1

Human Step Detection and Counting with a
Microphone (June 2015)
Juan Carlos Batista Sundström*, Hossam Salman** and Hossein Yazbek***

Abstract—nowadays it is common with the alternative use of sensors for detecting human movements such as footsteps and that is done for various reasons such as security or just for lightning a lamp automatically. We developed a Simulink model in Matlab to simulate a system that analyses the footsteps of three 25 years old men. Those men had different heights and weights. The data were recorded and analyzed using filtering and conditioning blocks of
Matlab. The System collected 3 sets of steps. The first set had 5 steps with 5 detections. The second set had 8 steps with 3 detection and the third set had 4 steps with 1 detections. In total, there were
17 steps where 9 steps were detected.
I.

C. Procedure
During the first part of the experiment, the footsteps were recorded on the main corridor of the first floor of the house 21, located in the University of Kristianstad. We also recorded the steps at our respective houses. The experiment was carried on a floor without carpet to allow a better collection of the data. This procedure was repeated several times, until satisfactory data without much noisy could be acquired.

INTRODUCTION

T

HE objective of this work was to detect the steps of a person using a microphone array embedded in our computer together with the Simulink library of Matlab.
We have some background of the idea after reading a few articles regarding this type of experiments.

Fig. 1. Simulink models used to record the steps

In the articles we read, the experiment was connected to fall detection. Whereas we experimented footsteps detection, a bit similar but not exactly the same.
An example is in the article “A microphone array system1”, where they used a camera to record the samples. We decided to only use a microphone to record the steps.
II. METHOD
A. Participants
Our experiment involved all members of our group where all hade their steps recorded. Our members consisted of three 25 years old men who weigh 95 kg, 79 kg, and 73 kg. Their heights were 176 cm, 168 cm, and 170 cm.
B. Material
The material used to collect the data in the experiment consisted of a microphone embedded in the computer used in the
Experiment. The software used to collect and save the data was
Matlab version R2015a. In Matlab the “Sinks” and the
“Sources” located in the DSP toolbox of the Simulink library were used to record and save the data in a file.

This Paper submitted June 08, 2015
* Juan Carlos Batista Sundström, student of the second year of the computerengineering program in the University of Kristianstad, Sweden.
** Hossein Yazbek student of the second year of the computer-engineering program in the University of Kristianstad, Sweden.

Fig 2. Simulink model of the system created to detect the footsteps in the signal.

During the second part of the experiment, we used Simulink in
Matlab to design a digital high-pass filter to make the signals from the data smoother. As we were 3 members, we recorded 3 different sets of steps. In total we had 17 steps. For each set of steps we had a different filter, because the signals were very different from each other. We used the “Digital Filter Design” block from simulink to design the filters. The filters were all finite impulse response filters. The calibration of the filters to allow better detection was made just by changing the parameters “Fstop” in the frequency and the parameters
“Apass” in magnitude. The desity factor was 17, The frequency

*** Hossam Salman, student of the second year of the computer-engineering program in the University of Kristianstad, Sweden.

2 units were kept in Hz, and The sample frequency in 44100 Hz.
The Magnitude units were kept in dB.
For the set of steps of the group member that weighs 95 kg and is 176 cm, the “Fstop” parameter was equal to 7000 Hz. The
“Apass” parameter was equal to 0.0000700 dB.
For the second set of steps which is from our member who weighs 79 kg and is 168 cm, the “Fstop” parameter was equal to 10600 Hz and “Apass” paramenter was equal to 0.0001 and the same parameters were kept in the third set of steps from the group member who weighs 73 kg and is 170 cm. In all sets of steps, the “Fpass” parameter was kept at 12000 Hz.

results were obtained with frequent and infrequent elevations in the amplitude of the signal. The high elevations that is mentioned bellow were the elevations with amplitude values equal or bigger than the constant value inside the “Compare to
Constant” block in Simulink.
After filtering the first set of steps, which comes from the group member who weighs 95 kg and is 176 cm, the signal was strong and consistent with frequent high elevations in the amplitude.

