This paper discussed about how shape-based object detection and tracking takes place using the MOUGH Transform. The shape-based voting algorithm proposed in this paper is mainly addressed one of the challenging problem, pedestrian tracking.
Shape-based detection and tracking:
MOUGH (Mixture of Union and Gaussians Hough) Transform is a new shape model designed and it designed based on a Gaussian Mixture model (GMM). As this method is shape-based, it is used to detect stationary objects and does not depend on movement. Shape-based algorithms use the object typical shape to perform tracking. Register complex 3d models and constraint optic flow methods are the part of shape based methods. This paper focused on edge-based methods for edge-based shape tracking. Active shape models are used in pedestrian tracking. But ASM has its own limitations which overcome by level set methods.
Generalized Hough Transform suggested an alternative approach to shape-based detection. GHT detects arbitrary shapes by using a single template image with a marked centroid. Since GHT has lot of drawbacks. Leibe’s implicit shape model uses a voting approach and has been applied to pedestrian tracking. MOUGH voting can be made up to sixteen times faster by an outlier distribution in training and using centroid adjustments.
MOUGH is a new shape-based model for shape-based detection and tracking. Voting from edges takes place in MOUGH using mixture of Gaussians. An analysis of noise characteristics given images of rigid objects gives this algorithm. Semi-parametric shape template from sample images has been learnt by this algorithm.
This algorithm uses silhouette shape, so it is very effective in detecting pedestrians. This algorithm shows high performance when compared to that of other Hough variants and also provides more advantages than other shape based tracking algorithms like active shape models. The shape-based model using this algorithm helps to locate and track objects accurately against complex backgrounds. This algorithm is more efficient in tracking pedestrians from a side view.