![](bg4.png)
ety of base maps and behaviors and estimate different
measures of effectiveness (such as queue lengths and
travel times).
One such essential simulation software is Sim-
ulation of Urban MObility (SUMO) (Lopez et al.,
2018). SUMO is an open-source, microscopic, agent-
based road traffic simulation package that is de-
signed to handle large road networks. SUMO uses
its file formats for traffic networks, but it can im-
port files encoded in other popular formats like Open-
StreetMap (OpenStreetMap contributors, 2017), VIS-
SIM (Lownes and Machemehl, 2006) etc. SUMO is
implemented in C++ and uses only portable libraries,
thus making it lightweight and fast. SUMO is single-
threaded, i.e., it uses only one CPU core, but several
parallel SUMO processes can be spawned, allowing
for parallel simulations.
3 FIELD IMPLEMENTATION
AND DESCRIPTIVE
STATISTICS
A critical intersection at a major urban center in
the United States of America was analyzed for this
demonstration. The intersection borders a large uni-
versity. Several residential complexes and commer-
cial establishments (especially restaurants) in the area
give rise to significant pedestrian traffic and vehicular
traffic. On the median, a single fisheye video camera
is mounted 13 feet (3.96 meters) above the intersec-
tion at the southbound approach.
Raw data from the intersection (fisheye video
data, high-resolution loop detector, and signal state
data) was captured and stored in a cloud database.
Three weeks of data, spanning October-November
2021, with time ranges from 6 am to 8 pm, was con-
sidered. This time range captures the AM (morning),
PM (evening), and midday traffic peaks but also en-
sures sufficient ambient light for the camera to func-
tion. The Major flow (i.e., North-South direction)
usually sees higher flows than the Minor flow (i.e.,
East-West direction), with the Major flow (North-
South combined) reaching 1400-1800 vph (vehicles
per hour) at peak and Minor flow (East-West com-
bined) reaching 800-1200 vph.
We then analyze the data given the hour-of-day
and day-of-week. We look for both variations and
similarities. These will help us identify periods where
significant pedestrian traffic co-exists with vehicu-
lar traffic and also inform us where segmented cus-
tomized signal timing plans may be needed.
4 APPLICATION TO IMPROVE
INTERSECTION SAFETY
Fisheye cameras provide a crucial piece of informa-
tion: turn-movement counts. This data is challeng-
ing to obtain from loop-detector-based ATSPM data
without exit detectors. Intersections frequently have
lanes that accommodate multiple movements, such as
the right-most lane allowing through and right-turning
traffic. Fisheye lens tracking can count these vehi-
cles, which loop detectors cannot differentiate. Fish-
eye data is necessary to analyze direction-wise flows
(see Figure 4). In addition, fisheye cameras also al-
low us to track individual pedestrians. This allows us
to estimate the number of pedestrians crossing in var-
ious directions. This contrasts with ATSPM data that
tracks the number of pedestrian calls but has no way
of estimating the actual number of pedestrians (Figure
5, Figure 7). With accurate trajectories of vehicles
and pedestrians, it is possible to study, both qualita-
tively and quantitatively, the safety aspects of the in-
tersection. Visualizations (Figure 4, Figure 5, Figure
7) can help determine overlapping times and direc-
tions where vehicles and pedestrians have intersect-
ing trajectories. This can support traffic authorities
in modifying intersection signaling (e.g., restricting
right turns, etc.) for those times and directions.
Standard safety metrics(Mullakkal Babu et al.,
2017) such as Time-To-Collision (TTC) and Post-
Encroachment Time (PET) can also be calculated.
Our intersection safety research (Banerjee et al.,
2022) has involved computing TTC and PET, along
with other factors such as speed, deceleration, prox-
imity between conflicting users, to determine severe
events. These metrics are solely obtainable from
trajectory datasets that are processed through video.
In (Mishra et al., 2022), we presented a study that
showed how the day of the week and time of day can
affect the frequency and severity of events at intersec-
tions. This analysis enables us to pinpoint the busiest
periods and conflict-prone locations at intersections.
The data can then be utilized to evaluate safety coun-
termeasures with minimal disruption to intersection
efficiency.
The fisheye video also lets us detect heavy motor
vehicles, i.e., large vehicles such as delivery trucks
and buses (Figure 6). These large vehicles often
have a long braking distance and significant blind
spots. Given that this intersection borders the Uni-
versity and has several shopping and eating locations
nearby, these large vehicles may pose a safety concern
for pedestrians. Their movement can be restricted or
rerouted based on such data. Using this information
to adjust signal timing plans for transit priority is also
Towards Effective Traffic Signal Safety and Optimization Using Fisheye Video
57