Performance optimization of rotation-tolerant Viola–Jones-based blackbird detection

Abstract

The research described in this paper investigates the rotational robustness of the Viola–Jones algorithm (VJA) object detection method when used for red-winged blackbird (Agelaius phoeniceus) detection. VJA has been successfully used for face detection, but can be adapted to detect a variety of objects. This work uses the histogram of oriented gradients (HOG) descriptor to train the blackbird classifier. Since VJA object detection is inherently not invariant to in-plane object rotation, additional effort is required during training and detection. The proposed method extends the object detection framework developed by Viola and Jones to efficiently handle rotated blackbirds and provide a balance between detection accuracy and computation cost.