ardupilot/Tools/LogAnalyzer/tests/TestOptFlow.py

296 lines
15 KiB
Python

from LogAnalyzer import Test,TestResult
import DataflashLog
from math import sqrt
import numpy as np
import matplotlib.pyplot as plt
class TestFlow(Test):
'''test optical flow sensor scale factor calibration'''
#
# Use the following procedure to log the calibration data. is assumed that the optical flow sensor has been
# correctly aligned, is focussed and the test is performed over a textured surface with adequate lighting.
# Note that the strobing effect from non incandescent artifical lighting can produce poor optical flow measurements.
#
# 1) Set LOG_DISARMED and FLOW_TYPE to 10 and verify that ATT and OF messages are being logged onboard
# 2) Place on level ground, apply power and wait for EKF to complete attitude alignment
# 3) Keeping the copter level, lift it to shoulder height and rock between +-20 and +-30 degrees
# in roll about an axis that passes through the flow sensor lens assembly. The time taken to rotate from
# maximum left roll to maximum right roll should be about 1 second.
# 4) Repeat 3) about the pitch axis
# 5) Holding the copter level, lower it to the ground and remove power
# 6) Transfer the logfile from the sdcard.
# 7) Open a terminal and cd to the ardupilot/Tools/LogAnalyzer directory
# 8) Enter to run the analysis 'python LogAnalyzer.py <log file name including full path>'
# 9) Check the OpticalFlow test status printed to the screen. The analysis plots are saved to
# flow_calibration.pdf and the recommended scale factors to flow_calibration.param
def __init__(self):
Test.__init__(self)
self.name = "OpticalFlow"
def run(self, logdata, verbose):
self.result = TestResult()
self.result.status = TestResult.StatusType.GOOD
def FAIL():
self.result.status = TestResult.StatusType.FAIL
def WARN():
if self.result.status != TestResult.StatusType.FAIL:
self.result.status = TestResult.StatusType.WARN
try:
# tuning parameters used by the algorithm
tilt_threshold = 15 # roll and pitch threshold used to start and stop calibration (deg)
quality_threshold = 124 # minimum flow quality required for data to be used by the curve fit (N/A)
min_rate_threshold = 0.0 # if the gyro rate is less than this, the data will not be used by the curve fit (rad/sec)
max_rate_threshold = 2.0 # if the gyro rate is greter than this, the data will not be used by the curve fit (rad/sec)
param_std_threshold = 5.0 # maximum allowable 1-std uncertainty in scaling parameter (scale factor * 1000)
param_abs_threshold = 200 # max/min allowable scale factor parameter. Values of FLOW_FXSCALER and FLOW_FYSCALER outside the range of +-param_abs_threshold indicate a sensor configuration problem.
min_num_points = 100 # minimum number of points required for a curve fit - this is necessary, but not sufficient condition - the standard deviation estimate of the fit gradient is also important.
# get the existing scale parameters
flow_fxscaler = logdata.parameters["FLOW_FXSCALER"]
flow_fyscaler = logdata.parameters["FLOW_FYSCALER"]
# load required optical flow data
if "OF" in logdata.channels:
flowX = np.zeros(len(logdata.channels["OF"]["flowX"].listData))
for i in range(len(logdata.channels["OF"]["flowX"].listData)):
(line, flowX[i]) = logdata.channels["OF"]["flowX"].listData[i]
bodyX = np.zeros(len(logdata.channels["OF"]["bodyX"].listData))
for i in range(len(logdata.channels["OF"]["bodyX"].listData)):
(line, bodyX[i]) = logdata.channels["OF"]["bodyX"].listData[i]
flowY = np.zeros(len(logdata.channels["OF"]["flowY"].listData))
for i in range(len(logdata.channels["OF"]["flowY"].listData)):
(line, flowY[i]) = logdata.channels["OF"]["flowY"].listData[i]
bodyY = np.zeros(len(logdata.channels["OF"]["bodyY"].listData))
for i in range(len(logdata.channels["OF"]["bodyY"].listData)):
(line, bodyY[i]) = logdata.channels["OF"]["bodyY"].listData[i]
flow_time_us = np.zeros(len(logdata.channels["OF"]["TimeUS"].listData))
for i in range(len(logdata.channels["OF"]["TimeUS"].listData)):
