Design And Validation Of Adaboost Face Detection Algorithm Based On FPGA

Face recognition has always been one of the hot spots in the field of artificial intelligence.And with the progress of computational methods and computational tools,it has been greatly developed.As the first step of face recognition system,the detection rate and speed of face detection directly affect the performance of the whole system.So the implementation of fast and accurate face detection is of great significance.Adaboost face detection algorithm is optimized in this thesis,which can be applied to the FPGA platform.After in-deep analysis of the face detection algorithm based on Haar feature,the strategies of optimization and improvement is determined to ensure the accuracy and speed.The image scaling method and multi-size detection window strategy are combined to avoid non-integer multiplication,making it easier for FPGA implementation.In order to accelerate the detection process,the detection windows slide in larger strides to minimize detection times.Small-scale fixed windows are used to slide over the image to save the memory space.The methods to calculate and store integral image are developed for further reducing the computing time.Validation will be performed for several times at the face location to reduce the false detection rate.In addition,a new feature threshold comparison method and floating point to fixed point measure are conduced to make the algorithm applicable for FPGA platform.Based on the Opal Kelly development board XEM6310-LX150,the Adaboost face detection algorithm is optimized and tested.The experimental results show that the face detection algorithm takes up 1834 Slices resources and 1495Kb on-chip BRAM resources of Xilinx Sparten-6 series XC6SLX150 FPGA,at the 62MHz system operating frequency.The false detection rate of the face detection system is less than 2%.The detection rate for HD images with only one face is 98%and the detection speed is 1.96 fps and the comprehensive performance of the face detection system is much better than the existing systems;the detection rate for HD images which contain 5 faces is 92.56%and the detection speed is 1.28 fps.The face detection system designed can quickly and accurately provide the face image information for the face recognition module,and further construct a complete face recognition system.