Driving-While-Impaired,
Remote-Monitoring of 

---

                By Glen A. Williamosn              1992 

All of the pylons are tied together, via a half-duplex (bidirectional) IR optical link to a central or supervisory computer. The host computer is a fast PC workstation with large disk storage and is able to process measurement data at a rapid rate. The system has the feature of adding workstations or servers, as the work load requires. 
 

Rolling Alley or Gauntlet

Disposable Instrumented Pylons

Parameters Measured
Because of its remote sensing nature, the system must deduce those parameters that would be directly measurable in the vehicle per se. Making measurements with sufficient resolution such that the front-end steering changes are distinct from the rear-end track of the vehicle is required. The analysis of such data should yield a sufficient characterization of the actual steering wheel changes, for example. 

Among the parameters measured are the following: 

Gross accelerations and decelerations. 

Micro-accelerations and micro-decelerations: 
 When attempting to hold a constant speed, especially a relatively slow speed, there is an attentiveness required to  maintaining that single speed which can be deduced. This "station-keeping," or lack there of, has a direct relationship to  the driver's condition. 

Gross yaw: gross steering 

Micro-directions: 
 Vector angle or yaw, and rate of vector swing or yawing are of interest because they represent fine steering (minute deviation from intended direction). 

Overall lane tracking score: 
 Scoring the driver's overall tracking (follow the white line) by plotting: should approximate an irregular sinusoid or smoothed trapezoid, about a zero axis. 
 

Directed and Non-directed Examination 
When evaluating drivers for DWI in a "Rolling-Ally" lane arrangement, there are two approaches that can be used: driver active and driver passive. The driver active approach requires that the driver go through several directed "problems," such as measuring how long it takes the driver to apply his/her brakes after a stop light turns red; and then when directed by the same light, starting off from a standing start, while being measured for smoothness of acceleration, as well as overshooting the desired top speed (25 MPH), or conversely, they may take an inordinately long time to reach top speed . Also, they might be directed--by signal--to change or shift lanes (simple slalom), hit the horn, etc., as they proceed. These "maze" type tests could be optionally invoked near the end of the gauntlet in those questionable cases where the computer (or the operator) needs more definitive data. One of the drawbacks to this approach is throughput, or the rate at which vehicles could be processed. Of course, this too could be adaptive: as the traffic gets heavier, the number of different "problems" are reduced. 

Driver passive, would require no directed driver actions other than following the lanes' path, while observing the speed limit, while being measured for constancy of speed--cruising. 

In order to tax the driver, sufficient to provoke better cues to performance: a (standardized) curved path or "meander," or even a mild form of Slalom might be appropriate. This type of layout would require the driver to negotiate the changing lane geometry with some finesse, while providing an opportunity to measure how well he/she maintains a track--as compared to others. 

--DWI Driver Profile
It is important to understand the drunk or impaired driver and their responses to various situations or stimuli. For example: An impaired driver may drive too fast because of their removed inhibitions, etc.; or they may drive too slowly, because of a feeling of paranoia or fear of arrest--being over-cautious. The problem is, that there are variously different and similar responses from an impaired driver. The trick is to catalog and categorize them is such a way that a reliable match can be made in the real world (with minimum false positives). 
Having said all that: the first order effect will be what they do with their vehicle, not their personality traits while impaired. The detection of the DWI driver will depend on statistical differences between what he/she does with their vehicle: as compared to what other impaired and unimpaired drivers do with theirs. 

___________ Evaluation Modality 

The data, when measured and evaluated against known norms of performance--standardize templates, for example, will give precise information as to the driver's deviations from a selected set of standards. 

These data will be quantified and examined statistically against statistically valid templates of average unimpaired and impaired drivers for a consistent grading of the driver's motor skill and coordination, performance. The results of such comparisons will yield a numbered value indicating the probability of impairment on some standardized rating scale. These scores are subject to the mitigation of various things related to the season of the year, weather, time of day, etc. 

There could also be a "dynamic" template, one that is a tally of the last n drivers' measured performance. It's a little like being graded on the curve. One possible advantage might be in cancelling out those mitigating variables. 

The Doppler transducer's output is an analog, audio frequency signal, that represents the "beat" difference between the ~ 50 kHz (24kHz to 66kHz) incident ultrasonic signal and the reflected--Doppler-shifted (velocity shifted)--signal. This signal is first band-limited by an anti-alias lowpass filter; digitized by an inexpensive Analog to Digital Convertor, ADC, (serial data out). This digital data is buffered and formatted, along with "pinger" data, and sent to the USART of the IR transponder, which in turn, transmits it (line-of-sight) to the system's computer at the appropriate time. 

