 So we can actually use that to plot the circle, plot the point here. 55 00:03:23,370 --> 00:03:25,020 That's how we actually get the trail. 56 00:03:25,710 --> 00:03:31,560 And if the radius is of a 25, which means if it's over a certain size, that's when we actually do 57 00:03:32,460 --> 00:03:33,420 grow in the circle. 58 00:03:33,430 --> 00:03:33,870 So we don't. 59 00:03:33,870 --> 00:03:36,780 We were drawing a circle at a point, a circle around the point. 60 00:03:36,780 --> 00:03:38,880 So one is supposed to be the center of the object. 61 00:03:38,910 --> 00:03:42,220 One is the little circle in closing the object. 62 00:03:43,050 --> 00:03:46,680 And then this is the area here where we just keep track of our points. 63 00:03:46,710 --> 00:03:51,600 So once the radius is over 25, we just store all the points here and then draw a line. 64 00:03:52,020 --> 00:03:55,110 So that line is basically the historical tracking. 65 00:03:55,770 --> 00:03:58,080 So this is a pretty cool system, pretty cool. 66 00:03:58,090 --> 00:03:58,950 Could, I should say. 67 00:04:00,180 --> 00:04:06,690 So we took the output that it produces from our trade frame converted into an empty before then played 68 00:04:06,690 --> 00:04:10,350 back here and we get the output video, which you saw previously. 69 00:04:10,980 --> 00:04:12,170 So that's just video here. 70 00:04:12,420 --> 00:04:14,330 So we can actually play it. 71 00:04:14,340 --> 00:04:14,970 Yeah, there we go. 72 00:04:14,970 --> 00:04:15,600 It's playing. 73 00:04:16,440 --> 00:04:18,000 So that's it for this lesson. 74 00:04:18,090 --> 00:04:24,900 What is a good exercise for you to do is if you want or if you do have open TV installed, you can run 75 00:04:24,900 --> 00:04:29,160 this locally and you can use a webcam and tweak these parameters. 76 00:04:29,520 --> 00:04:33,870 It's going to be much easier to tweak it in a live video running the court locally than on collab. 77 00:04:33,870 --> 00:04:41,100 And that's one of the disadvantages of collab in in terms of working with webcam video, but otherwise 78 00:04:41,400 --> 00:04:43,740 collab is a godsend for so many things. 79 00:04:44,490 --> 00:04:46,290 So I hope you enjoy this lesson. 80 00:04:46,290 --> 00:04:53,040 And then we'll move on now to the 27th lesson, which gets into a very cool topic, which is facial 81 00:04:53,040 --> 00:04:55,830 unlock detection using the dylib library. 82 00:04:56,460 --> 00:04:58,170 So I'll see you in the next lesson. 83 00:04:58,290 --> 00:04:58,770 Thank you.