TA的每日心情 | 开心
2024-10-14 08:33 |
|---|
签到天数: 2451 天 连续签到: 1 天 [LV.Master]伴坛终老
司徒
- 积分
- 33848
 
|
CurieNeurons，即Curie神经元。Neurons能通过示例学习，不用编程，可以通过数据离线学习。
general vision提供了两个版本可供下载，免费版和收费19美刀的专业版。
下载地址:http://www.general-vision.com/software/curieneurons/
为了尝鲜我们先体验一下免费版吧，
相关的文档：
详细的信息可以看文档里的介绍，这里我们就直接切入正题了。
需要的材料很简单，Curie Nano板和一块AD Keyborad板。
这个AD Keyborad DFROBOT商城有卖，不过太贵了买不起，好在提供原理图，我手工做了一个。
5个按键只占用一个模拟IO口，很巧妙的设计。
这里VCC接3.3V，IO口接A5。
示例代码：- /*
- Name: CurieNeurons_andIMU2.0.ino
- Created: 7/21/2016 11:08:34 AM
- Author: HansYang
- */
- #include <CurieIMU.h>
- #include <BMI160.h>
- #include <CurieNeurons.h>
- #include <stdlib.h>
- #define DEBUG1
- #define ACC_FEATURE_ONLY
- #define MANUAL_SCALING
- #ifdef ACC_FEATURE_ONLY
- #define SAMPLENBR 3
- #else
- #define SAMPLENBR 6
- #endif // ACC_FEATURE_ONLY
- #define MAX_SAMPLING_TIME 3000//ms
- #define SAMPLING_INTERVAL 50//ms
- #define MAX_VECTOR_LEN 60//MAX_SAMPLING_TIME/SAMPLING_INTERVAL
- #define DOWNSAMPLER_SEGS 10
- //ADkeyboard
- int adc_key_val[5] = { 850,900, 940, 960, 980 };
- #define NUM_KEYS 5
- int adc_key_in;
- int key = -1;
- //CurieNeurons
- CurieNeurons hNN;
- int cat, prevcat;
- int dist, nid;
- //IMU data
- int16_t ax, ay, az;
- int16_t gx, gy, gz;
- int minax, minay, minaz, maxax, maxay, maxaz;
- int mingx, mingy, mingz, maxgx, maxgy, maxgz;
- int Sax, Say, Saz, Sgx, Sgy, Sgz;
- //Others
- uint64_t timer;
- short vector_len;
- byte vector[MAX_VECTOR_LEN*SAMPLENBR];
- byte pattern[SAMPLENBR * 17];
- byte learnflag;
- void feature_extraction(byte* src_vector, int src_len, byte *dst_vector)
- {
- int seg_len = src_len / (DOWNSAMPLER_SEGS * SAMPLENBR);
- double res[17 * SAMPLENBR];
- for (int icol = 0; icol < SAMPLENBR; icol++)
- {
- int sum = 0;
- for (int i = 0; i < src_len; i++)
- {
- sum += src_vector[i*SAMPLENBR + icol];
- }
- //mean
- int aver = sum / src_len;
- int e = 0;
- int mad = 0;
- //int rms = 0;
- //byte max = src_vector[icol];
- //byte min = src_vector[icol];
- int ratiocount[5] = { 0 };
- for (int i = 0; i < src_len; i++)
- {
- e += (src_vector[i*SAMPLENBR + icol] - aver)*(src_vector[i*SAMPLENBR + icol] - aver);
- mad += abs(src_vector[i*SAMPLENBR + icol] - aver);
- //rms += src_vector[i*SAMPLENBR + icol] * src_vector[i*SAMPLENBR + icol];
- //if (src_vector[i*SAMPLENBR + icol] > max)
- // max = src_vector[i*SAMPLENBR + icol];
- //if (src_vector[i*SAMPLENBR + icol] < min)
- // min = src_vector[i*SAMPLENBR + icol];
- if (src_vector[i*SAMPLENBR + icol] <= 200)
- {
- if (src_vector[i*SAMPLENBR + icol] <= 150)
- {
- if (src_vector[i*SAMPLENBR + icol] <= 100)
- {
- if (src_vector[i*SAMPLENBR + icol] <= 50)
- {
- ratiocount[4]++;
- }
- else
- ratiocount[3]++;
- }
- else
- ratiocount[2]++;
- }
- else
- ratiocount[1]++;
- }
- else
- ratiocount[0]++;
- }
- //standard deviation
- res[icol] = sqrt(e / (src_len - 1)) * 2;
- //MAD
- res[icol + 1 * SAMPLENBR] = mad / src_len * 2;
- //RMS
- //res[icol + 3 * SAMPLENBR] = sqrt(rms / src_len);
- for (int i = 0; i < 5; i++)
- {
- res[icol * 5 + 6 + i] = 0;//ratiocount[i];// * 255 / src_len;
- }
- for (int iseg = 0; iseg < DOWNSAMPLER_SEGS; iseg++)
- {
- int sum = 0;
- for (int ismp = 0; ismp < seg_len; ismp++)
