Feature points extraction is crucial for image segmentation and moving object modeling, and uniform sampling always causes lost of details. Illumination variation and noise also make the extracted feature points unreliable because of the sensitivity of classical detector to noise. A method based on 2D Gabor wavelet is presented to extract features, and a dynamic feature clustering based on feature GWT energy is used to compute the local feature points. The method is robust to noise and the features points extracted are very close to true local features. Active mesh based on image features can be generated by feature points extration.