Real-time detection system for parked vehicles
The present invention discloses a real-time detection system based on hybrid background modeling for detecting parked vehicles along the side of a road. The hybrid background model consists of three components: 1) a scene background model, 2) a computed restricted area map, and 3) a dynamic threshold curve for vehicles. By exploiting the motion information of normal activity in the scene, we propose a hybrid background model that determines the location of the road, estimates the roadside and generates the adaptive threshold of the vehicle size. The system triggers a notification when a large vehicle-like foreground object has been stationary for more than a pre-set number of video frames (or time). The present invention is tested on the AVSS 2007 PV dataset. The results are satisfactory compared to other state-of-the-art methods.
1. An offline hybrid background modeling system for online detection, comprising:
a processor;
a memory device communicatively coupled with the processor; and
a non-transitory storage medium including an application configured for execution by the processor that when executed, enables the processor to:
load initialization frames from the memory device or the non-transitory storage medium acquired by a camera and stack each frame into a 1-D sample vector, wherein the initialization frames contain traffic activities;
form an observation matrix by using sample vectors and set a balancing coefficient;
solve a low-rank representation and output a low-rank matrix to the memory device;
calculate a row-wise median values based on an estimated low-rank matrix and reshape it into the matrix; and
obtain a scene background image based on the matrix.
2. The system according to claim 1 , wherein the general form of the low-rank representation is
min
L
,
E
(
L
*
+
λ
E
0
)
s
.
t
.
X
=
L
+
E
,
wherein L is the-rank matrix, E is a sparse matrix, and λ is the balancing coefficient.
3. The system according to claim 1 , wherein the balancing coefficient λ is set to 0.1.
4. The system according to claim 1 , wherein the processor solved the low-rank representation by using augmented Lagrange multiplier method.
5. A real-time detection system for parked vehicles, comprising:
a processor;
a memory device communicatively coupled with the processor; and
a non-transitory storage medium including an application configured for execution by the processor that when executed, enables the processor to:
set a stationary matrix S 0 =0 and receive a number of frames entered by a user and temporarily stored in the memory device;
perform a background subtraction on frame i acquired by a camera and calculate foreground object labels by using a precalculated restricted area map, wherein i is a natural number;
update a stationary matrix S i for obtaining S i by thresholding a stationary criteria;
extract connected components from S i as candidates, and calculate centroids and average widths of the candidates;
calculate adaptive thresholds using y coordinates of the candidates' centroids and an adaptive threshold function; and
send a notification to output user devices if the average width of the candidate is larger than the adaptive threshold.
6. The system according to claim 5 , wherein the stationary matrix is two dimensional.
7. The system according to claim 5 , wherein the processor precalculated the restricted area by following steps:
loading initialization frames from the memory device or the non-transitory storage device medium acquired by the camera and computing motion information of every two consecutive frames, wherein the initialization frames contain traffic activities;
performing mean shift clustering on motion information matrices, extracting moving objects and outputting foreground moving object masks;
accumulating the motion matrices and outputting a motion map;
transforming the motion map by using difference of sigmoid functions; and
obtaining the restricted area based on the transformed motion map.
8. The system according to claim 7 , wherein the difference of sigmoid functions is
dsigmf
(
x
,
[
a
1
c
1
a
2
c
2
]
)
=
1
1
+
e
-
a
1
(
x
-
c
1
)
-
1
1
+
e
-
a
2
(
x
-
c
2
)
,
wherein x is an element in the motion map; a1 and a2 control the left slope and right slope of the bell shape, respectively, and c1 and c2 control where the stiffs should begin on the left- and right-hand side, respectively.
9. The system according to claim 7 , wherein the adaptive threshold function is calculated by polynomial regression using the motion map obtained in claim 7 .