├── util
    ├── __init__.py
    ├── util.pyc
    ├── __init__.pyc
    ├── __pycache__
    │   ├── util.cpython-34.pyc
    │   └── __init__.cpython-34.pyc
    └── util.py
├── screenshots
    ├── freq.png
    ├── trace.png
    └── filter_time_domain.png
├── README.md
└── acc.py


/util/__init__.py:
--------------------------------------------------------------------------------
1 | from .util import*
2 | 


--------------------------------------------------------------------------------
/util/util.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/util/util.pyc


--------------------------------------------------------------------------------
/util/__init__.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/util/__init__.pyc


--------------------------------------------------------------------------------
/screenshots/freq.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/screenshots/freq.png


--------------------------------------------------------------------------------
/screenshots/trace.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/screenshots/trace.png


--------------------------------------------------------------------------------
/screenshots/filter_time_domain.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/screenshots/filter_time_domain.png


--------------------------------------------------------------------------------
/util/__pycache__/util.cpython-34.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/util/__pycache__/util.cpython-34.pyc


--------------------------------------------------------------------------------
/util/__pycache__/__init__.cpython-34.pyc:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/KChen89/Accelerometer-Filtering/HEAD/util/__pycache__/__init__.cpython-34.pyc


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # Accelerometer data filtering
 2 | Filter 3D accelerometer data [1] with median and low pass filter. Median filter generally remove big spikes. But, high frequency noise exists at very low amplitude. Low pass filter only preserves low frequency which creates undesirable distortions. Combing median and low pass filter generally recovers a better trace (since there is no ground truth signal to verify which one is better, just follow orignal data description [1] and verify with eyes).</br>
 3 | ![time domain](screenshots/filter_time_domain.png)
 4 | Time domain comparsion. 
 5 | ![3d trace](screenshots/trace.png)
 6 | Accelerometer data 3D trace. 
 7 | ### Dependencies
 8 | - Numpy
 9 | - Scipy
10 | - Matplotlib
11 | 
12 | :question: However, according the original data source description [1], the above 3D trace is suppose to be the displacement which is the integration of integration of accelerometer. When I integrated the data twice, it does not show a meaningful result. :confused:  
13 | 
14 | #### More
15 | - [x] filter data to create a clear trace
16 | - [ ] figure out the confusion.
17 | 
18 | ### Reference
19 | [1] https://www.shimmersensing.com/support/sample-data/
20 | 


--------------------------------------------------------------------------------
/acc.py:
--------------------------------------------------------------------------------
 1 | '''
 2 | 3D accelerometer filtering with median and low pass filter
 3 | @author: Kemeng Chen
 4 | '''
 5 | import os
 6 | import sys
 7 | import numpy as np
 8 | import matplotlib.pyplot as plt
 9 | from util import* 
10 | 
11 | def test_data(file_name):
12 | 	cur_dir=os.getcwd()
13 | 	fs=512
14 | 	cutoff=10
15 | 	file_path=os.path.join(cur_dir, 'data', file_name)
16 | 	data=read_data(file_path, [1,2,3])
17 | 	plot_lines(data, fs, 'Raw data')
18 | 	fft_plot(data, fs, 'Raw data')
19 | 	median_data=median_filter(data, 155)
20 | 	lpf_data=freq_filter(data, 155, cutoff/fs)
21 | 	comb_data=freq_filter(median_data, 155, cutoff/fs)
22 | 	plot_lines(median_data, fs, 'median filter')
23 | 	plot_lines(lpf_data, fs, 'low pass filter')
24 | 	plot_lines(comb_data, fs, 'median+low pass filter')
25 | 	fft_plot(lpf_data, fs, 'low pass filter')
26 | 	fft_plot(median_data, fs, 'median filter')
27 | 	fft_plot(comb_data, fs, 'median+low pass filter')
28 | 	plot3D(data, 'raw data')
29 | 	plot3D(median_data, 'median filter')
30 | 	plot3D(lpf_data, 'low pass filter')
31 | 	plot3D(comb_data, 'median+low pass filter')
32 | 	plt.show()
33 | 
34 | if __name__ == '__main__':
35 | 	if len(sys.argv)<2:
36 | 		raise ValueError('No file name specified')
37 | 	test_data(sys.argv[1])


