1 00:00:02,029 --> 00:00:07,550 In this video, Olga writes a program which converts this video. 2 00:00:11,740 --> 00:00:13,390 Into this. 3 00:00:16,590 --> 00:00:22,860 So as you can see in the article, the video of the face of the person speaking is blurred. 4 00:00:23,670 --> 00:00:25,530 So this is a technique I'm going to use. 5 00:00:25,530 --> 00:00:31,140 I'm going to blur the area around the face so that it's not visible. 6 00:00:31,320 --> 00:00:38,430 You can also use other techniques, such as writing a rectangle, which is filled in. 7 00:00:39,360 --> 00:00:42,040 But this is a more standard practice. 8 00:00:42,050 --> 00:00:49,080 So let me show you how I can do this using, of course, open CV. 9 00:00:49,410 --> 00:00:52,380 So import quota and load the video. 10 00:00:52,390 --> 00:00:56,700 You can find the video attached in the lecture resources. 11 00:00:56,700 --> 00:01:04,110 So the input video, which is smile that's aimed before and this is the output VIDEO So I'm going to 12 00:01:04,290 --> 00:01:10,050 leave the output video and recreate it using open CV. 13 00:01:10,680 --> 00:01:18,060 You also have this face is the example, which is the Cascade Fire, which contains the blueprints. 14 00:01:18,060 --> 00:01:22,350 Oh, our face looks like in example code. 15 00:01:24,660 --> 00:01:31,470 So the program needs to have that file in order to recognize faces in a video. 16 00:01:31,980 --> 00:01:37,050 So video is legal to see to the video capture. 17 00:01:37,110 --> 00:01:44,250 What I'm doing is referring to Smile does MP Fool and creates a video object. 18 00:01:45,340 --> 00:01:53,740 In Python and then create two valuable success and frame a sequel to video reads. 19 00:01:54,040 --> 00:02:01,720 So what I'm doing here is I'm reading the first frame of the video just to check if the video is read 20 00:02:01,720 --> 00:02:02,350 correctly. 21 00:02:02,620 --> 00:02:09,580 So success will gets through or falls as a value and frame will give the first frame as a value. 22 00:02:10,360 --> 00:02:14,440 So if the frame is read successfully, success will be equal to true. 23 00:02:14,440 --> 00:02:16,690 Otherwise, it will be equal to falls. 24 00:02:17,170 --> 00:02:27,310 Then we have faced Cascade, a variable which points to servitude that cascade classifier, and that's 25 00:02:27,310 --> 00:02:32,020 points to faces that example the path of the XML file. 26 00:02:33,190 --> 00:02:38,680 So we have the video, we have the first frame, we have the face cascades object and then we create 27 00:02:38,680 --> 00:02:39,430 an output. 28 00:02:41,810 --> 00:02:43,340 Which is empty for now. 29 00:02:43,760 --> 00:02:46,010 So that's a sequel to video writer. 30 00:02:49,310 --> 00:03:02,420 Let's name this file outputs smile, MP, fool or blurred smile if you preferred or blur face. 31 00:03:02,660 --> 00:03:14,000 So that is the ultimate video now this requires also another argument here that is CV to dot video right 32 00:03:14,000 --> 00:03:23,960 to underscore for CC, which is a class by itself. 33 00:03:24,410 --> 00:03:31,340 It expects now this argument, which is the context of the video you want to create. 34 00:03:32,030 --> 00:03:34,840 The effects of this is what works. 35 00:03:34,850 --> 00:03:38,180 I've tried this and three frames. 36 00:03:38,900 --> 00:03:40,010 It's another argument. 37 00:03:41,310 --> 00:03:48,060 So we've got the path to the ultimate file, the context of the video as a second argument. 38 00:03:48,810 --> 00:03:51,810 The frames per second of the article, the video. 39 00:03:52,050 --> 00:03:59,670 And lastly, the dimensions of the output feed you know of this should be in the form of a tipple, 40 00:03:59,670 --> 00:04:03,510 which could be something like a three hundreds for hundreds of bullets. 