Central Control and Communications 
Each instrumented pylon is a self-contained unit which is interchangeable with any other similar pylon. It has the ability to establish bidirectional (half-duplex) communication with the instrumentation computer (in the van), via an infrared (IR) transponder mounted on top. Each pylon's transponder is uniquely addressable (by dip switch) in a round-robin of sequentially polled pylons. Due to the nature of the Doppler transducer's time constraints (low beat frequencies for small changes in velocity: take longer to capture), and the possibility of interference from adjacent transducers: no two neighboring Doppler transducers are active at the same time. To further minimize interference, the transducers use frequency diversity or channelization (eight different bands: 24kHz to 66kHz, in 8 increments of 6 kHz each) , which also allows longer on-time; thus greater resolution. The information is polled in a fashion that facilitates this separation; in addition, it might be possible to poll in a pseudo-random sequence, such that it will maintain proper separation while helping to reduce sub-sample aliasing. When interrogated, the pylon turns on its ultrasonic emitter (starts its measurement) and simultaneously confirms this addressed command by sending back an acknowledgment in the form of its own address, the previously sampled data (any number of previously stored samples, held by the pylon), and after completing this measurement it sends that data, and then shuts down and awaits its next interrogation (the previous data is for redundancy and costs nothing). To minimize errors, the data is block coded such that if an error occurs that cannot be corrected, the data will be resent. Channalization may allow the system to run (Doppler RADAR) continuously, without interference problems. 

 --Grand Central Station
The heart of this system is the Command, Control and Communication, C3, or centralized control room (or van), where all of the data is received and processed, and the tickets are handed out! 

The center is run by operators (officers) who run the semiautomatic monitoring system with minimum intervention. Their job is to setup, install and operate the temporary installation and to oversee its, mostly automated operation: intervening only occasionally if needed. The real heart of the operation are the computers that are programmed to do the data reduction and to put the decision making information at the authorities' disposal. 

Data Reduction & Signal Analysis using DSPs 
The data reduction will be done on fast PC workstations equipped with ranks of 32 bit DSP coprocessor boards, along with several large hard disk files (486/50, 586/66 DX computer). The data must be reduced fast enough in order not to slow the traffic appreciably: this is the job of the DSP boards, while the host computer supervises the operations and handles the communications and control. Additional computers, or servers can be added as needed. This flexibility (modularity) allows the same hardware that was used in a small operation yesterday, to be part of a larger operation today. 

When the data is first received it must be separated out and tagged as belonging to specific identified vehicles (an operator IDs each vehicle as it enters the pipeline, by typing in the tag number, state and vehicle category (pick up, van, coupe, etc.). The data, once tagged, is further separated into Doppler signature data and position or pinger data. The Doppler data is processed by the DSP boards using FFT (fast Fourier transform) algorithms to decipher the complex spectrum of the Doppler shifted signals at each pylon. 

When analyzing the Doppler-shift data, separating the acceleration vector data from the yaw vector data will be helped by the pinger information, in identifying how much of the vector sum is lateral movement and how much is due to acceleration. As "scores" are deriving, they are segregated into separate bins (memory), each having a particular vehicle's ID on it. These bins are further separated into lateral and forward directions data (steering and acceleration). Simultaneously, as the data accumulates, it is being checked against the base line templates for a running-measure of delta or deviation from the norm. As these numbers mature, the trends are reported, in a continuous fashion, to the operators at Grand Central Station--via their GUI (Graphical User Interface)--for what ever action is appropriate. 

The measurement quantities can be peak to peak instantaneous deviations, average and Root-Mean-Square (RMS), deviations; Power spectral density distributions. 

Purity of Measurements 
When the FFTs are performed on the Doppler data, there is an attempt to segregate the data by direction, i.e., the lateral data from the forward data. The purity of this separation is of some importance, but it is not clear what the degree of purity should be. If data of some known purity is used in a "template match" comparison against a template of similar purity: will the answer be any different in a similar comparison if only the purity of the data and the template changes? An interesting question. 

Conclusion
Determining the level of driver Impairment to an equivalency of having a blood-alcohol level considered in violation, may turn out to be relatively easy--or not. 
 

Four Areas of Effort in Detecting DWI 

There are four discrete categories or disciplines that, when used in concert, work toward solving the problem at hand--DWI detection. Each discipline can be changed or improved upon without altering, either the other disciplines or the nature of the experiment. 
There is overlap or fuzzy boundaries between each, but that insures nothing is missed. 

Sensing: 
Doppler Radar and DME pinger. And "time-to-stop" timer for directed stopping. 

Data gathering, measuring and quantifying (XYZ): 
ADC, IR link, FFT (DSP), quantities, etc. 

Data reduction, identifying trends: 
peak--peak, average, RMS, integration, differentiation, etc. 

Template matching: Solving for DWI: 
Correlation, Histogram, Subtraction, Summing, Standard Deviation, etc. 

