Fast window fusion using fuzzy equivalence relation

Current window fusion of the sliding window based human detection is rather slow. This paper proposes a fast fuzzy equivalence relation based method (FER). It merges candidate windows based on the fuzzy equivalence relation structured from the normal fuzzy similarity relation. Experimental results demonstrate that the method can merge candidate windows faster than the popular non-maximum suppression based method (NMS) and the bounding region method (BR), while maintaining the detection quality.

► Study the fast window fusion of the sliding window based human detection.
► Propose a fuzzy equivalence relation based fusion method, FER.
► Experimental results show that FER is faster than previous methods.