--------------------------------------------------------------------------------
/util/util.py:
--------------------------------------------------------------------------------
  1 | '''
  2 | utility function to read data, integration, and plot
  3 | @author: Kemeng Chen
  4 | '''
  5 | 
  6 | import os
  7 | import sys
  8 | import numpy as np 
  9 | import math
 10 | import matplotlib.pyplot as plt
 11 | from scipy.fftpack import fft
 12 | from scipy import signal
 13 | from mpl_toolkits.mplot3d import Axes3D
 14 | 
 15 | def read_data(file_path, columns):
 16 | 	'''
 17 | 	read files according to file_path and columns
 18 | 	args:
 19 | 		file_path, columns (list)
 20 | 	return:
 21 | 		data (numpy array)
 22 | 	'''
 23 | 	if not os.path.isfile(file_path):
 24 | 		raise AssertionError(file_path, 'not found')
 25 | 	mode='r'
 26 | 	with open (file_path, mode) as f:
 27 | 		lines=f.readlines()
 28 | 		num_rows=len(lines)
 29 | 		print(str(num_rows), ' rows')
 30 | 		num_cols=len(columns)
 31 | 		data=np.zeros([num_rows, num_cols])
 32 | 		for indice, line in enumerate(lines[3:]):
 33 | 			row=line.rstrip().split('\t')
 34 | 			# print(row)
 35 | 			for ii, i in enumerate(columns):
 36 | 				data[indice,ii]=row[i]
 37 | 		# data[:,2]-=10
 38 | 		# for i in range(3):
 39 | 		# 	data[:,i]=calibration(data[:,i])
 40 | 	f.close()
 41 | 	return data
 42 | 
 43 | def median_filter(data, f_size):
 44 | 	lgth, num_signal=data.shape
 45 | 	f_data=np.zeros([lgth, num_signal])
 46 | 	for i in range(num_signal):
 47 | 		f_data[:,i]=signal.medfilt(data[:,i], f_size)
 48 | 	return f_data
 49 | 
 50 | def freq_filter(data, f_size, cutoff):
 51 | 	lgth, num_signal=data.shape
 52 | 	f_data=np.zeros([lgth, num_signal])
 53 | 	lpf=signal.firwin(f_size, cutoff, window='hamming')
 54 | 	for i in range(num_signal):
 55 | 		f_data[:,i]=signal.convolve(data[:,i], lpf, mode='same')
 56 | 	return f_data
 57 | 
 58 | def fft_plot(data, fs, title):
 59 | 	lgth, num_signal=data.shape
 60 | 	fqy=np.zeros([lgth,num_signal])
 61 | 	fqy[:,0]=np.abs(fft(data[:,0]))
 62 | 	fqy[:,1]=np.abs(fft(data[:,1]))
 63 | 	fqy[:,2]=np.abs(fft(data[:,2]))
 64 | 	index=np.arange(int(lgth/2))/(int(lgth/2)/(fs/2))
 65 | 	fig, ax=plt.subplots()
 66 | 	labels=['x','y','z']
 67 | 	color_map=['r', 'g', 'b']
 68 | 	for i in range(3):
 69 | 		ax.plot(index, fqy[0:int(lgth/2),i], color_map[i], label=labels[i])
 70 | 	ax.set_xlim([0, fs/2])
 71 | 	ax.set_xlabel('Hz')
 72 | 	ax.set_title('Frequency spectrum: '+title) 
 73 | 	ax.legend()
 74 | 
 75 | def plot_lines(data, fs, title):
 76 | 	num_rows, num_cols=data.shape
 77 | 	if num_cols!=3:
 78 | 		raise ValueError('Not 3D data')
 79 | 	fig, ax=plt.subplots()
 80 | 	labels=['x','y','z']
 81 | 	color_map=['r', 'g', 'b']
 82 | 	index=np.arange(num_rows)/fs
 83 | 	for i in range(num_cols):
 84 | 		ax.plot(index, data[:,i], color_map[i], label=labels[i])
 85 | 	ax.set_xlim([0,num_rows/fs])
 86 | 	ax.set_xlabel('Time [sec]')
 87 | 	ax.set_title('Time domain: '+title)
 88 | 	ax.legend()
 89 | 
 90 | def acc_integration(data):
 91 | 	num_rows, num_cols=data.shape
 92 | 	int_data=np.zeros(data.shape)
 93 | 	for i in range(num_cols):
 94 | 		int_data[:,i]=TZ_integration(data[:,i])
 95 | 	return int_data
 96 | 
 97 | def plot3D(data, title):
 98 | 	fig=plt.figure()
 99 | 	ax=fig.add_subplot(111, projection='3d')
100 | 	ax.plot(xs=data[:,0], ys=data[:,1], zs=data[:,2], zdir='z')
101 | 	ax.set_title(title)
102 | 
103 | def calibration(signal):
104 | 	inc_eng=np.sum(np.clip(signal, a_min=0, a_max=None))
105 | 	der_eng=-1*np.sum(np.clip(signal, a_max=0, a_min=None))
106 | 	if inc_eng==0 or der_eng==0:
107 | 		raise ValueError('Calibration rule does NOT hold')
108 | 
109 | 	if inc_eng>der_eng:
110 | 		beta_p=1
111 | 		beta_n=inc_eng/der_eng
112 | 		c_signal=(beta_n-1)*np.clip(signal, a_max=0, a_min=None)+signal
113 | 	elif der_eng>inc_eng:
114 | 		beta_n=1
115 | 		beta_p=der_eng/inc_eng
116 | 		c_signal=(beta_p-1)*np.clip(signal, a_min=0, a_max=None)+signal
117 | 	else:
118 | 		c_signal=signal
119 | 	return c_signal
120 | 
121 | def TZ_integration(in_signal):
122 | 	lgth=in_signal.shape
123 | 	integral=np.zeros(lgth)
124 | 	c=0
125 | 	for indice, s in enumerate(in_signal):
126 | 		if indice==0:
127 | 			integral[indice]=c+in_signal[indice]
128 | 		else:
129 | 			integral[indice]=integral[indice-1]+in_signal[indice]
130 | 	return integral
131 | 
132 | 


--------------------------------------------------------------------------------