A smart CMOS camera for autonomous navigation systems

In this thesis I present my research into the implementation of an edge point detection algorithm within a Smart CMOS Camera. The research includes the development and implementation of a new edge detection algorithm. The algorithm was designed for implementation in a Near Sensor Image Processing (NSIP) structure. This structure was integrated onto a CMOS substrate alongside a random access image-sensing array. The random access array employed pixels with integral gain. Operational specifications for the Smart CMOS Camera were derived from the spatial resolution, the frame rate and edge acuity, necessary to implement corridor autonomous navigation at a walking pace of lm/s. The architecture used to implement NSIP structure is referred to as the Scanned Layer Architecture (SLA). This reflects the layered processing adopted to overcome the connection restrictions of the CMOS substrate. The new edge detector was labelled as the SLA detector. This detector was developed from a study of the gradient based edge detection algorithms. Its integration into a mixed signal CMOS processor was facilitated by limiting the spatial derivative convolution coefficients to integer values, and by minimising the number of product terms. The SLA edge detector was designed to retain edge sense and edge direction information. Two directional edge sets were exported from each processed image. These were a vertical edge set and a horizontal edge set. Within these sets the edge information was encoded in a 3-stateformat to retain the edge sense information. A new edge point metric was developed for the quantitative assessment of the SLA algorithm results.This allowed the detector to be assessed against the requirements of a vision based navigation algorithm. Simulation results demonstrate the use of the SLA edge data to locate a robot's floor position within a corridor environment.