A hardware-friendly object detection algorithm has been developed using directional-edge histograms as key local-image-feature descriptors. In the recognition window, a number of small blocks having varying sizes and shapes are prepared for pattern matching. Four directional edges in vertical, horizontal and ±45 degrees are extracted from a local image enclosed in the block and their spatial distribution is summarized in the form of a histogram, then being utilized to represent the local image features. In order to put more emphasis on regions where significant features are more likely to appear, AdaBoost algorithm was employed. Taking the pedestrian detection as a test vehicle, the performance of the algorithm was tested. Three kinds of feature vector generation methods were proposed and their performances were compared using MIT CBCL pedestrian data base. An efficient VLSI hardware architecture for implementing such a system is presented and the key component of the system, the variable-block-size vector generator, is described. The component was implemented on a FPGA (Altera Cyclone® II 2C70 FPGA device), and 60 times enhancement in the processing efficiency has been demonstrated by measurement.