<
back to Services & Technology list 
Incident
Management > Incident
Detection >
What
Is It?
-
Automatic
Incident Detection (AID) systems use algorithms to analyze traffic
data and quickly detect incidents so as to reduce their adverse
effects.
-
Since
the early 1970s a variety of freeway incident algorithms have
been developed based on traffic flow theory, pattern recognition,
statistical techniques and recently using artificial intelligence
and fuzzy logic.
Key Results
Performance
Evaluation
The most commonly
used performance measures are the ability of an algorithm to detect
an incident (detection rate) as opposed to its false alarm rate.
The detection rate is the number of incidents detected as a percentage
of the number of incidents occurred. The false alarm rate is the
number of false alarm signals as a percentage of tests performed
by the algorithm. A summary of the comparative performance of different
algorithms is given below.
Reported
Algorithm Performance Summary Table
| Algorithm |
Detection Rate
[%] |
False Alarm Rate
[%] |
Average Detection Time
[minutes] |
| California |
Basic |
82 |
1.73 |
0.85 |
| California #7 |
67 |
0.134 |
2.91 |
| California #8 |
68 |
0.177 |
3.04 |
| APID |
86 |
0.05 |
2.5 |
| Standard Normal Deviate |
92 |
1.3 |
1.1 |
| Bayesian |
100 |
0 |
3.9 |
| Time Series ARIMA |
100 |
1.5 |
0.4 |
| Exponential Smoothing |
92 |
1.87 |
0.7 |
| Low-Pass Filter |
80 |
0.3 |
4.0 |
| Modified McMaster |
68 |
0.0018 |
2.2 |
| Neural Networks |
MLF |
89 |
0.01 |
0.96 |
| PNN |
89 |
0.012 |
0.9 |
| Fuzzy Set |
Good |
Good |
Up to 3 minutes quicker
than conventional algorithms |
| Wave Analysis |
Good |
Good |
Good |
| Dutch |
Good |
Poor |
Good |
| Monica |
Poor |
Good |
Good |
| Low-Volume Algorithm |
49-78 |
Volume < 400vph: 1 per 7 hrs
Volume 900-1000 vph: 1 per 2 hrs |
N/A |
| Logit- Based |
x |
96.3 |
5.3 |
Good |
Factors
Affecting Algorithm Performance
Several factors affect
the performance of all types of incident detection algorithms. The
key factors are shown in the table below.
1. Operating Conditions of the Highway-
At
capacity
Well
below capacity
Heavy,
Medium or Light Traffic |
| 2. Duration of the Incident. |
3. Geometric Factors
Grade
Lane
drops
Ramps |
4. Environmental
Snow,
Ice or Fog
Road
surface - Dry or Wet |
| 5. Severity of the incident. |
| 6. Detector spacing |
| 7. Location of the incident relative to the
detector station. |
| 8. Heterogeneity of the vehicle fleet. |
New
Developments
-
Pattern recognition
techniques recognize traffic patterns according to their common
characteristics. Recently introduced algorithms, Fuzzy ART(Adaptive
Resonance Theory) and Fuzzy ARTMAP work on this principle. A
recent study evaluating the algorithms using online techniques
(Haitham Al-Deek, 1999), showed that Fuzzy ART produces significantly
higher detection rate at the same false alarm rate, compared
to California Algorithms #7 and #8.
-
The recently
developed Logit - based algorithm attempts to recognize incident
patterns by using the incident index. The incident index represents
the probability of occurring incidents and is estimated by multinomial
logit model. The model was evaluated on a typical signalized
arterial in Seoul and it is reported that most of the incidents
could be detected accurately.
Author:
Indu Sreedevi
|