41 00:04:04,290 --> 00:04:08,160 You can just declare these numbers like that. 42 00:04:08,520 --> 00:04:18,660 You want to have the same width and height as the original video so that you don't end up with a distorted 43 00:04:18,660 --> 00:04:19,050 video. 44 00:04:19,920 --> 00:04:25,260 Therefore, we need to get the dimensions of the original video. 45 00:04:25,950 --> 00:04:32,400 So what I'm going to do is after we create the video object, after we get the first frame, we get 46 00:04:32,400 --> 00:04:34,500 the dimensions of that first frame. 47 00:04:35,220 --> 00:04:43,410 It's a single to frame dot shape and we get the first item of that table. 48 00:04:43,410 --> 00:04:51,120 So that's creates a table with the height and the width, and the width is equal to frame its shape. 49 00:04:53,840 --> 00:04:54,440 One. 50 00:04:56,160 --> 00:05:01,560 So then we can specify here the width which comes first and the height. 51 00:05:02,430 --> 00:05:05,760 So those are the dimensions of the ultimate video. 52 00:05:07,110 --> 00:05:09,840 Then we were ready to build the wire loop. 53 00:05:10,240 --> 00:05:14,070 Well, success, for example, which is equal to true. 54 00:05:14,100 --> 00:05:26,160 So if success is true, then we build the while loop and ride frames in this output video object. 55 00:05:26,580 --> 00:05:32,610 So while success, we did take the faces from using face cascades. 56 00:05:33,780 --> 00:05:43,770 So that object, which has a detect multi scale glass and this gets a frame as inputs. 57 00:05:44,340 --> 00:05:46,200 So we were working on the first frame. 58 00:05:46,200 --> 00:05:50,010 For now, we detect faces in the first frame. 59 00:05:50,260 --> 00:05:52,170 And these are the coefficients. 60 00:05:52,280 --> 00:05:58,350 And so I've covered those one point one and four is a good set of numbers. 61 00:05:59,130 --> 00:06:05,970 So if you change these, you can make the algorithm more sensitive or less sensitive. 62 00:06:06,390 --> 00:06:13,230 So if you make it over sensitive, it means you will detect false positives. 63 00:06:13,380 --> 00:06:17,340 So false faces objects which which are not faces, but. 64 00:06:18,520 --> 00:06:27,070 Are classified as faces from this algorithm, so that will give us a list of faces because we might 65 00:06:27,070 --> 00:06:29,650 have more than one faces in the frame. 66 00:06:30,310 --> 00:06:35,260 But then we can iterate over the faces for. 67 00:06:36,600 --> 00:06:37,680 X or Y? 68 00:06:37,710 --> 00:06:41,790 So each face has an X and Y and the width and height. 69 00:06:42,870 --> 00:06:46,680 So these are the dimensions of the rectangle around the detected phase. 70 00:06:47,520 --> 00:06:50,010 So we see for that to pull interfaces. 71 00:06:51,540 --> 00:06:54,870 And then we modify the current frame. 72 00:06:55,570 --> 00:07:04,680 So the current frame, but we only modify a particular slice of the frame that is a slice, the portion 73 00:07:04,680 --> 00:07:06,000 where the face is. 74 00:07:06,210 --> 00:07:12,200 So with this coordinate, so y so from y until y plus h. 75 00:07:12,930 --> 00:07:20,520 So that means where the rectangle of the face towards and until somewhere in here. 76 00:07:21,390 --> 00:07:24,810 So we get that portion and then we get the other portion as well. 77 00:07:25,200 --> 00:07:31,080 So we go again from X to X plus width. 78 00:07:31,380 --> 00:07:33,240 So that will give us basically that. 79 00:07:33,240 --> 00:07:38,130 And that's so it will give us the entire rectangle where the face is. 80 00:07:38,580 --> 00:07:42,540 So the entire frame is something like this, right? 81 00:07:43,170 --> 00:07:45,810 And the phase has these coordinates. 