CSE 190A Projects in Vision & Learning Winter 2006 - Project proposal

A proposal for CSE 190A Projects in Vision & Learning, Winter 2006 is presented. The proposal is intended for the suggested project in recognizing cars, with specific attention to the tasks of detection of car manufacturer logos/insignias and classify car colors into familiar color categories. The paper contains the student background information and the general approach conceived in order to cope with the problem, besides open questions to be answered during the project and a list of closely related papers taken as bases to build on. 1 Statement of qual i f ications As a fifth year student at the faculty of Telecommunications Engineering at the University of Trento, Italy (www.ing.unitn.it), Michele Merler took classes connected to Machine vision and learning, including : • Digital Signal Processing • Digital Image Processing (class plus lab) • Introduction to Pattern Recognition Moreover, in order to graduate after three years according to the Italian University system, he developed the thesis work “Video Processing and Analysis Techniques for Smart Environment Application”. Such thesis was based on a three months work on a project at the international research institute Create-Net (www.create-net.org). The project addressed the tasks of people detection and tracking in a smart room, abnormal behavior detection and system-user interaction. 2 Milestones and questions to be answered during the project 2 .1 Cars co lor se t The first step will the creation of a set of possible car colors based on statistics on a database of 150 images observed and analyzed by 5 people of different nationalities. A statistical description will be presented by histograms of each color distribution in different chromatic spaces and chrominance planes, specifically: • Normalized rg [3] • St [12] • CbCr [7] • Lab [7] Distances between colors using different metrics will be evaluated and one metric will be chosen to be used in the car color recognition task. Deadline : 01/15/06 2 .2 LogosIns ign ia s descr ip t ion After researching the current most common car logos, their characteristics will be extracted and analyzed basing on shape descriptors. The different descriptors found in literature will be then evaluated, starting from moment based functions [6, 12], reflection metric[6, 12] and eventually shape contexts[1]. As final step for this task, a database of insignias to be used in further stages of the project will be provided. Deadline : 02/01/06 2 .3 System imp lementa t ion The practical realization of the recognition system will be done according to the following block diagram: Figure 1 : Block Diagram of the System ORIGINAL IMAGE