(line, flow_time_us[i]) = logdata.channels["OF"]["TimeUS"].listData[i]
flow_qual = np.zeros(len(logdata.channels["OF"]["Qual"].listData))
for i in range(len(logdata.channels["OF"]["Qual"].listData)):
(line, flow_qual[i]) = logdata.channels["OF"]["Qual"].listData[i]
else:
FAIL()
self.result.statusMessage = "FAIL: no optical flow data\n"
return
# load required attitude data
if "ATT" in logdata.channels:
Roll = np.zeros(len(logdata.channels["ATT"]["Roll"].listData))
for i in range(len(logdata.channels["ATT"]["Roll"].listData)):
(line, Roll[i]) = logdata.channels["ATT"]["Roll"].listData[i]
Pitch = np.zeros(len(logdata.channels["ATT"]["Pitch"].listData))
for i in range(len(logdata.channels["ATT"]["Pitch"].listData)):
(line, Pitch[i]) = logdata.channels["ATT"]["Pitch"].listData[i]
att_time_us = np.zeros(len(logdata.channels["ATT"]["TimeUS"].listData))
for i in range(len(logdata.channels["ATT"]["TimeUS"].listData)):
(line, att_time_us[i]) = logdata.channels["ATT"]["TimeUS"].listData[i]
else:
FAIL()
self.result.statusMessage = "FAIL: no attitude data\n"
return
# calculate the start time for the roll calibration
startTime = int(0)
startRollIndex = int(0)
for i in range(len(Roll)):
if abs(Roll[i]) > tilt_threshold:
startTime = att_time_us[i]
break
for i in range(len(flow_time_us)):
if flow_time_us[i] > startTime:
startRollIndex = i
break
# calculate the end time for the roll calibration
endTime = int(0)
endRollIndex = int(0)
for i in range(len(Roll)-1,-1,-1):
if abs(Roll[i]) > tilt_threshold:
endTime = att_time_us[i]
break
for i in range(len(flow_time_us)-1,-1,-1):
if flow_time_us[i] < endTime:
endRollIndex = i
break
# check we have enough roll data points
if (endRollIndex - startRollIndex <= min_num_points):
FAIL()
self.result.statusMessage = "FAIL: insufficient roll data pointsa\n"
return
# resample roll test data excluding data before first movement and after last movement
# also exclude data where there is insufficient angular rate
flowX_resampled = []
bodyX_resampled = []
flowX_time_us_resampled = []
for i in range(len(Roll)):
if (i >= startRollIndex) and (i <= endRollIndex) and (abs(bodyX[i]) > min_rate_threshold) and (abs(bodyX[i]) < max_rate_threshold) and (flow_qual[i] > quality_threshold):
flowX_resampled.append(flowX[i])
bodyX_resampled.append(bodyX[i])
flowX_time_us_resampled.append(flow_time_us[i])
# calculate the start time for the pitch calibration
startTime = 0
startPitchIndex = int(0)
for i in range(len(Pitch)):
if abs(Pitch[i]) > tilt_threshold:
startTime = att_time_us[i]
break
for i in range(len(flow_time_us)):
if flow_time_us[i] > startTime:
startPitchIndex = i
break
# calculate the end time for the pitch calibration
endTime = 0
endPitchIndex = int(0)
for i in range(len(Pitch)-1,-1,-1):
if abs(Pitch[i]) > tilt_threshold:
endTime = att_time_us[i]
break
for i in range(len(flow_time_us)-1,-1,-1):
if flow_time_us[i] < endTime:
endPitchIndex = i
break
# check we have enough pitch data points
if (endPitchIndex - startPitchIndex <= min_num_points):
FAIL()
self.result.statusMessage = "FAIL: insufficient pitch data pointsa\n"
return
# resample pitch test data excluding data before first movement and after last movement
# also exclude data where there is insufficient or too much angular rate
flowY_resampled = []
bodyY_resampled = []
flowY_time_us_resampled = []
for i in range(len(Roll)):
if (i >= startPitchIndex) and (i <= endPitchIndex) and (abs(bodyY[i]) > min_rate_threshold) and (abs(bodyY[i]) < max_rate_threshold) and (flow_qual[i] > quality_threshold):
flowY_resampled.append(flowY[i])
bodyY_resampled.append(bodyY[i])
flowY_time_us_resampled.append(flow_time_us[i])
# fit a straight line to the flow vs body rate data and calculate the scale factor parameter required to achieve a slope of 1
coef_flow_x , cov_x = np.polyfit(bodyX_resampled,flowX_resampled,1,rcond=None, full=False, w=None, cov=True)
coef_flow_y , cov_y = np.polyfit(bodyY_resampled,flowY_resampled,1,rcond=None, full=False, w=None, cov=True)