82 00:07:46,170 --> 00:07:59,070 So now what we do to this portion of the frame here is we apply a CV to the blur methods and this goes 83 00:07:59,070 --> 00:07:59,580 like this. 84 00:07:59,700 --> 00:08:01,620 So frame again, y. 85 00:08:02,730 --> 00:08:04,170 Y plus h. 86 00:08:05,410 --> 00:08:09,520 And ex until X plus width. 87 00:08:10,390 --> 00:08:16,120 And then we also provides the intensity of the blurriness. 88 00:08:16,570 --> 00:08:18,520 So 50 50 is a good number. 89 00:08:18,670 --> 00:08:25,090 You can go less than that if you want the face to be more visible and you can increase these if you 90 00:08:25,090 --> 00:08:27,850 want the blurring effect to be more intense. 91 00:08:28,600 --> 00:08:29,320 And that's it. 92 00:08:29,800 --> 00:08:34,840 So now we have modified that portion of the frame by blurting it out and. 93 00:08:36,520 --> 00:08:46,540 Then we exit the loop and we say, Oh, foods dot right frame, so we are right, that's current frame 94 00:08:46,810 --> 00:08:49,780 that's modified frame into the outputs. 95 00:08:49,780 --> 00:08:56,680 VIDEO Then what we do is so far we worked on the very first frame of the video, then we want to give 96 00:08:56,690 --> 00:08:57,880 the other frames. 97 00:08:58,750 --> 00:09:05,440 So we say again, success that and frame is sequel to VIDEO That's right. 98 00:09:06,190 --> 00:09:13,290 So that we'll read the next frame of the video, and that's it. 99 00:09:14,090 --> 00:09:22,780 So the while loop will be repeating itself and we exit the loop and we say outputs that release. 100 00:09:24,400 --> 00:09:31,030 To release the video and make it able for writing it on disc. 101 00:09:31,480 --> 00:09:33,400 So I'm going to execute this now. 102 00:09:42,500 --> 00:09:48,950 Now, if you get this error, there is some mismatch between the formats, the video formats. 103 00:09:49,280 --> 00:09:57,290 That means you want to change the outputs from MP for to delete a VR and run again. 104 00:09:58,220 --> 00:10:06,710 You can see that blur smile the same before it's created, but the script has not ended execution yet. 105 00:10:06,980 --> 00:10:10,460 So we want to wait until this file is written. 106 00:10:11,030 --> 00:10:17,300 And of course, for the next time, you can improve the experience, the execution, the program execution 107 00:10:17,300 --> 00:10:26,780 experience by writing here a variable count is equal to zero and then you do see counts as increased 108 00:10:26,780 --> 00:10:32,570 by one and then you print the cones so that you can see some outputs here. 109 00:10:32,750 --> 00:10:37,940 So you see the while loop being executed, so you see these numbers. 110 00:10:39,840 --> 00:10:45,180 Actually, I can stop this and Ronnie's again so that we can see some outputs in here. 111 00:10:45,390 --> 00:10:50,130 So now we see the cones, which basically represent each frame. 112 00:10:52,400 --> 00:10:54,930 So every frame now is being processed. 113 00:10:54,950 --> 00:10:57,950 We are in the eighth frame, nine, 10 and so on. 114 00:11:00,480 --> 00:11:06,750 There will be 90 frames in total because the video is three seconds long. 115 00:11:08,280 --> 00:11:10,470 So there are 30 frames per second. 116 00:11:10,950 --> 00:11:13,050 Three times three, it's 90. 117 00:11:16,780 --> 00:11:24,670 Yes, of the execution has finished, I can download the Lord's mine, that's a AIVI file. 118 00:11:27,010 --> 00:11:27,820 And here it is. 119 00:11:28,030 --> 00:11:36,760 So you can only open this, at least in my case, I can only open it with VLCC program VRC Media Player. 120 00:11:37,450 --> 00:11:39,970 So the other ones might not work. 121 00:11:42,570 --> 00:11:45,570 And yeah, that's the outputs, and that's the final codes. 122 00:11:46,830 --> 00:11:49,450 Thanks for following this, and I'll talk to you the next videos. 123 00:11:49,450 --> 00:11:49,770 So you.