Vehicle Borne, DWI Detection and Alerting 

A vehicle borne method would enlist the vehicle's own engine management computer to monitor the driver's performance. 
From an instrumentation standpoint, the vehicle borne method is superior to the remote monitoring method: because of the direct sensing of those things (steering wheel, fine steering; brake pressure; gas pedal position, acceleration; etc.) that a remote monitor is attempting to deduce. The resolution of the direct measurement approach is absolute (relatively), as opposed to the resolution that is limited by the several constraints, including the limited time of measurement (Doppler resolution), as well as the finite resolution of that type of measurement. 

The data derived from this, continuous measurement, would be used in a fashion that would continually update--and compare against--an adaptive template, one derived from the present driver's own past driving performance (an historical template), as well as a standardized performance template (one statically based on the population as a whole, samples of impaired and unimpaired). Using the drivers own evolving record of performance and continually testing for quantifiable deviations, holds great promise of being an effective technique. However, in the unlikely event that the driver operated the vehicle in an impaired condition a significant part of his/her driving life: it would fail. But, there is still the possibility it would be caught by the simultaneous testing against the standardized performance template. 

When the vehicle's computer notes a differential in performance profile, above some predetermined value, it can initiate a protocol of "notification of DWI." This protocol could range from a silent alarm which could causes, for example, the CHMSL (LED brake light) light send short data bursts that cannot be seen (too short in duration), but can be detected by law enforcement using simple and inexpensive detectors on their vehicles. Another alerting method might be the use of the vehicle's integral cellular phone (which could be circumvented by disabling the phone). It could automatically take control of the phone and dial a special number and identify the vehicle (prerecorded voice or cellular modem burst) by tag number. 

A final and certainly controversial scheme, would be to have that vehicle's parking lights blink at some identifying rate or pattern: signifying to the world, its status (this will keep the lawyers busy). Case law would have to determine what level constitutes "probable cause," in order to have a police officer detain, interview and test the driver. 

In deciding this type of thing, some Jurist will have to weight the difference between the feelings of an outraged citizen at being publicly embarrassed and "humiliated;" and the outrage and desolation of parents at hearing that their child has just been hit and killed by that unchallenged driver who was Driving While Impaired.
 

______________Testing the Design 

Determining the degree of impairment of a driver at a distance is an imponderable. First of all: can it be done? If so, then how do you do it? After designing--on paper--a way that is deemed to be the best derivable from the finite choices that technology has available: it must stand the test, of--the test. There is great power in, just starting. The design must be tested, and the results of such testing used to improve the original design: which must be tested, and the results of such testing used to improve the original design: 
which must be tested, and the results of such testing used to improve the original design ... and so it goes, until an optimum alley has emerged along with a "good" experiment. 

Part of the evolution of system design is testing, and the results being fed back to alter and improve the design. The separation between research and testing the design in this area is fuzzy, to say the least. 
 

 --"Rolling Alley" Standardization 
The lane or "rolling alley ," configuration, what ever it turns out to look like, should always be setup in the same arrangement or arrangements (there may be more than one preferred way) when used as a roadside monitor. 

Questions to be answered: 
 1.. What is a typical Gauntlet configuration: is "n" feet in length, and has a total of "m" instrumented pylons; spaced at "k" foot    intervals. 

 2.. The length of the alley is to be determined. 200' --> 350' ??? 

 3.. The spacing between pylons, in car lengths: .75cl --> 2.0cl ??? 

 4.. Total number of pylons (n per side x 2 = T): 200' @ 2.0cl = 16, 200' @ .75cl = 32, 350' @ 2.0cl = 22, 350' @ .75cl = 58 

 5.. What is the optimum speed through the gauntlet? 25MPH ??? 

 6.. Staggering of the pylons (two parallel rows): what effect will that have on the total number needed (will it reduce the number),   will it effect the required spacing between pylons, and finally, will it alter the effectiveness of the overall system? 

 7.. Vehicle spacing will be important. How do you prevent tail-gating? 

The passive driver will follow the marked lane's path while observing the speed limit of the examination: there could be a standardized curving path or "meander." 

Directed stopping 
Time to braking where the subject driver is timed as to how long it takes him/her to start applying brakes (as indicated by their brake light turning on and being sensed remotely, which "stops the clock") after getting a visual signal to do so (traffic light). 

Directed starting 
This "problem," is a great opportunity for gaining data on the driver's performance: 

1.. Smoothness of acceleration from a standing start. 

2.. Overshooting the requested top speed in scenario. 

3.. Conversely, a driver may take an inordinately long time to reach this top speed. 

  A plot of acceleration over time would yield great information: paying special attention to the "impulse response" of this function; it will be either over-damped, under-damped, or normal. Normal is good! 

