Research by Abel Ricardo Marcao Ribeiro and Prof. Paulo Lobato Correia.
Research Paper
The aim of this article is to represent a video-based traffic management system that has fast processing speed which allows real time count of the vehicles from acquired Traffic Surveillance Videos (TSV).
The use of video cameras in real time, for monitoring purposes is becoming more popular, therefore this research paper proposed an architecture to counting vehicles using Image processing methods. These type of systems has to be robust, because of the different operational conditions on the same scene like weather or illumination changes.
This video based classification method is started by foreground estimation of the video. By removing the background, it is easy to highlight the objects. This vehicle detecting process has many struggles like shadows of objectives, inclement weather and vehicle occlusion.
Shadows
Detection of shadows affects the system. If the shadow and the real vehicle detected together, it seems bigger than actual vehicle. It can cause a problem to vehicle count. Dealing with shadows divided into two parts. They are property-based and model-based. Property based method use extracted features from the objects. Model based method use the foreground objects or the light sources to detect the presence of the moving shadows on the objects.
Inclement weather
The rainy weather may cause puddles on the roads which cause reflection of objects. The foggy weather can harm to the visibility of the traffic from the camera. This problems may cause the detection of false vehicles. Those type of changes usually are made by lowering thresholds or reducing the size of smoothing filters used to reduce the noise, such as the Gaussian filter.
The vehicles going through the places there are no street light at night cause detection problem because of the illumination changes. Those vehicles can be detected at night using their headlights and taillights.
Vehicle occlusion
When the view of the object is partially or completely obstructed by another objective, the vehicle occlusion occurs. The reasons of this problem can be high traffic density or poor angle of the camera placement. To overcome this problem the Virtual Detection Lines (VDL) and a modified background estimation method is used. In this method, Temporal-Spatial Images (TSI) are generated using the luminace values of pixels of the moving objects which cross the VDLs. Afterwards, the moving objects are detected when they cross these VDLs by applying the background subtraction to the TSIs.
System implementation is a collection of different steps. Initialization is the first step. It consists of using the application’s GUI to draw the VDLs on the TSV’s initial frame, from which the ST images will be automatically created. The user can draw an unlimited number of VDLs, in order to analyse each road lane individually or all the lanes together. To obtain the best results, the VDL should be drawn perpendicularly to the road lanes.
Temporal-spatial Images of the VDL defined as individually generated images. The length of the TSI depends on the number of frames presented in the video. The width of TSI depends on the length of corresponding VDL.
Background estimation and subtraction
In background subtraction method, the researchers used a histogram of the TSI, to give score to each pixel’s intensities. The score is based on the frequency of that intensity on the TSI. More frequent intensities have higher scores. After that, sum of the scores in each column is calculated. Then the column with the highest score is selected. This column is the column of the TSI which has the pixels that are more frequent in the whole TSI. Therefore they assume that it has the highest probability of being a good representation of the background of the VDL. (This process is explained with details in the research paper using equations)
Edge detection
The edge detection is the step that can really involve with finding moving objects in the acquired video. It works on the modified TSI by using background subtraction. In this research, the researchers used Canny Edge Detector to extract edges of the foreground. It can act like a solution to overcome problems like image blurring, arising of fog and unclear edges in an image.
To remove false vehicle detections, they used three different processes.
The first process
The objective of the first proposed process is to eliminate the objects which have small and unusual dimensions, namely a small width and an extremely long height or a small height and an extremely long width. These objects are noise that reduces the vehicle counting accuracy. To do so, the width and height of the objects are analysed and noise is removed.
The second process
Used to verify the objects’ area. Its purpose is to reject small objects, usually noise, detected as vehicles.
The third process
The False Vehicle Elimination (FVE) developed in this dissertation searches for the duplication of objects for the same vehicle, due to multiple detections of the same vehicle. The multiple detections can usually be removed by eliminating the detected objects which are at least partially overlapped. The strategy used to tackle this is based on the comparison of the overlapping area, with the total area of each of the individual rectangles. When this overlapping area is larger than a given threshold, the rectangle is rejected. However, both rectangles cannot be simultaneously rejected; hence when a rectangle is rejected the one with a bigger area is always kept.
The vehicle counting step
The vehicle counting step of this proposed architecture can count the vehicles that cross the selected lane for the considered time duration of the video sequence. The final number of vehicles is simply the sum of the vehicles that where associated between the TSIs and the occlusions which were found.
Finally this research article presented the result of benchmark of the proposed system with different vehicle counting algorithms under the categories of day mode and the night mode. Considering this result comparison, they obtained clear accuracy and improvement of the proposed system. The proposed method achieves its goals to be efficient and flexible to counting vehicles.
