Learning convolutional filters for interest point detection

We present a method for learning efficient feature detectors based on in-situ evaluation as an alternative to hand-engineered feature detection methods. We demonstrate our in-situ learning approach by developing a feature detector optimized for stereo visual odometry. Our feature detector parameterization is that of a convolutional filter. We show that feature detectors competitive with the best hand-designed alternatives can be learned by random sampling in the space of convolutional filters and we provide a way to bias the search toward regions of the search space that produce effective results. Further, we describe our approach for obtaining the ground-truth data needed by our learning system in real, everyday environments.

Authors

• Andrew Richardson 0002
• Edwin Olson

Keywords

• Detectors
• Feature extraction
• Visualization
• Cameras
• Discrete cosine transforms
• Three-dimensional displays
• Optimization
• Feature Point Detection
• Random Sampling
• Detection Efficiency
• Feature Detection
• Visual Odometry
• Gradient Descent
• Fast Fourier Transform
• Error Function
• Scale-invariant
• Iterative Optimization
• Feature Matching
• Target Application
• Motion Estimation
• Cubicle
• Fiducial Markers
• Rotation Invariance
• Random Sample Consensus
• Discrete Cosine Transform
• Coordinate Descent
• Camera Motion
• Learned Filters
• Lower Cut-off Frequency
• Reprojection Error
• Sum Of Absolute Differences
• Haar Wavelet
• Convolution
• Coordinate Descent Method
• Filtration Performance
• Objective Function
• Frequency Domain Representation