Tracking-based semi-supervised learning

By Alex Teichman, Sebastian Thrun. 2011

85 citations recorded by the source corpus.

We consider a semi-supervised approach to the problem of track classification in dense three-dimensional range data. This problem involves the classification of objects that have been segmented and tracked without the use of a class-specific tracker. This paper is an extended version of our previous work. We propose a method based on the expectation–maximization algorithm: iteratively (1) train a classifier, and (2) extract useful training examples from unlabeled data by exploiting tracking information. We evaluate our method on a large multiclass problem in dense range data collected from natural street scenes. When given only three hand-labeled training tracks of each object class, the final accuracy of the semi-supervised algorithm is comparable to that of the fully supervised equivalent which uses two orders of magnitude more. Further, we show experimentally that the accuracy of a classifier considered as a function of human labeling effort can be substantially improved using this method. Finally, we show that a simple algorithmic speedup based on incrementally updating a boosting classifier can reduce learning time by a factor of three.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Electrical and Electronic Engineering

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Electrical and Electronic Engineering

Citations in the corpus

Voting for Voting in Online Point Cloud Object Detection (2015, background): “[19, 18], for example, first segment input laser scans and then track the segments obtained in a sequence of scans.”
Watch and learn: Semi-supervised learning of object detectors from videos (2015, methodology): “Temporally consistent detections: We ﬁrst remove detections that are temporally inconsistent using a smoothness prior on the motion of detections.”
What could move? Finding cars, pedestrians and bicyclists in 3D laser data (2012, background): “[7], [10], who classify complete tracks of segmented objects into one of the classes car, pedestrian, bicyclist or background.”
Guaranteed Safe Online Learning via Reachability: tracking a ground target using a quadrotor (2012, background): “As an illustrative example, consider an autonomous car that uses a machine learning algorithm to learn online what objects around it are obstacles that must be avoided, such as pedestrians in crosswalks (not unlike [7]).”
Performance of histogram descriptors for the classification of 3D laser range data in urban environments (2012, background): “Hence, we cannot exploit the range data in terms of first generating a segmentation and then classifying the segments [1], or even use tracking information to segment dynamic objects of interest [2].”
Generalizing Skills with Semi-Supervised Reinforcement Learning (2016, unknown): “However, while semi-supervised learning has been successful in domains such as object tracking and detection (Teichman & Thrun, 2007), applications to action and control have not been applied to the objective of the task itself.”