Hough Forests for Object Detection, Tracking, and Action Recognition

Abstract

Abstract—The paper introduces Hough forests, which are random forests adapted to perform a generalized Hough transform in an efficient way. Compared to previous Hough-based systems such as implicit shape models, Hough forests improve the performance of the generalized Hough transform for object detection on a categorical level. At the same time, their flexibility permits extensions of the Hough transform to new domains such as object tracking and action recognition. Hough forests can be regarded as task-adapted codebooks of local appearance that allow fast supervised training and fast matching at test time. They achieve high detection accuracy since the entries of such codebooks are optimized to cast Hough votes with small variance and since their efficiency permits dense sampling of local image patches or video cuboids during detection. The efficacy of Hough forests for a set of computer vision tasks is validated through experiments on a large set of publicly available benchmark data sets and comparisons with the state-of-the-art.

Propagative Hough Voting for Human Activity Recognition

“…To avoid enumerating all the possible sub-volumes and save the computational cost, Hough voting has been used to locate the potential candidates. In [7], Hough voting has been employed to vote for the temporal center of the activity while the spatial locations are pre-determined by human tracking. However, tracking human in unconstrained environment is a challenging problem.…”

“…In our algorithm, both spatial and temporal centers can be determined by Hough voting and the scale can be further refined with back-projection. Besides, the trees in [7] are supervisedly constructed for the classification purpose while our trees are trying to model the underlying data distribution in an unsupervised way. Furthermore, our trees can be built on the test data which allows our propagative Hough voting to well handle the limited training data problem.…”

“…For example, [7] requires sufficient labeled local features to train the random forest. When limited training examples are provided, e.g., given one or few query examples to search similar activity instances, the performance of previous methods is likely to suffer.…”

“…The accumulated Hough voting score can be used for recognition and detection. By applying the random projection trees, our proposed method is as efficient as the existing Hough-voting based activity recognition approach, e.g., the random forest used in [7]. However, our method does not rely on human detection and tracking, and can well handle the intra-class variations of the activity patterns.…”

Exploiting Pedestrian Interaction via Global Optimization and Social Behaviors

“…Tracking is often divided in two steps: detection, finding the objects of interest on every frame, and data association, matching the detections to form complete trajectories in time. Researchers have presented improvements on the object detector [1][2][3] as well as on the optimization techniques [4,5] and even specific algorithms have been developed for tracking in crowded scenes [6,7]. Though each object can be tracked separately, recent works have proven that tracking objects jointly and taking into consideration their interaction can give much better results in complex scenes.…”