Human Detection Based On Adaboost Algorithm

Human detection is an important area of computer vision. Due to the non-rigid characteristic of human body and the variety of shapes, human body detection is considered a difficult task. The applications of human body detection include military alarm, visual safety and ensurance, intelligent building, intelligent transportation system, etc.. It is an important issue to improve the detection rate and to reduce the rate of false alarm. A number of human detection algorithms have been proposed, e.g. detection based on histogram of oriented gradient (HOG), CAMSHIFT, skin-color based human detection. Adaboost algorithm is known for its simplicity, reliablity and high accuracy, and is used for face detection successfully. This paper uses Adaboost algorithm for human detection. Haar features are used for the classifier. The use of integral image allows the features to be computed quickly. A cascade of classifiers is acquired through extensive training. With the cascade of classifiers, many images are tested and good detection results achieved.In the end, this paper presents the implementation of Adaboost algorithm on the DM642 platform.