Degree Name

Master of Engineering (Research)


School of Electrical, Computer and Telecommunications


Visual impairment is a disability that affects a large number of people. TheWorld Health Organization estimates that there are more than 284 million people suffering from visual impairment. Moving in an unconstrained environment is a vital activity but potentially dangerous for a visually impaired person. Detecting obstacles in a given scene using digital devices can help the blind navigate. Obstacle detection also has applications in road safety and autonomous vehicles.

This project investigates a vision system for sensing pedestrians using a 3-D time-of-flight camera. The range image captured by the time-of-flight camera is used for object segmentation and classification. Distance data calculated from the object region is used to estimate the velocity and collision time of moving pedestrians.

Different image segmentation algorithms are investigated in this project. The histogram thresholding with mean-shift grouping is proposed to segment range images. Object features are extracted by the Fourier transform, Gabor filter (GIST), scale invariant feature transform (SIFT) and histogram of oriented gradient (HOG). These features are used to describe pedestrians in range images. The texture features of corresponding intensity images are combined with features from range images. The support vector machine is used to classify the pedestrian and non-pedestrian objects. The proposed segmentation algorithm compares favorably with several existing techniques, such as the graph cut and Otsu’s algorithm, tested on theMESA database, which is acquired by the time-offlight SwissRanger 4000 camera. The proposed pedestrian classification system is evaluated on the same database.