The local image descriptor cores compute the energy, phase, and orientation for every input image. The algorithm starts from a multi-orientation analysis given by a set of Gabor filters.

Local features such as energy or phase can be used for on-road line detection and orientation information (computed in a multiscale fashion). They are typically used for many object recognition or complex local descriptor developments. In addition, phase-based approach can be interpreted as being a robust, complete low-level image representation basis for the development of a complete image analysis framework. An example is shown in Fig 1.

Cores are described in Handel-C and achieve a throughput of 36 fps for an image resolution of 800 by 600 pixels. Hardware details are described in [1].

Fig 1. Whole system results (optical flow, disparity, and local descriptors).

Contact