EMERGENCY VEHICLE ALERT SYSTEM STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH
This invention was made with government support under Grant 1R43DC03017-01 Al by the United States Department of Health and Human Services, Public
Health Service. The government has certain rights in this invention.
BACKGROUND OF THE INVENTION
The present invention relates generally to an apparatus for alerting automotive vehicle operators to the presence of an emergency vehicle and, more specifically, to an apparatus for detecting the approach of emergency vehicles by analyzing the sound from the sirens of an
emergency vehicle.
Hearing impaired people cannot hear sirens of an emergency vehicle as it
approaches. Unless the hearing impaired person sees the warning lights of the emergency
vehicle, they do not realize the presence of the emergency vehicle. Many hearing impaired drivers feel vulnerable because they cannot hear the emergency vehicle's siren.
Automakers desire the interior noise level in the vehicle to be reduced as much
as possible. Road noise, for example, contributes significantly to the interior noise level of the
vehicle. Other noises such as wind noise, sounds from adjacent cars and sounds from emergency
vehicle sirens are also significantly reduced within the interior the vehicle. When a radio is
playing within the interior of the vehicle, a siren from an emergency vehicle may be difficult to
detect. Thus, drivers of certain vehicles become in a sense "hearing impaired" with respect to emergency vehicle sirens.
Emergency service officials have recognized the problem that their vehicles may
go unnoticed. Drivers unaware of emergency vehicles pose a hazard to themselves as well as to the drivers of the emergency vehicles. Many alerting devices have been developed and tested. At present, several devices are on the market. No device, however, presently exists that fulfills the desired function satisfactorily. For example, NASA has developed a system where an RF transponder is installed in an emergency vehicle and a receiver is installed in the vehicle of the
hearing impaired driver. While the NASA system functions well, the cost of implementation is
expensive and therefore not practical except in very limited areas.
Other known emergency vehicle detection systems use sound detection to provide
a warning signal to drivers. Such devices use filtering techniques, including the use of a bank
of linear filters, or a correlation method (template method) with the storing of siren sounds.
These devices have proved unreliable. One device, for example, generates a high rate of false
positives. That is, the device responds to sounds other than emergency vehicle sirens such as
whistling or hand clapping. Another known device generates a high rate of false negatives. That is, the device does not respond when a siren is present.
In order for emergency vehicle alerting devices to be of any practical use to
hearing impaired drivers, significant improvement in the accuracy of the detection of emergency vehicles is required.
SUMMARY OF THE INVENTION
It is therefore one object of the present invention to provide an emergency vehicle alert system generating a reduced number of false and missed alarms.
One advantage of the present invention is that the emergency vehicle alert system is relatively inexpensive. Thus, it is believed that hearing impaired drivers will rapidly
implement the system on their vehicles thus making the roads safer for themselves and others.
One feature of the invention is that information may be displayed about the
direction from which an emergency vehicle is approaching while providing approximate distance of the approaching vehicle.
In one aspect of the invention, an emergency vehicle alert system has a sound
sensor that receives an audible sound and converts the sound into an electrical analog sound
signal corresponding to the sound. An analog-to-digital converter is coupled to the sound sensor
and converts the electrical analog sound signal into a digital sound signal corresponding to the
sound. A trained classifier has a data input coupled to the analog to digital converter and a set
of predetermined internal coefficients for processing the digital sound signal to generate an output signal indicative of the detection of an emergency vehicle. An alerting device is coupled
to the trained classifier for alerting a vehicle driver of the presence of an emergency vehicle.
In another aspect of the invention, the display has a plurality of rings divided in quarters. Each quarter represents a relative direction to the emergency vehicle. The concentric rings represent the relative distance to the emergency vehicle.
In another aspect, the present invention provides a method of alerting the presence
of an emergency vehicle comprising the steps of: converting an audible sound into an analog
electrical signal corresponding to the audible sound, converting the analog electrical signal into a digital sound signal, processing the digital sound signal through a trained classifier to generate
an output signal indicative of the detection of an emergency vehicle, and indicating the presence of an emergency vehicle.
