Abstract— Vehicle speed detection in urban roadways plays an integral role for the government and the society in upholding the propriety of the road for a smooth and accident-less travelling experience. In this research paper, we present an incisive overview of the vehicle speed detection methods and the license plate detection techniques which are used in building this software. We have devised a system that helps in overcoming the limitations of the existing system which uses a radar gun or some other specialized hardware that needs to be managed manually thus inefficient. It uses basic camera (ex. CCTV) footage and processes it with various image processing techniques in order to forecast the accurate speed of the over speeding vehicle. The proposed system was tested in various weather conditions with legitimate ground speed of the vehicles. The measured speeds have an average error of 0-5 km/h which is in acceptable limit proposed by the regulatory bodies. The system was able to achieve desired and precise output with respect to the data set.
Over speeding is considered as one of the most lethal weapon that can take away our lives in just few seconds. According to the statistics provided by the National Highway Traffic Safety Administration (NHTSA) over 35,092 people died in 32,166 fatal accidents reported in the year 2015.
It gives one pause about the dreadful surge in the rate of accidents due to over speeding and triggers us to take a step forward in dwindling the rate.
With an increase in the rate of over speeding it has become laborious for the officials to look after each and every vehicle that over speeds and catch hold of them. The present technology makes use of a radar gun that points at the vehicle and if the vehicle exceeds the threshold limit then the report is send to the nearest police station so that they can catch hold of them. As it requires manual efforts to handle the radar gun therefore it is evident that it may lead to compromised results.
In order to fully automate the process we have designed a system that captures live photos of the over speeding vehicles from cameras that are installed at the highways or at traffic signals. These cameras possess no special qualities, in fact they can also be replaced by any of the ordinary cameras in use. Through the technique of image processing we process those images and the current speed is displayed on the screen. The vehicle whose speed exceeds the threshold speed is further send for number plate detection technique. Thus it definitely is a better substitute of the existing system as it requires considerably less amount of manpower and it proves to be economical.
II. LITERATURE REVIEW
A. Speed detection
Background Subtraction and Blob Detection are the techniques used for vehicle speed detection.
Background subtraction is a method wherein an image’s foreground is extracted for further processing. It is a widely used approach for detecting moving objects in videos from static cameras.
Blob Detection is a method used to mark and identify blobs. A Blob is a group of connected pixels in an image that share some common property (ex. grayscale value).
B. Licence Plate Detection
Licence plate detection is mainly done by Machine learning. Machine learning is a subset of artificial intelligence in the field of computer science that often uses statistical techniques to give computers the ability to “learn” (i.e., progressively improve performance on a specific task) with data, without being explicitly programmed.
III. EXISTING SYSTEM
The existing system uses the Speed Radar Gun to detect the speed of the vehicles on road. The speed radar gun can be handheld, static or entrenched. When the Radar Gun is pointed towards the vehicle, the gun through the RADAR SIGNAL determines the speed by using the principle of Doppler Effect which involves scrutinizing the changing frequency to calculate the output.
Now a days more modern LIDAR (Light Imaging Detection and Ranging) speed guns replaced the traditional speed guns that use the pulsed laser light instead of the radar signal because of the limitations of the small radar system.
However it is tedious system as it involves an employee to impinge the gun at a static point and thoroughly conduct the process to determine the speed and it does not involves the detection of the vehicle through number pate and hence has more manual labour.
In First, numerous camera inputs needs to be taken from the user at the time of installation for which vehicle speed is tracked, which includes its height from the road used to calculate actual distance from the vehicle, field of view used to calculate pixels per inch, minimum speed above which vehicle image is to be recorded, vehicle area for vehicle specific selection (example truck has greater area than a scooter or a car), destination folder where over speeding vehicle images are to be stored.
Then bounded area from the footage is selected which is to be monitored, any vehicle passing through it is under surveillance.
Edges of this region is selected as bounding lines, which are considered as entry and exit points for any vehicle.
Now through background subtraction method moving objects are detected, and then through blob detection centroid of that object is obtained which is further used for speed calculation.
Finally the speed is calculated by the basic formula
Speed = Distance/Time
Where distance is calculated by
Distance = Pixels per inch * Distance covered by blob in bounded region in inches
And time by
Time = Time at which blob left bounded region – Time at which blob entered bounded region
Lastly, image of the over speeding vehicle with its speed is saved in the destination folder which is then further processed for License Number detection.
B．Licence Plate Detection
We take image as an input here, which we obtain from the output of speed detection algorithm, discussed earlier. Final output will be the exact License Number in text format.
Number plate detection is done using 3 different processes.
1. Licence Plate Exposure: It is the most vital process of all. It detects position of Number plate and separates it from whole image. Here image is taken as input and number plate is the output.
2. Character Disjunction: In this process characters are segmented and mapped out as individual character images.
3. Character Recognition: In it we recognise the characters from the image mapped earlier using machine learning.
1) Licence Plate Exposure:
First we convert image into greyscale where each pixels are from 0 to 255, then we convert this to binary image where pixels are either 0 or 1 i.e. either black or white.
Then we identify connected regions by checking adjacent pixel values, if it is same then region can be stated as adjoint. This will also result into many unwanted regions which we aren’t interested in thus, we eliminate them by following strategies,
1. Region must be rectangular in shape.
2. Width must be greater than height.
3. Width and Height of the number plate should be proportionate to the Width and Height of the image respectively. Example, Width should be 20-30% whereas Height should be 10-15% of the image’s Width and Height.
It may still happen that there are multiple regions which satisfy given conditions, for which we select the region with maximum pixel values, matter of fact that characters are written on it.
2) Character Disjunction:
Here we map each character out of number plate. We again use the same method used in Licence plate detection, where we check pixel values of adjacent pixels to mark different regions.
This results into segmentation of each character which is then resized to 20pixel by 20pixel.
3) Character Recognition:
We take use of Supervised Learning of Machine Learning to detect the exact character on the Plate. First we train machine with sufficient amount of images (example 20) of each characters. We take use of 20pixel by 20pixel dataset as we have already converted each character image into it.
Lastly we predict the characters segmented earlier.
This paper caters the needs of the megalopolitan phenomenon which include the determination of the over speeding vehicles on road. Our paper thus presents a quick snapshot summarising the detailed study of the project. The main prospect of the project is to determine the over speeding vehicle on the road. This is done by mainly tracking the vehicle first and then identifying it to the optimum limit.
The technology used in the experimental setup is python, OpenCV library and machine learning. The vehicular detection is done by using python and OpenCV libraries. Our system uses an efficient license plate detection method, a classifier that is novel enough to capture the gradient distribution that comprise the characters of the number plate. All of this is achieved by machine learning at optimum level. Through machine learning the characters and numbers of the detected vehicle are replicated precisely. The input given to the system is the footage of the camera’s installed on the roads already. We achieved an accuracy of 94%-97% by the experimental set up.
We believe that this project extends a great deal of scope as the purpose meets. This can be applied to the practical scenario expecting prominent outputs. Our future works include to send the resultant output to the RTO department over the online database which would help them taking the required action. The technology used suits the requirement best. In future more ideas like these should be implemented that serve as a boon to arising urbane problems.
This essay has been submitted by a student. This is not an example of the work written by our professional essay writers. You can order our professional work here.