Motivation • In recent years, the number of traffic accidents have increased exponentially. Monitoring speeding vehicles manually is not feasible round the clock everywhere. • Automation of speed detection is need of the hour. • We detect the speed of the moving vehicles, given a traffic surveillance video. 1. Dataset - https://www.youtube.com/watch?v=tqeDene7r74 2. cvBlob - Fabrice de Gans-Riberi 3. Vehicle Detection, Tracking and Counting - Andrews Sobral We found in our experimentation that the Background Subtraction works well in detecting moving vehicles. Each phase involved worked fairly and there is a scope for improvement. Our approach doesn't consider vehicles changing lanes, overtaking other vehicles and handling multiple lanes which can be done in future. Overall, it can be made as a full proof real world application. Conclusion References Results / Evaluation Speed Detection of Moving Vehicles in a Video The following steps would be followed to detect the speed of a moving vehicle in a video obtained from a still mounted camera on a highway. 1. Extract frames from the given traffic surveillance video. 2. Find the foreground image by using Background Subtraction method. 3. Detect the blob and tracks by comparing consecutive foreground images. 4. Filter out the blobs outside the selected lane or having irregular size. 5. Track the entry and exit frames for each moving vehicle using its centroid. 6. Calculate the number of frames between the entry and exit points of the moving vehicle. 7. Compute the speed of the vehicle using the equation below. The mapping of the frame coordinates to real world coordinates of the video is given ሺሻ. Speed Detection: • For each lane, we maintain a queue which contains the entry point of the vehicle. • As the vehicle leaves the exit point, we look at its entry frame from the queue and determine the number of frames elapsed ሺሻ. • Speed is computed using the equation = ∗ . Methodology Bipra De, Ghanshyam Malu, Suhas Gulur Ramakrishna a - Frames from the original video. b - Foreground Image using Background Subtraction Method. c - Blob detected shown in blue. d - Start line, end line, detected speed shown in cyan, green, yellow. Video - 1 Video - 2 Video - 3 a b c d For testing, we have considered a video 1 for which the distance between two points given. It also has the speed of the moving vehicles which acts as the ground truth for our evaluation. Video - 2 Speed Computed (kmph) Actual Speed (kmph) Error % Car 1 112.000 112.500 0.44% Car 2 107.000 115.244 7.15% Car 3 102.000 105.000 2.86% Car 4 126.000 131.250 4.00% Video - 3 Speed Computed (kmph) Actual Speed (kmph) Error % Car 1 98.000 102.717 4.59% We expected a margin of error as the detection of starting and ending point of the car varies by a margin that affects the accuracy of the calculation of the speed which is evident from the above results. Video - 1 Speed Computed (kmph) Actual Speed (kmph) Error % Car 1 126.000 127.703 1.33% Car 2 86.000 82.894 3.75%
Transcript
Motivation
• In recent years, the number of traffic accidents have
3. Vehicle Detection, Tracking and Counting - Andrews Sobral
We found in our experimentation that the Background
Subtraction works well in detecting moving vehicles.
Each phase involved worked fairly and there is a scope
for improvement. Our approach doesn't consider
vehicles changing lanes, overtaking other vehicles and
handling multiple lanes which can be done in future.
Overall, it can be made as a full proof real world
application.
Conclusion
References
Results / Evaluation
Speed Detection of Moving Vehicles in a Video
The following steps would be followed to detect the speed of a
moving vehicle in a video obtained from a still mounted
camera on a highway.
1. Extract frames from the given traffic surveillance video.
2. Find the foreground image by using Background
Subtraction method.
3. Detect the blob and tracks by comparing consecutive
foreground images.
4. Filter out the blobs outside the selected lane or having
irregular size.
5. Track the entry and exit frames for each moving vehicle
using its centroid.
6. Calculate the number of frames between the entry and exit
points of the moving vehicle.
7. Compute the speed of the vehicle using the equation
below.
The mapping of the frame coordinates to real world
coordinates of the video is given .
Speed Detection:
• For each lane, we maintain a queue which contains the
entry point of the vehicle.
• As the vehicle leaves the exit point, we look at its entry
frame from the queue and determine the number of frames
elapsed � .
• Speed is computed using the equation �� = � ∗ ���� .
