Object detection on aerial image using cascaded binary classifier

Ricardo Fonseca

Research output: Contribution to conferencePaper


© 2016 IEEE. Image analysis is essential through a wide range of scientific areas and most of them have one task in common, i.e. object detection. Thus automated detection algorithms had generated a lot of interest. This proposal identifies objects with similar features on a frame. The inputs are the image where to look at, and a single appearance of the object we are looking for. The object is searched by a sliding window of various sizes. A positive detection is given by a cascaded classifier that compares input patches from sliding window to the object model. The cascaded classifier has three stages: variance comparison, layers of pixel comparisons and patch correlation. Object model is a collection of templates which are generated from scales and rotations of the first appearance. This algorithm is capable to handle change in scale, in plane rotation, illumination, partial occlusion and background clutter. The proposed framework was tested on high cluttered background aerial image, for identifying palm oil trees. Promising results were achieved, suggesting this is a powerful tool for remote sensing image analysis and has potential applications for a wide range of sciences which require image analysis.
Original languageEnglish
Publication statusPublished - 14 Aug 2017
EventProceedings - Applied Imagery Pattern Recognition Workshop -
Duration: 14 Aug 2017 → …


ConferenceProceedings - Applied Imagery Pattern Recognition Workshop
Period14/08/17 → …

Fingerprint Dive into the research topics of 'Object detection on aerial image using cascaded binary classifier'. Together they form a unique fingerprint.

  • Cite this

    Fonseca, R. (2017). Object detection on aerial image using cascaded binary classifier. Paper presented at Proceedings - Applied Imagery Pattern Recognition Workshop, . https://doi.org/10.1109/AIPR.2016.8010601