BRIEF DESCRIPTION OF THE DRAWINGS
Other features and advantages of the present invention will become apparent in
the following detailed description which should be read in conjunction with the drawings, in which:
Figure 1 is a cut-away side view of an automotive vehicle having an emergency vehicle alert system of the present invention;
Figure 2 is a block diagram of the emergency vehicle alert system according to
the present invention;
Figure 3 is a schematic of the microphones coupled to the processing circuit and
display;
Figure 4 is a schematic of the display unit;
Figure 5 is a top view of an alternative embodiment of a microphone holding unit;
Figure 6 is a side view of the microphone holding unit of Figure 5;
Figure 7 is a flow chart of the signal processing system of the present invention;
Figure 8 is a frequency response of a conventional narrow band filter compared
with the effective frequency response of one neural network; and
Figure 9 is a plot of the superimposed effective frequency band of neural network
systems to detect various emergency vehicle sirens.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
In the following figures, like reference numerals will be used to represent like
components. The figures illustrate an emergency vehicle alert system having an illuminated
display. However, other types of display devices such as audible may be used according to the
vehicle driver's needs.
Referring now to Figure 1, an automotive vehicle 10 has a power supply 12 such as a 12 volt battery. An emergency vehicle alert system 14 has a sound sensor 16, processing circuitry 18 and an alerting device 20. Sound sensor 16 is preferably installed on the outside of
automotive vehicle 10. As shown, sound sensor 16 is installed on the deck lid of vehicle 10. Other locations such as on the roof, on the hood, or incorporated into a side rearview mirror may be suitable locations for sound sensor 16.
Sound sensor 16 is preferably a microphone unit. Preferably, three microphones are installed to form the sound sensor 16. In the preferred embodiment, electret microphones are used due to optimum sensitivity and frequency response. Sound sensor 16 houses the
microphones so that they are sealed against water, road spray and dust. Sound sensor 16
preferably also reduces vibration from the vehicle. Such adverse conditions may reduce the
reliability of the microphones.
A cable 22 couples sound sensor 16 to processing circuitry 18. Processing
circuitry 18 is also coupled to power supply 12. Power supply 12 is represented as the vehicle battery. However, power supply 12 may, for example, be a stepped down, regulated, and
conditioned power supply commonly used for electronic circuitry. Processing circuitry 18 will
be described further below.
Alerting device 20 is used to alert the driver of vehicle 10 to the presence of an
emergency vehicle siren. Alerting device 20 may, for example, provide an audible sound, a
visual display or a combination of the two used to signal the presence of an emergency vehicle.
For people with limited hearing, alerting device 20 may, for example, emit an audible tone or
sounds within the hearing range of the vehicle driver. Alerting device 20 preferably uses a visual display such as an LED, liquid crystal display or incandescent warning lights to signal the driver
of the presence of an emergency vehicle. Alerting device 20 may, for example, be incorporated into the instrument panel of the automotive vehicle. Alerting device 20 may also be coupled to the audio system of motor vehicle 10. Through the audio system, alerting device 20 may
broadcast a warning signal or reduce the volume of the radio.
Alerting device 20 may also be integrated into a commonly known heads-up
display system used in vehicles. Alerting device 20 may provide a simple warning such as a warning light on/off in the presence/absence of an automotive vehicle. More elaborately, alerting
device 20 may provide an indication of distance and direction to the emergency vehicle.
Referring now to Figure 2, sound sensor 16 preferably has three microphones 24.
By using three microphones 24, the system may detect direction. If only distance and presence
of an emergency vehicle is desired, only one microphone may be used. Sound sensor 16 is
coupled to processing circuitry 18, power supply 12 and alerting device 20.
Processing circuitry 18 preferably has an amplifier 26 and a band pass filter 28
for each microphone 24. Processing circuitry 18 has an analog-to-digital converter 30 coupled
between band pass filters 28 and a trained classifier 32. Trained classifier 32 is coupled to
alerting device 20. Trained classifier 32 preferably has a memory 34 and a microcomputer 36.
