The illumination variation and cast shadow cause serious problems while detecting and tracking people due to misclassification of these points as foregrounds. We propose a robust framework for detecting people by eliminating the lighting effects and shadow hierarchically. The DWT and human visual system are adopted to classify the object and the light effect noise such as shadow and spotlight. A multiresolution framework for tracking people is presented using wavelet transform and kalman filter in unconstrained environments. The kalman filter is adopted to be the estimator in this system. The tracking system is efficient and effective framework by combining the corrections between color and position information as the recognition technique of the feature tracking. Experiments show that the system we proposed achieves optimal performance based on the complicated backgrounds.
展开▼