Wei Yeang Kow

Image segmentation via normalised cuts and clustering algorithm

Image segmentation has been widely applied in image analysis for various areas such as biomedical imaging, intelligent transportation systems and satellite imaging. The main goal of image segmentation is to simplify an image into segments that have a strong correlation with objects in the real world. Homogeneous regions of an image are regions containing common characteristics and are grouped as single segment. One of the graph partitioning methods in image segmentation, normalised cuts, has been recognised producing reliable segmentation result. To date, normalised cuts in image segmentation of various sized images is still lacking of analysis of its performance. In this paper, segmentation on synthetic images and natural images are covered to study the performance and effect of different image complexity towards segmentation process. This study gives some research findings for effective image segmentation using graph partitioning method with computation cost reduced. Because of its cost expensive and it becomes unfavourable in performing image segmentation on high resolution image especially in online image retrieval systems. Thus, a graph-based image segmentation method done in multistage approach is introduced here.

Graph-Based Image Segmentation Using K-Means Clustering and Normalised Cuts

Image segmentation with low computational burden has been highly regarded as important goal for researchers. Various image segmentation methods are widely discussed and more noble segmentation methods are expected to be developed when there is rapid demand from the emerging machine vision field. One of the popular image segmentation methods is by using normalised cuts algorithm. It is unfavourable for a high resolution image to have its resolution reduced as high detail information is not fully made used when critical objects with weak edges is coarsened undesirably after its resolution reduced. Thus, a graph-based image segmentation method done in multistage manner is proposed here. In this paper, an experimental study based on the method is conducted. This study shows an alternative approach on the segmentation method using k-means clustering and normalised cuts in multistage manner.

Adaptive Tracking of Overlapping Vehicles Via Markov Chain Monte Carlo with CUSUM Path Plot Algorithm

Vehicle detection and tracking is essential in traffic surveillance and traffic flow optimization. However, occlusion or overlapped vehicle tracking is difficult and remain a challenging research topic in image processing. In this paper, a conventional Markov Chain Monte Carlo (MCMC) is enhanced via Cumulative Sum (CUSUM) path plot in order to track vehicles in overlapping situation. By calculating the hairiness of CUSUM path plot, MCMC can be diagnosed as converged based on its sampling outputs. Varying sample size of MCMC provides enhancement to the tracking performance and capability of overcoming the limitation of conventional fix sample size algorithm. In addition, implementation of m-th order prior probability distribution and fusion of color and edge distance likelihood have further improved the tracking accuracy. MCMC with fixed sample size and CUSUM path plot are implemented and their corresponding performances are analyzed. Experimental results show that MCMC with CUSUM path plot has better performance where it is able to track the overlapped vehicle accurately with lesser processing time.

Enhancement of Particle Filter Resampling in Vehicle Tracking Via Genetic Algorithm

Vehicle tracking is an essential approach that can help to improve the traffic surveillance or assist the road traffic control. Recently, the development of video surveillance infrastructure has incited the researchers to focus on the vehicle tracking by using video sensors. However, the amount of the on-road vehicle has been increased dramatically and hence the congestion of the traffic has made the occlusion scene become a challenge task for video sensor based tracking. Conventional particle filter will encounter tracking error during and after occlusion. Besides that, it also required more iteration to continuously track the vehicle after occlusion. Thus, particle filter with genetic operator resampling has been proposed as the tracking algorithm to faster converge and keep track on the target vehicle under various occlusion incidents. The experimental results show that enhancement of the particle filter with genetic algorithm manage to reduce the particle sample size.

Kernel-based object tracking via particle filter and mean shift algorithm

One of the critical tasks in object tracking is the tracking of fast-moving object in random motion, especially in the field of machine vision applications. An approach towards the hybrid of particle filter (PF) and mean shift (MS) algorithm in visual tracking is proposed. In this proposed system, complete occlusion and random movement of object can be handled due to its ability in predicting the object location with adaptive motion model. In addition, the PF is capable to maintain multiple hypotheses to handle clutters in background and temporary failure. However PF requires a large number of particles to approximate the true posterior of the target dynamics. Therefore, MS algorithm is applied to the sampling process of the PF to move these particles in gradient ascent direction. Consequently a small sample size will be sufficient to represent the system dynamics accurately. The proposed approach is aimed to track the moving object in random directions under varying conditions with acceptable computational time.

Overlapping vehicle tracking via adaptive particle filter with multiple cues

Vehicle tracking is a vital approach to assist the on-road traffic surveillance system. Since the on-road vehicles is increasing, occlusion and overlapping of vehicles is often happen in the traffic surveillance scene. Therefore, segmentation and tracking of the occlusion or overlapped vehicle can be a challenging task in surveillance system via image processing. In this paper, a multiple cues overlapping vehicle tracking algorithm is proposed to continuously track the occluded vehicle effectively. The earlier vehicle tracking systems are normally based on colour feature which will leads to inaccurate results when the background colour is complex or too similar with the target vehicle. On the other hand, shape feature will increase the accuracy but consume more computation time in the resampling process during overlapping. The experimental results show that enhancement of the particle filter resampling process with multiple cues is capable to track the overlapped vehicle with higher accuracy and without compromising the processing time.

CUSUM-Variance Ratio based Markov Chain Monte Carlo algorithm in overlapped vehicle tracking

Markov Chain Monte Carlo (MCMC) is one of the algorithms that have been widely implemented in tracking vehicle for traffic surveillance purposes. The sampling efficiency of the algorithm is essential to determine the vehicle position accurately. However, the sample size of the algorithm is still remaining an issue as non-optimal sample size will defect the tracking accuracy, especially when the moving vehicle is overlapped. Adaptive sample size of MCMC has been implemented using CUSUM Path Plot and Variance Ratio algorithms to perform vehicle tracking. CUSUM Path Plot determines the samples convergence rate by calculating the hairiness of the sample size whereas Variance Ratio method computes two sets of MCMC to determine the samples steady state. This paper proposes the fusion of CUSUM-Variance ratio algorithm to enhance the tracking efficiency. Experimental results shows that the CUSUM-Variance Ratio method have a better performance in tracking the overlapping vehicle with higher accuracy and more optimal sample size compared to the standalone CUSUM Path Plot and Variance Ratio approaches.

Implementing manifold learning in adaptive MCMC for tracking vehicle under disturbances

In recent years, tracking vehicle with overlapping and maneuvering disturbances has become a challenging task in visual tracking. Markov Chain Monte Carlo (MCMC) is proved to be effective in tracking vehicle under disturbances by probabilistically estimating the vehicle position. However the sampling based tracking algorithm is highly depending on the sampling efficiencies where adequate chain length is necessary to sustain the tracking accuracy. Therefore variance ratio (VR) based MCMC has been implemented in this study to adapt the chain length according to the disturbances encountered. Isomap manifold learning is further implemented to update the vehicle model and accurately track the vehicle with maneuvering disturbances. Multiple vehicle models with different viewing angles are represented by Isomap under low dimensional manifold. The suitable vehicle model will be selected according to the estimated vehicle position. Experimental results have shown that Isomap-VR-MCMC have better tracking performances compared to VR-MCMC with smaller RMSE value.