Real-time video analytics for traffic conflict detection and quantification
A video analytics algorithm, system, and method for use in real time allowing accurate, reliable, and timely warnings that facilitate traffic safety and efficiency. The system and method are readily implemented with minimal computational resources, providing broad applicability. In an embodiment, a video analytics method may include (a) obtaining a sequence of real-time images as input from a traffic monitoring system; (b) identifying a plurality of vehicles within a pre-defined region of interest; (c) tracking vehicles within the predefined region of interest; and (d) detecting a conflict event when the vehicles are located within a pre-determined maximum separation threshold based on each vehicles coordinates in a spatial-temporal domain.
1. A video analytics method comprising the steps of
obtaining a sequence of real-time images as input from a traffic monitoring system;
identifying a plurality of vehicles within a pre-defined region of interest; and
detecting, using the sequence of real-time images, a conflict event when trajectories of at least two of the vehicles are within a pre-determined maximum separation threshold in a spatial-temporal domain.
2. The video analytics method of claim 1 further comprising the step of classifying the conflict event.
3. The video analytics method of claim 1 further comprising determining a collision probability based on a computed minimum separation and a pre-determined safe separation threshold.
4. The video analytics method of claim 1 where the step of detecting a conflict event includes determining a separation measure between trajectories of at least two vehicles and comparing a computed separation measure to the pre-determined maximum separation threshold, the separation measure being determined according to the following equation:
d
=
min
x
,
y
,
t
{
ω
[
(
x
i
-
x
j
)
2
+
(
y
i
-
y
j
)
2
]
+
(
1
-
ω
)
(
t
i
-
t
j
)
2
}
subject to: (x i , y i , t i )∈ trajctory i and (x j , y j , t j )∈ trajctory j
where, ω is a weighting factor between spatial separation and temporal separation, (x i , y i , t i ) and (x j , y j , t j ) are points on the trajectories of the vehicles.
5. The video analytics method of claim 1 further comprising the step of comparing respective velocities of the vehicles.
6. The video analytics method of claim 5 further comprising the step of determining a velocity vector difference Δv using the equation
|Δ v |=√{square root over (| v 1 | 2 +|v 2 | 2 −2| v 1 ||v 2 |cos(θ))}
where v 1 is the velocity vector of a first vehicle, v 2 is the velocity vector of a second vehicle, and θ is the angle between the first velocity vector and the second velocity vector.
7. The video analytics method of claim 6 further comprising the steps of
determining a collision probability based on the temporal-spatial separation of the vehicles; and determining a conflict severity value by multiplying the collision probability by the magnitude of
the velocity vector difference.
8. The video analytics method of claim 6 further comprising the step of implementing a corrective action with respect to the region of interest to reduce the collision probability.
9. A method comprising the steps of
obtaining a spatial and temporal position (x, y, t) of a plurality of moving objects from a video source over a defined tracking prism comprising successive frames;
tracking center points (x, y) of all moving objects and assigning a time stamp (t) relative to the tracking window;
representing the center points (x, y) on a plan view by geospatially referencing the view from the video source onto the plan view, and generating a corresponding sequence of dot-featured image frames on the plan view;
extracting trajectories of moving objects in the tracking prism;
determining a conflict by inspecting trajectories in the tracking prism every time step and determining the separation of the trajectories in the joint spatial-temporal (x, y, t) domain; and
characterizing the conflict based on conflict points in the trajectories.
10. The method of claim 9 where conflict points are points on conflicting trajectories having a minimum separation measured in the (x, y, t) domain, and wherein the approaching velocity of each moving object is computed based on the portion of trajectory prior to its respective conflict point.
11. The method of claim 9 further comprising characterizing the conflict based on a conflict type and a conflict severity measure.
12. The method of claim 11 where the conflict severity measure is determined from a magnitude of difference in approaching velocities of the moving objects defined in claim 10 and a collision probability function conditional on the separation measure in the (x, y, t) domain.
13. The method of claim 9 implemented iteratively by sliding a tracking prism one frame at a time in synchronization with at least one of a live video source and a streaming video source.
14. The method of claim 9 further comprising calibrating a probability-of-collision function based on a plurality of conflict events logged over a specified period of observation, where calibrating the probability-of-collision function includes the following steps:
logging a separation of vehicles (d min ) in conflict for each conflict for a pre-determined observation period;
obtaining a frequency and a distribution of the logged separations over the pre-determined observation period; and
using the frequency and the distribution of d min to calibrate the probability-of-collision function as follows:
P
(
collision
|
d
min
=
s
)
=
P
(
d
min
=
0
)
P
(
d
min
=
s
)
=
Frequency
of
d
min
=
0
(
i
.
e
.
,
Freqency
of
colision
)
Frequency
of
d
min
=
s
where P indicates the probability-of-collision function and s indicates a particular d min .
15. A video analytics system comprising
at least one camera producing successive images of a region of interest;
a processor configured to access the images from the at least one camera; and
a memory operatively coupled to the processor, the memory including instructions that, when executed by the processor, cause the processor to perform the following steps:
obtaining a sequence of real-time images as input from a traffic monitoring system;
identifying a plurality of vehicles within a pre-defined region of interest; and
detecting, using the sequence of real-time images, a conflict event when a minimum separation of the trajectories of conflicting vehicles is within a pre-determined maximum separation threshold in a spatial-temporal domain.
16. The video analytics system of claim 15 where the memory includes additional instructions that, when executed by the processor, cause the processor to perform the step of determining a collision probability based on a computed minimum separation and a pre-determined safe separation threshold.
17. The video analytics system of claim 15 where the step of detecting a conflict event includes determining a separation measure between trajectories of two or more of the vehicles captured in the tracking prism and comparing the separation measure to the predetermined maximum separation threshold, the separation measure being determined according to the following equation
d
=
min
x
,
y
,
t
{
ω
[
(
x
i
-
x
j
)
2
+
(
y
i
-
y
j
)
2
]
+
(
1
-
ω
)
(
t
i
-
t
j
)
2
}
.
18. The video analytics method of claim 15 where the memory includes additional instructions that, when executed by the processor, cause the processor to perform the step of comparing respective approaching velocities of the vehicles.
19. The video analytics method of claim 18 where the memory includes additional instructions that, when executed by the processor, cause the processor to perform the step of determining a velocity vector difference Δv using the equation
|Δ v |=√{square root over (| v 1 | 2 +|v 2 | 2 −2| v 1 ||v 2 |cos(θ))},
where v 1 is the approaching velocity vector of a first vehicle, v 2 is the approaching velocity vector of a second vehicle, and θ is the angle between the first velocity vector and the second velocity vector.
20. The video analytics method of claim 19 where the memory includes additional instructions that, when executed by the processor, cause the processor to perform the steps of
determining a collision probability based on the temporal-spatial separation of the vehicles; and determining a conflict severity value by multiplying the collision probability by the magnitude of
the velocity vector difference.