- {
- int offset = icol + (iseg * seg_len + ismp)*SAMPLENBR;
- sum += (int)src_vector[offset];
- src_vector[offset] = 0;
- }
- res[20 + icol * DOWNSAMPLER_SEGS + iseg] = sum / seg_len;
- }
- }
- for (int i = 0; i < SAMPLENBR * 17; i++)
- {
- pattern[i] = (int)(res[i] + 0.5) * 0x00FF;
- }
- #ifdef DEBUG1
- Serial.print("\n");
- for (int i = 0; i < SAMPLENBR*17; i++)
- {
- Serial.print(pattern[i]);
- Serial.print("\t");
- }
- //Serial.print("\n");
- #endif // DEBUG1
- }
- int get_key(unsigned int input) //1 2 3 4 5
- {
- int k;
- for (k = 0; k < NUM_KEYS; k++)
- {
- if (input < adc_key_val[k])
- {
- return k + 1;
- }
- }
- if (k >= NUM_KEYS)
- k = -1; // No valid key pressed
- return k + 1;
- }
- void read_data_once()
- {
- CurieIMU.getMotion6(&ax, &ay, &az, &gx, &gy, &gz);
- /*int R = sqrt(ax*ax + ay*ay + az*az);
- double tX = acos(ax / R);
- double tY = acos(ay / R);
- double tZ = acos(az / R);
- ax = ax - 8192 * cos(tX);
- ay = ay - 8192 * cos(tY);
- az = ax - 8192 * cos(tZ);*/
-
- vector[vector_len*SAMPLENBR + 0] = (byte)(((ax - minax) * 255 / Sax) & 0x00FF);
- vector[vector_len*SAMPLENBR + 1] = (byte)(((ay - minay) * 255 / Say) & 0x00FF);
- vector[vector_len*SAMPLENBR + 2] = (byte)(((az - minaz) * 255 / Saz) & 0x00FF);
- #ifndef ACC_FEATURE_ONLY
- vector[vector_len*SAMPLENBR + 3] = (byte)(((gx - mingx) * 255 / Sgx) & 0x00FF);
- vector[vector_len*SAMPLENBR + 4] = (byte)(((gy - mingy) * 255 / Sgy) & 0x00FF);
- vector[vector_len*SAMPLENBR + 5] = (byte)(((gz - mingz) * 255 / Sgz) & 0x00FF);
- #endif // !ACC_FEATURE_ONLY
- /*vector[vector_len*SAMPLENBR + 0] = ax;
- vector[vector_len*SAMPLENBR + 1] = ay;
- vector[vector_len*SAMPLENBR + 2] = az;
- #ifndef ACC_FEATURE_ONLY
- vector[vector_len*SAMPLENBR + 3] = gx;
- vector[vector_len*SAMPLENBR + 4] = gy;
- vector[vector_len*SAMPLENBR + 5] = gz;
- #endif // !ACC_FEATURE_ONLY */
- if (vector_len < (MAX_VECTOR_LEN - 2)*SAMPLENBR ) {
- vector_len++;
- Serial.print("*");
- }
- else
- {
- Serial.print("O");
- }
- }
- // the setup function runs once when you press reset or power the board
- void setup() {
- Serial.begin(9600);
- while (!Serial);
- Serial.print("CurieNeurons initializing...");
- hNN.begin();
- hNN.forget();
- Serial.println("Done");
- Serial.print("CurieIMU initializing...");
- CurieIMU.begin();
- Serial.println("Done");
- Serial.println("Initializing IMU device...");
- CurieIMU.begin();
- // use the code below to calibrate accel/gyro offset values
- Serial.println("Internal sensor offsets BEFORE calibration...");
- Serial.print(CurieIMU.getXAccelOffset());
- Serial.print("\t"); // -76
- Serial.print(CurieIMU.getYAccelOffset());
- Serial.print("\t"); // -235
- Serial.print(CurieIMU.getZAccelOffset());
- Serial.print("\t"); // 168
- Serial.print(CurieIMU.getXGyroOffset());
- Serial.print("\t"); // 0
- Serial.print(CurieIMU.getYGyroOffset());
- Serial.print("\t"); // 0
- Serial.println(CurieIMU.getZGyroOffset());
- Serial.println("About to calibrate. Make sure your board is stable and upright");
- delay(2000);
- // The board must be resting in a horizontal position for
- // the following calibration procedure to work correctly!
