Motion estimation on homogenous surface for around view monitoring system

• Main Authors: Hanizam, Syahirah, Nik Hashim, Nik Nur Wahidah, Zainal Abidin, Zulkifli, Mohd Zaki, Hasan Firdaus, Abdul Rahman, Hasbullah, Mahamud, Nurul Hidayah
• Format: Conference or Workshop Item
• Language: English; English
• Published: Institute of Electrical and Electronics Engineers Inc. 2020
• Subjects: QA75 Electronic computers. Computer science
• Online Access: http://irep.iium.edu.my/78402/1/78402_Motion%20Estimation%20on%20Homogenous%20Surface%20_complete.pdf; http://irep.iium.edu.my/78402/2/78402_Motion%20Estimation%20on%20Homogenous%20Surface%20_scopus.pdf; http://irep.iium.edu.my/78402/; https://ieeexplore.ieee.org/abstract/document/8952062/keywords#keywords
• Institution: Universiti Islam Antarabangsa Malaysia

Around View Monitoring (AVM) system uses multiple input cameras mounted on different positions of a vehicle to display 360° bird-eye-view around the vehicle that is not readily visible to the driver. The development of this system will contribute to the reduction of parking accidents by monitoring its surroundings, detecting lanes and identifying obstacles. With AVM, we can significantly decrease the number of minor accidents. AVM will not only be used for parking assistance but can also assist navigation in the narrow path area. Conventional AVM systems developed in the market using four or six cameras and requires an additional sensor for detection in order to minimise stitching error or to reduce the time to calibrate the output display image. The procedure is time-consuming and increases the cost of development. We propose to develop two ultra-wide-angle cameras located on the front and rear vehicle integrated with the motion estimation (ME) algorithm to produce a parking bird eye view and forward/backward trajectory lines. From our ablative analysis, optical flow is not suitable to be used for real-time ADAS systems as it fails at least 25.5% of the time. However, block matching algorithm based on normalized cross-correlation (CCORR NORMED) and normalised correlation coefficient (CCOEFF NORMED) were able to detect all templates correctly with 0% of false detection on our dataset.