To reduce the amount of memory of the HOG feature for object detection, this paper proposes the Relational HOG feature (R-HOG) and masking of the binary by using a wild-card "*" with Real AdaBoost. HOG features are effective for human detection, but their focus on local regions makes them high-dimension features. Therefore, to reduce the memory size for the HOG features, this paper proposes R-HOG that creates binary patterns from the HOG features extracted from two local regions. This approach enables the created binary patterns to reflect the relationships between local regions. However, since R-HOG features contain binary values not needed for classification, we have added a process to the Real AdaBoost learning algorithm in which wild-card "*" permits the two binary values of "0" and "1", and so unnecessary binary can be masked. The results of our evaluation experiment demonstrated that our method offered better detection performance than the conventional method (HOG feature) despite managing to reduce the amount of memory.
展开▼