Development of video image acquisition for traffic surveillance using open source software

• Main Author: Zainab Nazar, Khalil Wafi
• Format: Thesis
• Language: English
• Published: Universiti Malaysia Perlis (UniMAP) 2014
• Subjects: Monitoring system; Image vision; Traffic Surveillance System (TSS); Smart camera; Open source software; OpenCV programming
• Online Access: http://dspace.unimap.edu.my:80/dspace/handle/123456789/31924

The Research works to presents an innovative approach in monitoring system by applying new programming methods which can be used for Traffic/Road Surveillance Systems. Nowadays, the development of traffic surveillance on the road is compulsory as highways and roads are getting crowded especially in the cities. These crowded situations would increase the number of accidents on the roads. The Traffic Surveillance System performs image recognition and object tracking processing. A camera is used to monitor specific roads (highways, motorways, junctions…etc) and communicate with the main server. The main server integrates the target image from the camera after processing it in several steps then communicates with another computer by sending the image to it. A prototype implementation of traffic surveillance is based on OpenCV programming under GNU-Linux supported by socket programming. The objectives of traffic surveillance system are tracking moving vehicles, counting them and detecting the abnormal movement in specific places. This is done by analyzing the frames captured and processing them using OpenCV functions. The use of open source resources such as OpenCV functions in GNU- Linux provides an easy method to use the computer vision framework besides it can run vision code in real time. The traffic surveillance works in real-time process for (5-33 fps) of video stream in a rainy day and a sunny day. Traffic surveillance system accepts video images either from the camera captures or from the files. It tracks the moving vehicles using image processing and recognition algorithms such as converting to Gray scale and Mixture of Gaussian method; it marks each moving vehicle with a rectangle box and counts them. The system consists of: capturing part, tracking part, detecting abnormal movement from specific points and streaming the view to the other connecting computer. Besides it could minimize the frame processing in the case of the frame capture with high resolution. The system has successfully been tested in three different processor speeds which are 1.2 GHz CPU, 2.0 GHz CPU and 2.6 GHz CPU. The results are quite accurate and are measured in mille second. The system works in outdoor environment which is complex with wavering tree branches and flow of rain. Besides it updates frame by frame in any module of background views especially from the upper view of the road in order to get a significant images of all the objects. The results could be recorded by saving the requested group of photos in any image format. The CPU processing speed with the frame size process represents one of the key factors of the performance analysis of traffic surveillance. Processing speed comparison of the processing steps results in different resolutions for frame size made. And, a significant result is found by reducing the calculation time of processing when the frames size captured are reduced. The experiment was done by using 9 different views each with 700 frames consist of different views of moving vehicles in sunny or rainy situations on the roads of Perlis state. These experiments successfully detected the movements of all types of vehicles.