Eliminating Crop Shadows in Video Sequences by Probable Learning Pixel Classification

Shadows have been one of the most serious problems for vegetation segmetation, espescially under conditions of natural random airflow and human or vehicle disturbance. A video sequence processing method has developed in this paper to identify and eliminate crop shadows. The method comprises pixel models and algorithms explained in a probable learning framework. Expectation maximization (EM) for mixture models is established and an incremental EM method is proposed. This method performs a probable reasoning unsupervised classification of pixels for real-time implementation. The results show that the method is quite robust and can successfully remove shadows under natural lighting conditions.