Cars detection in stitched image using morphological approach

• Main Author: Joselyn, Jok.
• Format: Final Year Project Report
• Language: English; English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2017
• Subjects: QA75 Electronic computers. Computer science; T Technology (General)
• Online Access: http://ir.unimas.my/id/eprint/25637/1/Cars%20detection%20in%20stitched%2024pgs.pdf; http://ir.unimas.my/id/eprint/25637/4/Joselyn%20Jok%20ft.pdf; http://ir.unimas.my/id/eprint/25637/

The techniques of image processing which capable to be implemented in object detection and classify accurately plays an important role in developing better computer vision applications such as visual surveillance for behavior analysis and image analysis. This project presents a method to detect cars in panoramic or stitched images which taken from top view angle. Image stitching is basically a process to align and combine few images to produce a panoramic view of the particular scene. The method use to stitch the image is Scale Invariant Feature Transform (SIFT) which perform more accurate matching compared to Speeded Up Robust Features (SURF). Related algorithms using mathematical approach such as alpha blending will be implemented in the project to enhance the image stitching method for a better quality stitched image. There are many type of researches on the topic of car detection are investigated in term of noise elimination at the initial part of the project to get the region of cars. The main functions of the car detection system in this project are capable to recognize and count the number of cars in the aerial type of stitched image by using morphological method which enhanced by using median filtering and convolution. The method is improved by using algorithm which removes noise by sizes. The methods integrate with bounding box to count the number of the cars. The performance of the proposed cars detection algorithm could detect the cars fairly accurate. In addition, this project could be a solution to detect trapped cars in a flood to provide useful information to the rescue department where the image sequences of the flood scene could be stitched together to get a wider view of the flood without zooming the view of the scene to get the whole view of the flood. At the end of the project, some future recommendation for the system’s development will be presented.