A Slalom, or markedly curved driving course that because of its layout, requires the driver to negotiate the changing lane geometry, and a measure of how well the driver maintains a track--as compared to others, is a good experiment. 
 

Testing & Template Construction 

The building of standardized templates will require data taking of two distinct populations, with gradations of each. The first population will be the normal or unimpaired group. This group will be stratified in to several categories relating to age, health or fitness criteria, visual acuity, etc. 

The second population will be the impaired group, and like the unimpaired group will be similarly stratified. This group of people--all volunteers--will be further segregated as to their drinking history. It should be noted, parenthetically, that alcohol is the only impairment precursor that can be legally used in these tests. Unfortunately, It is not a faithful representation of the impaired population. If at all possible, all impairment precursors should be represented in these tests, after all, the success of this research is not purely academic--it can save countless lives or not save lives. 

Under a controlled (limited access) environment these various populations and subgroups of each, will be tested using a prototype "Roadside Monitoring System for Impaired Drivers" (RMSFID). Preliminary data taken will be evaluated relative to its impact to system architecture. After any modifications to the system or the experiment: the process will continue until the design matures, with the data taken, being of a more permanent nature (more meaningful). 

Dynamic Template 
There could be a "dynamic" template, one that is a tally of the last n drivers' measured 
performance. It's a little like being graded on the curve. One advantage it may have, is to help cancel out the variables related to the season, weather, time of day, etc. 

Adaptive Template
A variation on the dynamic template is the adaptive template used in vehicle borne systems: the data derived from the continuous measurement of the driver's performance, would be used in a fashion that would continually update--and compare against--an historical template of the present driver's own performance. Using the drivers own evolving record of performance and continually testing for quantifiable deviations, holds great hope of being an effective technique. However, in the unlikely event the driver operated the vehicle in an impaired condition, a significant part of their driving life, it fails and is checked by the simultaneous testing against the standard, or global, performance template. 
 

Impairment Precursors, a List: 

Impairment precursors are those mechanisms, ingested or environmental, that work to alter or weaken one's ability to operate a motor vehicle, to a circumscribed level of performance. The self ingested precursors fall in several broad categories: Intoxicants, Psychotropics, hallucinogens, some over-the-counter (OTC) medications--some with codeine or similar affecting agents. 
 

	Alcohol		Narcotics
	Beer, 3.2%, > 6%		Marijuana
	Wine, 9% > 27%		Cocaine
	Distilled Spirits		Amphetamine
			Heroin
			PCP

 

Motion Vector Components 

Motion vector component data: each have different characteristics and common characteristics, and it's the separation and identification of these three components that is key to the system's usefulness. 

In this system we are concerned with three basic classes of motion vector components: longitudinal, lateral, and oblique. The longitudinal includes the approach vector and depart vector (vehicle coming and going); the lateral includes the front-side vector, right and left and the rear-side vector, right and left (resulting from coarse and fine steering). The oblique is what's left, those remaining motion components that are of sufficient magnitude, but will not fall into either category. 

There are are certain characteristics or physical laws that help in identifying the source of a particular motion component. For example, the rate at which the vehicle can be made to move laterally is significantly faster than achieving the same quick or rapid movement longitudinally--accelerating or decelerating. This can be understood if one thinks of the inertia to be overcome--thus reducing acceleration perturbations to a relatively low frequency function. Considerations of fine steering (that steering that is a result of the driver's attention to small tracking errors and compensating for them) and the differences on the front lateral and rear lateral movements. The front end is very responsive to fine steering, as the rear wheels tend to integrate those movements and tend to track the vector sum of the steering. 

Ultrasonic Doppler-Shift Transducer Design 

The ultrasonic transducer will operate at a frequency range of from 24kHz to 66kHz. Its beam shape or pattern should be adaptable (within a limited range) for the application. 

A single emitter could be used with multiple detectors on one pylon. If the source were at the end of a PCB plastic tube and fed a combination, 50% power divider "T" with rotating joint assembly: each output would feed a beam-forming feed horn. These horns are able to swivel, a limited amount, for better aiming (for a given lane layout). Complementing this, are two highly directional receivers (one on either side). See Drawing 

The proper control of output power and receiver sensitivity is a trade-off between achieving high quality measurements and limiting interference from neighboring transducers. Also, maintaining good directivity is important (narrow, at least in the x direction) by virtue of the improved motion vector component discrimination, i.e., separating the steering induced vehicle motion from the acceleration induced motion. 

Outline of Proposed Research 
 

_____________Doppler Numbers 

Speed of Sound @ sea level: 742 MPH =13,059 inches/sec. 

Doppler, incident sound wave, wavelength = 0.326480 inch, @40,000 Hz. 

Doppler beat = 53.9084 Hz/MPH