Real-time Sound Analysis to Count Opening Cycles of Automatic Doors

Abstract

Counting opening cycles on an automatic sliding door is of great interest for a company manufacturing doors, such as ASSA Abloy. These metrics could be used for consumer statistics or for door diagnostics. Counting opening cycles is seemingly trivial when there is access to the door’s internal diagnostics or having adequate sensors. Problems start to arise when these are not present, which is the case when working with third party door vendors, and sensors can often be expensive and time consuming to install.

This report investigates the process of counting opening cycles using real-time sound analysis on a microcontroller. Due to microphones having low cost and lack need for precise placement, ASSA Abloy is considering implementing this on their IoT gateway hardware, ready to be mounted on any third party door.

To achieve this, two different algorithms were tested - signal energy analysis and Mel Frequency Cepstrum Coefficient (MFCC) analysis. An implementation was also tested were both algorithms were used. These were then tested on four different kinds of automatic doors.

Signal energy detection performs well on all types of doors but is prone to false positives from external noise. MFCC detection is more resistant to false positives from a noisy environment, but often detects false positives during the closing of the door, and only gives good results for two doors. The combined algorithm similarly only performed well for two of the doors, but had the fewest false positives. These results show that in an ideal environment signal energy detection will suffice, but in noisy scenarios the combined algorithms show promise, assuming the MFCC detection can be improved or an initial calibration to the door is added.

Popular Abstract

Using sound analysis to count automatic door opening cycles:
Counting door cycles is useful for a company manufacturing automatic doors, especially if there is a solution that works for any automatic door. Different methods of sound analysis were tested on different types of doors to determine how best to detect opening cycles for any door.

Counting how many times an automatic door opens - the number of opening cycles, is a seemingly trivial task. For a door manufacturer, for example ASSA Abloy, this task is simple, as this information can be accessed with the internal electronics of the door. In certain cases, such as if ASSA Abloy needed to do maintenance on a door from a different company, this might not be possible. This can be problematic if usage statistics are desired by a customer, or if ASSA Abloy is contracted to perform maintenance for an entire door building containing various types of doors and want to use opening cycles as a diagnostic tool. Because of this, the need for external sensors arises. Placing sensors to detect when the door is opening can however be time consuming and expensive when doing it for many doors.

This is why this paper investigates the use of microphones for listening for door openings instead. Using microphones has two big advantages - low cost and ease of installation. Although an algorithm for doing this is inherently more complex to develop, it has the potential to reduce cost when applied at scale. The sound analysis is done on a microphone connected to a microcontroller which analyses the audio in real-time. By recording the sound of multiple opening cycles from different doors suitable parameters could be programmed into the microcontroller when tuning the algorithms.

Two different methods were tested - signal energy analysis and Mel Frequency Cepstrum Coefficient (MFCC) analysis. The signal energy analysis is a very simple algorithm detecting sustained periods of high signal energy - prone to many false positives from background noise, but rarely misses an opening cycle. The other algorithm is more sophisticated and is capable of differentiating between several different kinds of sound signatures. The MFCC algorithm is designed to mimic how humans interpret sound. This means that the difference in the MFCC output varies distinctly when for example a human says different vowels, which is why this algorithm is often used in speech processing, but has proven to be very useful when detecting door opening sounds as well. A downside to this algorithm is that they detect both the opening and closing of the doors, when only the door opening should be detected.

The energy-based algorithm worked best in quiet environments, while the MFCC-based algorithm only performed well on specific doors. This is due to how different the sound is between different doors, and it is difficult to make a generalised algorithm which works between multiple door types while not generating false positives. Using both algorithms at the same time also only had good results for half of the doors, but was the most resistant to false positives from background noise. In more realistic environments, the combined algorithms may be the most useful, if the MFCC algorithm can be improved for all doors, or if the microcontroller was calibrated for a doors opening sound before being put to use.

• author: Renman, Teodor and Ringström, Charlie
• supervisor: Maria Sandsten
• organization: Department of Electrical and Information Technology
• course: EITM01 20231
• year: 2023
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: Sound analysis, signal processing, event detection, MFCC, real-time, automatic door, embedded programming, stm32, microcontroller
• report number: LU/LTH-EIT 2023-944
• language: English
• id: 9128330
• date added to LUP: 2023-08-29 09:58:10
• date last changed: 2023-08-29 09:58:10