Object detection using boosted local binaries

• Propose a new binary descriptor with variable patterns.
• Design a structure-aware framework to balance the discriminative ability and generalization power of the proposed descriptor.
• 5% accuracy improvement compared to traditional binary descriptors.
• Effective for large scale dataset and general object detection task.

This paper presents a novel binary descriptor Boosted Local Binary (BLB) for object detection. The proposed descriptor encodes variable local neighbour regions in different scales and locations. Each region pair of the proposed descriptor is selected by the RealAdaBoost algorithm with a penalty term on the structural diversity. As a result, confident features that are good at describing specific characteristics will be chosen. Moreover, the encoding scheme is applied in the gradient domain in addition to the intensity domain, which is complementary to standard binary descriptors. The proposed method was tested using three benchmark object detection datasets, the CalTech pedestrian dataset, the FDDB face dataset, and the PASCAL VOC 2007 dataset. Experimental results demonstrate that the detection accuracy of the proposed BLB clearly outperforms traditional binary descriptors. It also achieves comparable performance with some state-of-the-art algorithms.