After passing through the filter, the signals would follow the following algorithm:
1. Input the signal n
2. Define the highest levels in the amplitude of the signal n for detection as DetectionLevel
3. for i=1 to n-length do
4. if ni >= DetectionLevel then
4.2 DetectedStepi ← 1
4.3 Downsample DetectedStepi by specific tested factor 4.4 DetectedStepSum =DetectedStepi +
DetectedStepi-1
5. Output DetectedStepSum
The Block “Compare to Constant” was used to implement the if-statement present in point 4. of the algorithm. The variables
“n” and “DetectionLevel” define the conditioning of the ifstatement in the algorithm. The value of n was the incoming signal and the DetectionLevel was predefined after analysis of the whole signal with the “Time Scope” block. This predefined value was filled in the parameter “Constant Value” of the
“Compare to Constant” block. The predefined value for the first set of steps was 0.6648, for the second and third sets the value was 0.18.
The implementation of the point 4.3 of the algorithm was done using the “Downsample” block of Simulink. The use of this block was necessary to decrease the number of samples in the signal before it could be summed. In that way, over summing could be avoid after the signal passed through the “Sum” block, which was used to implement step 4.4 algorithm. This block requires a factor and different values for this factor, was used for the different sets. In the first set the value was 60, in the second the value was 13 and in the third the value was 15.
The difference between the signals is the reason why different values were used in both the “DownSample” and “Compare to
Constant” blocks
III.

RESULTS

After the whole process of filtering and conditioning the signal for detection and summing, a link to the signal output after the conditioning, was connected together with a link of the summing into a “Time-Scope” Block of Matlab. Different

Fig. 3. Signal of the first set of steps after filtering.

After passing the “Compare to Constant” block, 5 high elevations in the amplitude were detected and summed by the summer block. The sum reached 5.

Fig. 4. The thick line in black represents the counting of the signal.
The lilac lines represents the signal after passing the conditioning.

After filtering the second set of steps, which comes from the group member who weighs 79 kg and is 168 cm, the signal was strong and consistent but with infrequent high levels in the amplitude. Fig. 5. Signal of the second set after filtering

After passing the “Compare to Constant” block, 3 high elevations in the amplitude were detected and summed by the summer block. The sum reached 3.

3 distance in which the microphone was recording in the different occasions. The number of microphones used in a single room is also a factor that could have influenced the bad quality of some signals. Our study also showed that the weight of our group members could have had some influence in the detection with the heaviest member of the group having the most accurate detection. Fig. 6. The second set could only detect and count 3 steps of 8.

After filtering the third set of steps, which comes from the group member who weighs 73 kg and is 170 cm, the signal was not consistent, with only one high elevation in the amplitude in relation to the other small elevations.

After passing through the “Compare to Constant” block, there were many elevations rising close to each other in some short intervals (see the lilac lines in fig.4 and 8). The created model saw them as only one elevation rising only once after summing, that because of the use of the “Downsample” block, which decreased the number of samples in that single interval to one, so it could interpret the whole interval of elevation as one step. we could learn that it is possible to detect steps, using conditioning of amplitude with the “Compare to Constant” block in Simulink. Our model detected 9 of 17 steps, it means
53% of success in the detection. That clear limitation exist because in our study, the steps could only be detected with signals reaching a specific amplitude. It means that some steps were missed by the detection. The solution for this problem could be the use of more than one “Compare to Constant” and
“Downsample” blocks with different constant values and factors values.

Fig. 7. Signal of the third set after filtering

After conditioning in the “Compare to Constant” block, 1 high elevations in the amplitude were detected and summed by the summer block. The sum reached 1.

Our study could have been more reliable if the experiments was applied with more people of different genders and ages. Besides the experiment, could even be done on a floor with carpet, with participants using different kind of shoes and sandals or even barefoot. REFERENCES

[1] Yun Li, K. Ho and M. Popescu, 'A Microphone Array System for Automatic
Fall Detection', IEEE Transactions on Biomedical Engineering, vol. 59, no. 5, pp. 1291-1301, 2012.

Fig. 8. The graph shows how the third set was inaccurate by showing only the detection of one step.

IV. DISCUSSION AND CONCLUSION
After analyzing the different graphs, it was possible to verify that the high levels in the signal amplitude represented the steps of the group members. The reason why the elevations were so high, frequent and infrequent in specific intervals is unclear.
Many reasons can be associated to this. Like the angle and

[2]S. Gasparrini, E. Cippitelli, S. Spinsante and E. Gambi, 'A Depth-Based Fall
Detection System Using a Kinect® Sensor', Sensors, vol. 14, no. 2, pp. 27562775,
2014.
[3]M. Salman Khan, M. Yu, P. Feng, L. Wang and J. Chambers, 'An unsupervised acoustic fall detection system using source separation for sound interference suppression', Signal Processing, vol. 110, pp. 199-210, 2015.
Basic format for periodicals: