ICTACT Journal on Image and Video Processing

Wei Leong Khong ¹, Wei Yeang Kow ¹, Ismail Saad ¹, Chung Fan Liau ¹, Kenneth Tze Kin Teo ¹

Affiliations
1 Modelling, Simulation and Computing Laboratory, Material and Mineral Research Unit, School of Engineering and Information Technology, Universiti Malaysia, Sabah, MY

Abstract

Optical sensors based vehicle tracking can be widely implemented in traffic surveillance and flow control. The vast development of video surveillance infrastructure in recent years has drawn the current research focus towards vehicle tracking using high-end and low cost optical sensors. However, tracking vehicles via such sensors could be challenging due to the high probability of changing vehicle appearance and illumination, besides the occlusion and overlapping incidents. Particle filter has been proven as an approach which can overcome nonlinear and non-Gaussian situations caused by cluttered background and occlusion incidents. Unfortunately, conventional particle filter approach encounters particle degeneracy especially during and after the occlusion. Particle filter with sampling important resampling (SIR) is an important step to overcome the drawback of particle filter, but SIR faced the problem of sample impoverishment when heavy particles are statistically selected many times. In this work, genetic algorithm has been proposed to be implemented in the particle filter resampling stage, where the estimated position can converge faster to hit the real position of target vehicle under various occlusion incidents. The experimental results show that the improved particle filter with genetic algorithm resampling method manages to increase the tracking accuracy and meanwhile reduce the particle sample size in the resampling stage.

Keywords

Vehicle Tracking, Particle Filter, Genetic Algorithm, Resampling, Occlusion.

Full Text

     

Lorita Angeline ¹, Hou Pin Yoong ¹, Hui Keng Lau ¹, Ismail Saad ¹, Kenneth Tze Kin Teo ¹

Affiliations
1 Modelling, Simulation and Computing Laboratory, Material and Mineral Research Unit, School of Engineering and Information Technology, Universiti Malaysia, Sabah, MY

Abstract

A new algorithm for license plate localisation and identification is proposed on the basis of Signature analysis. Signature analysis has been used to locate license plate candidate and its properties can be further utilised in supporting and affirming the license plate character recognition. This paper presents Signature Analysis and the improved conventional Connected Component Analysis (CCA) to design an automatic license plate localisation and identification. A procedure called Euclidean Distance Transform is added to the conventional CCA in order to tackle the multiple bounding boxes that occurred. The developed algorithm, SAICCA achieved 92% successful rate, with 8% failed localisation rate due to the restrictions such as insufficient light level, clarity and license plate perceptual information. The processing time for a license plate localisation and recognition is a crucial criterion that needs to be concerned. Therefore, this paper has utilised several approaches to decrease the processing time to an optimal value. The results obtained show that the proposed system is capable to be implemented in both ideal and non-ideal environments.

Keywords

Vehicle Localisation, Automatic License Plate Recognition, Signature Analysis, Adaptive Searching, Euclidean Distance Transform.

Full Text