Spatial Location Constraint Model in Object Detection

A spatial location constraint model is proposed to deal with the problem of part absence commonly occurred in object detection.The object center is selected as a virtual landmark to describe the spatial relation of all the local features instead of some special point in other similar models.Both the appearance and the spatial relationship are employed in this model.First,appearance information is used to generate hypotheses.Then,the probability of each candidate center is evaluated.Besides,a fast algorithm is also introduced for searching the object center.Experimental results demonstrate the good robustness of the proposed model to part absence in object detection.