Memory 34 may be part of microcomputer or a separate component. Microphones 24 convert
the sound signal to an analog electric signal corresponding to the sound.
Amplifiers 26 receive the analog sound signal and amplify the signal to a predetermined level. Amplifiers 26 may be coupled to microcomputer 36. Microcomputer 36 may be used to change the gain of amplifiers 26. The gain of amplifiers 26 may be adjusted to
a desired level. It is preferred that the gain of amplifiers 26 is adjusted so that the analog sound
signal does not saturate analog-to-digital converter 30.
Once amplified by amplifier 26, the analog sound signal is filtered by band pass
filters 28. Because most road noise resides in the lower frequency region of the audible spectrum, that is below one kilohertz, the band pass filter is design to filter out frequencies below one kilohertz. Also, it has been found that trained classifier 32 responds primarily to the
harmonics of the siren in the one kilohertz to five kilohertz band of the audible spectrum. Thus, band pass filter 28 is designed to also suppress sound above five kilohertz. In the flat band of
band pass filter 28, the acoustic region between one kilohertz and five kilohertz is either flat or only slightly increased in that region. Band pass filter 28 provides the greatest signal to noise
ratio in the presence of ambient noise.
Analog-to-digital converter 30 samples analog sound signals from band pass filter 28. Analog-to-digital converter 30 converts the analog sound signal into a digital sound signal.
The digital sound signal provides input vectors for trained classifier 32. As will be described
below, depending on the trained classifier 32, various timings between the input vectors from
analog-to-digital converter 30 may be used. A time interval equal to 32 analog-to-digital output
samples may be appropriate input vectors. For example, an input vector consisting of 256
consecutive output samples from analog-to-digital converter 30 may be fed into trained classifier
32 and then 32 sample points later, another vector consisting of the next 256 samples may be fed to the trained classifier 32.
Trained classifier 32 is preferably of a neural network configuration. One such
configuration is a conventional feed-forward network. The other configuration is a finite impulse response neural network. The neural network can be simulated using preexisting digital signal processing (DSP) chips. Microcomputer 36 may, for example, be a TMS 320C50 DSP designed to load classification coefficients as needed from memory 34 into an on-chip memory. Once the classification coefficients are loaded, the microprocessor may access them at full speed. Of course, the neural network may be masked permanently onto a custom made DSP chip. Memory
is preferably an EPROM.
Trained classifier 32 receives input from analog to digital converter 30 to
determine the presence of an emergency vehicle based upon the characteristics of the sound from the siren of the emergency vehicle. Microcomputer 36 may also be used to calculate the direction
and distance from the emergency vehicle and initiated an alert by alerting device 20.
Trained classifier 32 may be trained in a conventional manner. That is various
types of sirens may be input to the system. The internal coefficients of trained classifier 32 are
recursively adjusted until the coefficients determine the presence of a siren within a desired value
of a percentage of correct determinations.
Referring now to Figure 3, an alerting device 20 is shown coupled to sound sensor
16. Alerting device 20 may also be in a "black box" with processing circuitry 18. Alerting
device 20 preferably has a display 38 used to convey distance and direction of the emergency
vehicle.
Sound sensor 16 has a base 40 used for mounting sound sensor 16 to automotive
vehicle 10. Base 40 has a plurality of arms 42 extending therefrom. As described above, emergency vehicle alert system preferably has three microphones 24. Microphones 24 are
located at the end of each arm 42. For ease of calculating techniques, microphones 24 preferably form the vertices of an equilateral triangle.
Referring now to Figures 3 and 4, display 38 is formed of an inner ring 44 and an
outer ring 46. Preferably inner ring 44 and outer ring 46 are concentric. Rings 44 and 46 provide
an indication of distance from emergency vehicle. For example, outer ring 46 may indicate that the emergency vehicle is less than 150 meters from the vehicle. Inner ring 44 may indicate that the vehicle is closer than 150 meters. A center portion 48 may also be used in conjunction with
the inner ring to illustrate that the vehicle is very close, for example, less than 50 meters away.
