Design and Optimization of Real-Time Boosting for Image Interpretation Based on FPGA Architecture

This paper presents a reconfigurable architecture of a classification module based on the Adaboost algorithm. This architecture is used for object detection based on the attributes of color and texture. The Adaboost algorithm module uses the technique of decision trees as weak classifiers. This high-performance architecture processes up to 325 dense images of size 640 × 480 pixels, classifying all the structured objects contained on the image. Classification results are provided on an image with the same size. Both architectures, Adaboost algorithm and decision trees, are discussed and compared with several studies found in the literature. The conclusions and perspectives of the project are provided at the end of this document.