Based On The Adaboost Algorithm, Real-time Pedestrian Detection System

Pedestrians are the important objects for intelligent video surveillance system. In this thesis the pedestrian detection based on supervised machine learning is studied in detail. The main contributions are as follows:1. The convergence and generalization capability of Adaboost algorithm as well as the effect of weight-updating approach on the classifier's performance are analysed comprehensively. On this basis, a new weight-updating approach of Adaboost is presented, which modifies the weight-adjusting process of traditional Adaboost algorithm and gives attention to both the misclassification rates of a certain object class and overall object classes, thus it can decrease the computation burden and meet the special needs of cascade classifier to the component strong classifiers.2. Four types of unsymmetrical rectangle filters are given, which can improve the discriminability in last scale and decrease the numbers of feature in the strong classifier dramatically. It will have an complementary advantage of optimizing the classifier's configuration to use both the symmetrical and unsymmetrical rectangle filters and will enhance the classifier's performance.3. A video sequences database is created, which consists of about 80 video sequences shot from several real scenes, including the condition of raining, snowing, shadows, highly textured, the traffic scenes with both static and dynamic background as well as people motion viewed from multiple cameras. This database can provide effective experiment data to test the algorithms of object detection, tracking, trajectory analysis, etc, and lay a good foundation of further researches.4. A real-time pedestrian detection system is developed. This system uses Gaussian distribution to model the background and then segment moving objects. In this way, the candidate regions can be reduced dramatically and realtime processing is realized. The experiment results show that the system is of satisfactory performance in conditions of those complex scences including shadows, raining, snowing, highly textured, multiple moving targets.