- Serial.print("Starting Gyroscope calibration...");
- CurieIMU.autoCalibrateGyroOffset();
- Serial.println(" Done");
- Serial.print("Starting Acceleration calibration...");
- CurieIMU.autoCalibrateXAccelOffset(0);
- CurieIMU.autoCalibrateYAccelOffset(0);
- CurieIMU.autoCalibrateZAccelOffset(1);
- Serial.println(" Done");
- Serial.println("Internal sensor offsets AFTER calibration...");
- Serial.print(CurieIMU.getXAccelOffset());
- Serial.print("\t"); // -76
- Serial.print(CurieIMU.getYAccelOffset());
- Serial.print("\t"); // -2359
- Serial.print(CurieIMU.getZAccelOffset());
- Serial.print("\t"); // 1688
- Serial.print(CurieIMU.getXGyroOffset());
- Serial.print("\t"); // 0
- Serial.print(CurieIMU.getYGyroOffset());
- Serial.print("\t"); // 0
- Serial.println(CurieIMU.getZGyroOffset());
- Serial.println("Enabling Gyroscope/Acceleration offset compensation");
- CurieIMU.setGyroOffsetEnabled(true);
- CurieIMU.setAccelOffsetEnabled(true);
- CurieIMU.setAccelerometerRange(4);
- //hNN.GCR(128); //LSUP distance
- //hNN.MAXIF(600);
- #ifdef MANUAL_SCALING
- minax = minay = minaz = maxax = maxay = maxaz = -32768;
- mingx = mingy = mingz = maxgx = maxgy = maxgz = 32767;
- Sax = Say = Saz = Sgx = Sgy = Sgz = 65535;
- #else
- Serial.println("\nAssessing the amplitude of motions...Press any key to continue");
- adc_key_in = analogRead(5);
- while (adc_key_in > 980)
- {
- adc_key_in = analogRead(5);
- CurieIMU.getMotion6(&ax, &ay, &az, &gx, &gy, &gz);
- if (ax>maxax) maxax = ax; if (ax<minax) minax = ax;
- if (ay>maxay) maxay = ay; if (ay<minay) minay = ay;
- if (az>maxaz) maxaz = az; if (az<minaz) minaz = az;
- if (gx>maxgx) maxgx = gx; if (gx<mingx) mingx = gx;
- if (gy>maxgy) maxgy = gy; if (gy<mingy) mingy = gy;
- if (gz>maxgz) maxgz = gz; if (gz<mingz) mingz = gz;
- }
- Sax = maxax - minax;
- Say = maxay - minay;
- Saz = maxaz - minaz;
- Sgx = maxgx - mingx;
- Sgy = maxgy - mingy;
- Sgz = maxgz - mingz;
- Serial.print("\nSa :\t"); Serial.print(Sax); Serial.print(","); Serial.print(Say); Serial.print(","); Serial.print(Saz);
- Serial.print("\nSg :\t"); Serial.print(Sgx); Serial.print(","); Serial.print(Sgy); Serial.print(","); Serial.println(Sgz);
- #endif
- timer = 0;
- prevcat = 0x7FFF;
- learnflag = 0;
- Serial.print("The program will begin in 2 seconds...\n");
- delay(2000);
- Serial.println("Start.");
- }
- // the loop function runs over and over again until power down or reset
- void loop() {
- timer = millis();
- adc_key_in = analogRead(5);
- key = get_key(adc_key_in);
- if (key!=0)
- {
- learnflag = key;
- read_data_once();
- }
- else if (learnflag != 0)
- {
- read_data_once();
- feature_extraction(vector, vector_len, pattern);
- //downsampler_means(vector, vector_len, pattern, SAMPLENBR*DOWNSAMPLER_SEGS);
- #ifdef DEBUG
- Serial.print("\n");
- for (int i = 0; i < SAMPLENBR*DOWNSAMPLER_SEGS; i++)
- {
- Serial.print(pattern[i]);
- Serial.print("\t");
- }
- Serial.print("\n");
- #endif //DEBUG
- if (learnflag != 5)
- {
- hNN.learn(pattern, SAMPLENBR * 17, learnflag);
- Serial.print("\nLearned motion #");
- Serial.print(learnflag);
- Serial.println("once, please repeat.");
- }
- else
- {
- int res = hNN.classify(pattern, SAMPLENBR * 17, &dist, &cat, &nid);
- cat &= 0x00FF;
- if (res != 0)
- {
- Serial.print("\nMotion Detected. #"); Serial.println(cat);
- }
- else
- {
- Serial.println("\nMotion Unknow.");
- }
- }
- learnflag = 0;
- vector_len = 0;
- }
- while ((millis() - timer) < SAMPLING_INTERVAL);
- }
动作学习：
按住黄色、绿色、蓝色或者红色按键做动作，动作完成松手。黄色对应动作1，绿色对应动作2，蓝色对应动作3，红色对应动作4。
每种动作可以多做几次，这样识别的准确率会高一些。
下图是我按住黄色按键1和绿色按键2完成的动作识别。
识别动作:
按住中间的白色按键5，做出刚才的动作，程序会识别属于刚才的哪个动作。
如下图所示，如果做出的动作不是刚才的两种之一会显示未知。
动作的采样时间可以在以下的语句修改，但是时间越来复杂度也就越高，识别的准确率也就越低。
程序说明:
在这个过程中，主要是采用按键按下作为动作的开始，松开表示动作的结束，再将采样的动作进行数据分析，再由CurieNeurons神经网络进行学习。识别动作的方式同理。
|
|
DFROBOT
6972 人阅读
|
6 人回复
|
2017-05-16
/3 