Inner ring 44 and outer ring 46 are divided into quadrants that preferably
correspond to the direction from which the emergency vehicle is approaching. As shown, the top
quadrant corresponds to the emergency vehicle approaching from forward of the vehicle, the right
quadrant indicates the emergency vehicle is approaching from the right, the left quadrant
indicates the emergency vehicle is approaching from the left and the lower quadrants indicates the emergency vehicle is approaching from behind the automotive vehicle.
Display 38 may be formed of a plurality of lights or LEDs which form the quadrants of inner ring 44 and outer ring 46.
Referring now to Figures 5 and 6, sound sensor 16 is shown having a unique
ornamental design. Microphones 24 are preferably enclosed within end caps 50. End caps 50 are rounded to improve the aesthetics of the sound sensor 15 as in Figure 3. End caps 50 and
therefore microphones 24 preferably form an equilateral triangle. Arms 42 may also be stylized. Arms 42 may, for example, be curved along their lengths as illustrated to improved aesthetics.
Base 40 may also be rounded to improve the overall appearance of sound sensor 16.
Referring now to Figure 7, the types of sirens are well regulated. There are three types of sirens that are almost exclusively used. The frequency modulation rate varies among
the types of sirens. For example, a high-low type siren possesses a frequency modulation rate of between 40-60 cycles per minute. The wail type siren has a modulation rate of between 15-30
cycles per minute. The yelp siren has a modulation rate of between 160-240 cycles per minute.
The National Institute of Justice sets standards for emergency vehicle sirens. Class A sirens are
required to generate 120 decibels (RMS) of acoustic power at 3 meters from the source. This
reduces to about 80 decibels at a distance of 1/4 mile. Normal highway road noise generates on
average between 74-78 decibels of acoustic power. Therefore, a siren sound of 80 decibels at a
distance of 300 meters may be detected in the absence of extremely loud noises. This response is similar to that of a human listener trying to detect a siren.
In Figure 7, step 100 represents signal processing that takes place before the
trained classifier. Signal processing may include amplification from amplifiers 26, filtering from
band pass filters 28 and analog-to-digital conversion by A-to-D converter 30 all of Figure 2. To detect the presence of a siren, only one microphone output need be used.
In step 110, a fast modulating siren is detected. A portion of the digital sound signal is processed. Preferably, an individual neural network system for each type of siren is
used. Either a feed forward or finite impulse response network may be used. To detect the fast
modulating siren such as a yelp siren, 128 inputs, 7 hidden layer units and 1 output unit are used in a dedicated neural network to detect the fast modulating siren. Step 112 sends the results to decision block 122. The results are an intermediate signal that indicates that the sampled portion of the digital signal indicates the presence or absence of a fast modulating siren.
Step 114 detects the presence of a medium modulating siren. A medium
modulating siren, for example, may be a high-low type siren. A neural network suitable for a
high-low type siren, for example, may be a 256 input, 7 hidden layer units and 1 output unit. As
above, each of the 256 inputs is an output of the analog-to-digital converter . Preferably, the same portion of the digital sound signal is used as in the fast modulating siren detection. Block
116 sends the results to decision block 122. The results are an intermediate signal that indicates that the sampled portion of the digital signal indicates the presence or absence of a medium
modulating siren.
In step 118, the presence of a slow modulating siren is determined. A slow
modulating is of the wail type siren. A suitable neural network for detecting a slow modulating
siren may have 256 inputs 14 hidden layer units and 1 output unit. Preferably, the same portion
of the digital sound signal is used as in the fast and medium modulating siren detection. The
results of step 118 are also sent to decision block 122. The results are an intermediate signal that indicates that the sampled portion of the digital signal indicates the presence or absence of
a slow modulating siren.
Because the portion of the digital sound signal is the same, the process may be run virtually simultaneously through steps 110, 114 and 118. This allows an overall faster processing
time.