Methodology
Bipra De, Ghanshyam Malu, Suhas Gulur Ramakrishna
a - Frames from the original video.
b - Foreground Image using Background Subtraction Method.
c - Blob detected shown in blue.
d - Start line, end line, detected speed shown in cyan, green, yellow.
Video - 1 Video - 2 Video - 3
a
b
c
d
For testing, we have considered a video1 for which the
distance between two points given. It also has the speed of
the moving vehicles which acts as the ground truth for our
evaluation.
Video - 2Speed Computed (kmph) Actual Speed (kmph) Error %
Car 1 112.000 112.500 0.44%
Car 2 107.000 115.244 7.15%
Car 3 102.000 105.000 2.86%
Car 4 126.000 131.250 4.00%
Video - 3Speed Computed (kmph) Actual Speed (kmph) Error %
Car 1 98.000 102.717 4.59%
We expected a margin of error as the detection of
starting and ending point of the car varies by a margin
that affects the accuracy of the calculation of the speed
which is evident from the above results.
Video - 1
Speed Computed (kmph) Actual Speed (kmph) Error %
Car 1 126.000 127.703 1.33%
Car 2 86.000 82.894 3.75%
Motivation• In recent years, the number of traffic accidents have increased
exponentially. Monitoring speeding vehicles manually is not feasible roundthe clock everywhere.
• Automation of speed detection is need of the hour.
• We detect the speed of the moving vehicles, given a traffic surveillancevideo.
• Pornpanomchai1 discusses about detecting the position of a movingvehicle in a frame and finding the reference points and calculating thespeed of each static image frame from the detected positions.
• Abbott and Williams2 discuss a method to extract the background image,find the foreground pixel mask to extract blob.
• Jun, Aggarwal, Gokmen3 discuss Background Subtraction and blobsegmentation using Gaussian background model (MoG).
Background Information
1. Vehicle speed detection system – C. Pornpanomchai2. Multiple Target Tracking with Lazy Background Subtraction and
Connected Components Analysis – Abbott and Williams3. Tracking and Segmentation of Highway Vehicles in Cluttered and
We found in our experimentation that the Background Subtractionworks well in detecting moving vehicles. Each phase involved workedfairly and there is a scope for improvement. Our approach doesn'tconsider vehicles changing lanes, overtaking other vehicles andhandling multiple lanes which can be done in future. Overall, it can bemade as a full proof real world application.
Conclusion
References
Results / Evaluation
Speed Detection of Moving Vehicles in a Video
The following steps would be followed to detect the speed of a moving vehiclein a video obtained from a still mounted camera on a highway.
1. Extract frames from the given traffic surveillance video.
2. Find the foreground image by using Background Subtraction method.
3. Detect the blob and tracks by comparing consecutive foreground images.
4. We use centroid of the blob as reference point.
5. Filter out the blobs outside the selected lane or having irregular size.
6. Track the entry and exit frames for each moving vehicle using its centroid.
7. Calculate the number of frames between the entry and exit points of themoving vehicle.
8. Compute the speed of the vehicle using the equation below.
The mapping of the frame coordinates to real world coordinates of the video isgiven .
Speed Detection:• For each lane, we maintain a queue which contains the entry point of the
vehicle.• As the vehicle leaves the exit point, we look at its entry frame from the
queue and determine the number of frames elapsed � .
• Speed is computed using the equation �� = � ∗ ���� .
Methodology
Bipra De, Ghanshyam Malu, Suhas Gulur Ramakrishna
a - Frames from the original video.
b - Foreground Image using Background Subtraction Method.
c - Blob detected shown in blue.
d - Start and end lines shown in green, cyan. Detected speed shown in yellow.
Video - 1 Video - 2 Video - 3
a
b
c
d
For testing, we have considered a video4 for which the distance between twopoints given. It also has the speed of the moving vehicles which acts as theground truth for our evaluation.
Video - 2
Speed Computed (kmph) Actual Speed (kmph) Error %
Car 1 112.000 112.500 0.44%
Car 2 107.000 115.244 7.15%
Car 3 102.000 105.000 2.86%
Car 4 126.000 131.250 4.00%
Video - 3
Speed Computed (kmph) Actual Speed (kmph) Error %
Car 1 98.000 102.717 4.59%
We expected a margin of error as the detection of starting and endingpoint of the car varies by a margin that affects the accuracy of thecalculation of the speed which is evident from the above results.
