Target Detection and Extraction Based on Motion Attention Model
Long Liu, Jie Ding · 2018
In the paper, a new motion attention temporal spatial fusion model is constructed for motion object detection and extraction in view of the limitations of target detection and extraction method under global motion scene according to motion attention formation mechanism. In the algorithm, motion vector fields undergo superposition, filtering and other pretreatment firstly. Then, a motion attention fusion model is defined according to temporal - spatial change characteristics of motion vector. The model is adopted for detecting the motion object region. Finally, morphology and boundary following method are utilized for accurate extraction of the object region. The test results of many different global motion video scenes show that the algorithm has better accuracy and real - time performance than other similar algorithms.