In step 122 if sufficient data has not been obtained to determine the presence of
one of the types of sirens, the system is directed back to step 110 wherein steps 110-120 are repeated with another portion of the digital sound signal. Because only a small portion of a
sound signal is used, several cycles through steps 110-122 may be needed to obtain a sufficient
portion of the sound signal to accurately detect the presence of an emergency vehicle. In one
constructed embodiment, the processing time for steps 110-122 was equivalent to about 32 bits of information from the signal processing step 100. The next set of 256 inputs to the neural
networks of steps 110, 114 and 118 were thus about 32 bits after the end of the previous set of
bits. One skilled in the art would recognize that the delay between samples may be adjusted to
have a greater delay or may be adjusted to have essentially no delay.
In step 122 if a signal display is used, step 130 may be executed to alert the driver of the presence of an emergency vehicle. If, however, more complete data such as distance and direction is desired to be displayed to the driver, step 124 is executed.
In step 124, data is obtained from all the microphones. As mentioned above, it is preferred that three microphones are used.
In step 126, the direction of the emergency vehicle may be determined. The direction may be obtained by well-known correlation techniques or by the following method. To obtain the direction, each of the inputs from the three microphones network are utilized. The digital sound signal of each of the microphones is used. First, the output data from a pair of
microphones are analyzed. The differences between corresponding points of the two signals are squared and then summed. The sum is proportional to the arrival time difference of the two
signals which is proportional to the sine function of a direction angle with respect to a line connecting the two microphones. In such a manner, a set of four data points is obtained, as the solution to the direction from which the sound is coming. Thus, a comparison using the third
microphone must be used to determine the direction from which the emergency vehicle is
approaching. The relationship between the arrival time Dt, and the angle (S) is given by S=
arcsine (D/(Dt*C)) where, where D is the distance between the two microphones and C is the
velocity of sound. The four data points obtained from the first comparison between two microphones may be narrowed by doing a sum of differences calculation using a time shifted
sample for one of the microphones. The output of one microphone may be compared with the
output of another microphone shifted one point earlier with respect to the other. Depending on
whether the sum of the difference squared is greater or less than the original sum, the four data
points may then be narrowed to two data points.
The above method has been found to be accurate within ten degrees using microphones placed six inches apart with the microphones at the vertices of an equilateral
triangle. Interpolation may be used if further accuracy is desired.
After direction is determined and if direction is the only information desired to
be displayed, then step 130 is executed to display only direction information. However, if the
distance of an emergency vehicle is desired to be transmitted to the vehicle driver, step 128 is performed. In step 128, the distance may be determined by monitoring the level of sound
received from the siren of the emergency vehicle. A simple distance estimate can therefore be established by monitoring the amplitude of the sound signal once the presence of the emergency
vehicle is detected. For example, for sound less than 86 decibels the far indicator (such as outer
ring 46) may be illuminated. If the sound amplitude is between 86 and 95.5 decibels, the distance is between 150 meters and 50 meters. The inner ring 44 would then be illuminated. Center 48
would then be illuminated. If the sound is greater than 95.5 decibels, the emergency vehicle is
at 50 meters or less.
If a single neural network is utilized to detect all types of sirens, steps 110 through
120 may be combined into a single step. Both a finite impulse response and feed forward
network may be implemented using a single network. The finite impulse response network
requires less computation than the feed forward configuration. However, accuracy may not be as great due to the fact that each of the three significantly differing sirens must be detected by the
single neural network.
In operation, the display should be located in an easily seen location such as on the instrument panel or incorporated as part of a "heads up" display. The preferred display merely alerts the driver to whether the emergency vehicle approaching from either side of the
vehicle, from the front of the vehicle or the back of the vehicle.
Referring now to Figure 8, the probability of detection is plotted versus modulation frequency for a linear matched filter type and a neural network. As shown, the neural detector determines proper identification of a siren frequency range better than that using a match filter of the type used in the prior art.
Referring now to Figure 9, the probability of detection is illustrated for three
frequency ranges using three networks as described above in a feed forward type neural network system.
It should be understood by those skilled in the art that variations and modifications
to the preferred embodiments described above may be made without departing from the true scope of the invention as defined by the following claims: