1 00:00:00,990 --> 00:00:09,150 Lesson eight project for people counting with ultrasonic sensors, specifically theory and data collection. 2 00:00:10,110 --> 00:00:16,350 In this article, we will create a people accounting system by using Real Terminal, an ordinary ultrasonic 3 00:00:16,350 --> 00:00:18,780 ranger and special deep learning saws. 4 00:00:18,810 --> 00:00:26,940 To top it off and actually make it to work, we will also utilize Microsoft Azure IoT Central Service 5 00:00:26,940 --> 00:00:31,500 to store the real occupancy data, the cloud and visualize it on ABC. 6 00:00:31,860 --> 00:00:33,480 Now that sounds interesting, right? 7 00:00:34,350 --> 00:00:40,020 The first thing we need to do is learn about the data we can get from the ultrasonic sensor and how 8 00:00:40,020 --> 00:00:42,420 we can use it to figure out the direction of things. 9 00:00:43,650 --> 00:00:48,810 This grow ultrasonic ranger is a module that doesn't need to be touched to measure distance. 10 00:00:49,560 --> 00:00:51,420 It works at 40 hertz. 11 00:00:52,170 --> 00:00:58,140 We then send a pulse signal through the signal pin that last more than 10 us through signal pin. 12 00:00:59,100 --> 00:01:05,640 Then the global supersonic ranger will send out eight cycles of 400 cycle level and listen for the Echo. 13 00:01:06,390 --> 00:01:10,200 The pulse width of the Echo signal changes as you get closer. 14 00:01:10,620 --> 00:01:14,310 If you want to know how to do this, here is what you need. 15 00:01:14,610 --> 00:01:22,980 Reflection time multiplied by sound speed The sound speed is equivalent to 340 meters per second squared. 16 00:01:24,240 --> 00:01:29,830 Now it is important to take note that do not hot block growth ultrasonic ranger. 17 00:01:29,850 --> 00:01:32,360 Otherwise, it will damage the sensor. 18 00:01:32,730 --> 00:01:37,020 The measured area must be no less than 0.5 square meters and smooth. 19 00:01:37,930 --> 00:01:44,310 Now, after installing, girls will just like Ranger Library for the Arduino HD and connecting ultrasonic 20 00:01:44,310 --> 00:01:47,640 ranger to wheel Terminal D one or two. 21 00:01:48,600 --> 00:01:54,150 We can upload this simple script with a terminal connected to growth will just like Ranger and then 22 00:01:54,150 --> 00:01:56,100 walk in and walk out of the room. 23 00:01:58,930 --> 00:02:06,160 Then after that, we can immediately see that for walking it, we can get relatively high values, which 24 00:02:06,160 --> 00:02:13,060 are corresponding to the distance from the object first, which then decrease, and for walking out, 25 00:02:13,060 --> 00:02:15,070 we get completely the opposite of it. 26 00:02:16,060 --> 00:02:21,730 It's possible that we could write an algorithm by hand that could help us figure out the right way to 27 00:02:21,730 --> 00:02:22,030 go. 28 00:02:22,570 --> 00:02:26,110 As it turns out, real life is it as simple as it looks? 29 00:02:26,770 --> 00:02:30,640 We have people who walk quickly and people who walk slowly. 30 00:02:31,150 --> 00:02:34,660 We have dinner people and people who aren't so thin. 31 00:02:35,230 --> 00:02:42,610 So our handwritten algorithm has to take all of this into an account, which will make it very complicated 32 00:02:43,090 --> 00:02:45,220 and hard to figure out at the same time. 33 00:02:45,910 --> 00:02:51,580 We have a project that needs to be done, but inferencing, not processing in a lot of noisy data. 34 00:02:51,760 --> 00:02:52,840 Big differences. 35 00:02:53,230 --> 00:03:01,240 Now there is a way to do this, which is called deep learning install ultrasonic reader library, Arduino 36 00:03:01,240 --> 00:03:01,930 IEEE. 37 00:03:02,350 --> 00:03:09,700 Step one is that download the Ultrasonic Ranger Library from GitHub, step to extract the archive and 38 00:03:09,700 --> 00:03:15,460 place it inside the Libraries folder and make sure we are terminal and ultrasonic sensor with screws 39 00:03:15,820 --> 00:03:17,680 to wooden or 3D printed frame. 40 00:03:18,070 --> 00:03:22,960 For example, to put the frame of the wall through metal, velcro strips were used. 41 00:03:23,530 --> 00:03:28,030 Additional options include using phone data, screws or nails. 42 00:03:28,750 --> 00:03:34,420 Now let's start a new project in the edge embossed dashboard and get ready to get the data from it. 43 00:03:35,200 --> 00:03:42,580 We can use the data for our tool from Edge impulsively to get the data, since we don't need very high 44 00:03:42,580 --> 00:03:43,690 sampling frequencies. 45 00:03:44,080 --> 00:03:48,510 We can just use this tool if you have already installed Edge impulsively. 46 00:03:48,910 --> 00:03:51,220 Follow these steps to upload the API. 47 00:03:51,220 --> 00:03:56,080 People counter data collection that inner script to the wheel terminal. 48 00:03:56,740 --> 00:04:00,850 Now this is the same script that was repeated above in this script. 49 00:04:01,180 --> 00:04:08,500 We set all the values above 200 centimeter to negative one, so we can see them actually not for your 50 00:04:08,500 --> 00:04:09,190 application. 51 00:04:09,460 --> 00:04:13,870 You might need to set this value lower or higher, depending on the setup. 52 00:04:14,440 --> 00:04:15,310 Start walking. 53 00:04:16,030 --> 00:04:19,570 You should try walking in and now try walking up. 54 00:04:21,300 --> 00:04:27,480 Now, try walking near the device, not getting closer or further away from you. 55 00:04:27,990 --> 00:04:34,950 For this lesson, we recorded one minute and 30 seconds of data for each class, each time we recorded 56 00:04:34,950 --> 00:04:40,350 five thousand millisecond samples and then cut them down to get one thousand five hundred millisecond 57 00:04:40,350 --> 00:04:40,920 samples. 58 00:04:41,490 --> 00:04:44,850 Make sure you have the samples where you or other people walk fast. 59 00:04:45,120 --> 00:04:47,130 Slow run and so on. 60 00:04:47,560 --> 00:04:49,140 Now try again for walking in. 61 00:04:49,710 --> 00:04:55,770 Then after that, try observing when walking out now for the non category. 62 00:04:56,220 --> 00:05:01,890 Apart from samples that have nobody in front of the device, it is a good idea to include samples that 63 00:05:01,890 --> 00:05:08,790 have a person standing close to the device and walking beside it to avoid any movement being falsely 64 00:05:08,790 --> 00:05:10,830 classified as in or out. 65 00:05:11,580 --> 00:05:17,370 Then lastly, try to change the position of the ultrasonic sensor, installing it lower or higher, 66 00:05:17,940 --> 00:05:20,310 or possibly near another door frame.