# taking the exisiting scale factor parameters into account, calculate the parameter values reequired to achieve a unity slope
flow_fxscaler_new = int(1000 * (((1 + 0.001 * float(flow_fxscaler))/coef_flow_x[0] - 1)))
flow_fyscaler_new = int(1000 * (((1 + 0.001 * float(flow_fyscaler))/coef_flow_y[0] - 1)))
# Do a sanity check on the scale factor variance
if sqrt(cov_x[0][0]) > param_std_threshold or sqrt(cov_y[0][0]) > param_std_threshold:
FAIL()
self.result.statusMessage = "FAIL: inaccurate fit - poor quality or insufficient data\nFLOW_FXSCALER 1STD = %u\nFLOW_FYSCALER 1STD = %u\n" % (round(1000*sqrt(cov_x[0][0])),round(1000*sqrt(cov_y[0][0])))
# Do a sanity check on the scale factors
if abs(flow_fxscaler_new) > param_abs_threshold or abs(flow_fyscaler_new) > param_abs_threshold:
FAIL()
self.result.statusMessage = "FAIL: required scale factors are excessive\nFLOW_FXSCALER=%i\nFLOW_FYSCALER=%i\n" % (flow_fxscaler,flow_fyscaler)
# display recommended scale factors
self.result.statusMessage = "Set FLOW_FXSCALER to %i\nSet FLOW_FYSCALER to %i\n\nCal plots saved to flow_calibration.pdf\nCal parameters saved to flow_calibration.param\n\nFLOW_FXSCALER 1STD = %u\nFLOW_FYSCALER 1STD = %u\n" % (flow_fxscaler_new,flow_fyscaler_new,round(1000*sqrt(cov_x[0][0])),round(1000*sqrt(cov_y[0][0])))
# calculate fit display data
body_rate_display = [-max_rate_threshold,max_rate_threshold]
fit_coef_x = np.poly1d(coef_flow_x)
flowX_display = fit_coef_x(body_rate_display)
fit_coef_y = np.poly1d(coef_flow_y)
flowY_display = fit_coef_y(body_rate_display)
# plot and save calibration test points to PDF
from matplotlib.backends.backend_pdf import PdfPages
output_plot_filename = "flow_calibration.pdf"
pp = PdfPages(output_plot_filename)
plt.figure(1,figsize=(20,13))
plt.subplot(2,1,1)
plt.plot(bodyX_resampled,flowX_resampled,'b', linestyle=' ', marker='o',label="test points")
plt.plot(body_rate_display,flowX_display,'r',linewidth=2.5,label="linear fit")
plt.title('X axis flow rate vs gyro rate')
plt.ylabel('flow rate (rad/s)')
plt.xlabel('gyro rate (rad/sec)')
plt.grid()
plt.legend(loc='upper left')
# draw plots
plt.subplot(2,1,2)
plt.plot(bodyY_resampled,flowY_resampled,'b', linestyle=' ', marker='o',label="test points")
plt.plot(body_rate_display,flowY_display,'r',linewidth=2.5,label="linear fit")
plt.title('Y axis flow rate vs gyro rate')
plt.ylabel('flow rate (rad/s)')
plt.xlabel('gyro rate (rad/sec)')
plt.grid()
plt.legend(loc='upper left')
pp.savefig()
plt.figure(2,figsize=(20,13))
plt.subplot(2,1,1)
plt.plot(flow_time_us,flowX,'b',label="flow rate - all")
plt.plot(flow_time_us,bodyX,'r',label="gyro rate - all")
plt.plot(flowX_time_us_resampled,flowX_resampled,'c', linestyle=' ', marker='o',label="flow rate - used")
plt.plot(flowX_time_us_resampled,bodyX_resampled,'m', linestyle=' ', marker='o',label="gyro rate - used")
plt.title('X axis flow and body rate vs time')
plt.ylabel('rate (rad/s)')
plt.xlabel('time (usec)')
plt.grid()
plt.legend(loc='upper left')
# draw plots
plt.subplot(2,1,2)
plt.plot(flow_time_us,flowY,'b',label="flow rate - all")
plt.plot(flow_time_us,bodyY,'r',label="gyro rate - all")
plt.plot(flowY_time_us_resampled,flowY_resampled,'c', linestyle=' ', marker='o',label="flow rate - used")
plt.plot(flowY_time_us_resampled,bodyY_resampled,'m', linestyle=' ', marker='o',label="gyro rate - used")
plt.title('Y axis flow and body rate vs time')
plt.ylabel('rate (rad/s)')
plt.xlabel('time (usec)')
plt.grid()
plt.legend(loc='upper left')
pp.savefig()
# close the pdf file
pp.close()
# close all figures
plt.close("all")
# write correction parameters to file
test_results_filename = "flow_calibration.param"
file = open(test_results_filename,"w")
file.write("FLOW_FXSCALER"+" "+str(flow_fxscaler_new)+"\n")
file.write("FLOW_FYSCALER"+" "+str(flow_fyscaler_new)+"\n")
file.close()
except KeyError as e:
self.result.status = TestResult.StatusType.FAIL
self.result.statusMessage = str(e) + ' not found'