ABSTRACT
In this paper we report on techniques for automatically learning foveal sensing strategies for an active pan-tilt-zoom camera. The approach uses reinforcement learning to discover foveal actions maximizing the performance of visual detectors, that are in turn assumed to be highly correlated with the task at hand. In our case,the main goal is to recognize people, hence a frontal face detection module is employed. The system uses reinforcement learning to learn if when and how to foveate on a subject, basedonits previous experience in terms or successful actions in similar situations. An action is successful if it leads to a correct face detection in the high resolution images obtained when the subject is zoomed in. In contrast with existing methods,the proposed approach obviates the need for camera calibration and camera performance modeling. Also, the method does not rely on active tracking of targets. Experimental results show how the system can be deployed in unconstrained surveillance environments, and is capable of learning foveation strategies without requiring extensive a priori information or environmental models. Results also illustrate how the system effectively learns a strategy that allows the camera to foveate only in situations where successful detection is highly likely.
- A. D. Bagdanov, A. del Bimbo, and W. Nunziati. Improving evidential quality of surveillance imagery through active face tracking. In Proceedings of ICPR 2006 2006.(to appear). Google ScholarDigital Library
- A. D. Bagdanov, A. del Bimbo, and F. Pernici. Acquisition of high-resolution images through on-line saccade sequence planning. In Proceedings of VSSN2005 pages 121--130, 2005. Google ScholarDigital Library
- C. J. Costello, C. P. Diehl, A. Banerjee,and H. Fisher. Scheduling an active camera to observe people. In Proceedings of VSSN2004 October 2004. Google ScholarDigital Library
- M. Erdem and S. Sclaroff. Look there! predicting where to look for motion in an active camera network. In Proceedings of IEEE Conference on Advanced Video and Signal Based Surveillance (AVSS)2005.Google Scholar
- L. P. Kaelbling, M. L. Littman, and A. P. Moore. Reinforcement learning: A survey. Journal of Artificial Intelligence Research 4: 237--285, 1996. Google ScholarDigital Library
- S.-N. Lim, L. S. Davis, and A. Elgammal. Scalable image-based multi-camera visual surveillance system. In Proceedings IEEE Conference on Advanced Video and Signal Based Surveillance pages 205--212, July 2003. Google ScholarDigital Library
- D. W. Murray, K. J. Bradshaw, P. F. McLauchlan, I. D. Reid, and P.Sharkey. Driving saccade to pursuit using image motion. IJCV 16(3): 205--228, 1995. Google ScholarDigital Library
- J. Peng and R. J. Williams. Incremental multi-step q-learning. Machine Learning 22(1-3):283--290, March 1996. Google ScholarDigital Library
- A. Senior, A. Hampapur, and M. Lu. Acquiring multi-scale images by pan-tilt-zoom control and automatic multi-camera calibration.In Proceedings of the IEEE Workshop on Applications on Computer Vision Breckenridge, Colorado, January 2005. Google ScholarDigital Library
- J. N. Tsitsiklis. Asynchronous stochastic approximation and q-learning. Machine Learning 22(1):185--202, May 1994. Google ScholarDigital Library
- C. J. C.H. Watkins and P. Dayan. Q-learning. Machine Learning 8(3-4): 279--292, May 1992. Google ScholarDigital Library
- X. Zhou, R. Collins, T. Kanade, and P. Metes. A master-slave system to acquire biometric imagery of humans at a distance. In ACM SIGMM 2003 Workshop on Video Surveillance Berkeley, CA, 2003. Google ScholarDigital Library
Index Terms
- A reinforcement learning approach to active camera foveation
Recommendations
Reward Shaping in Episodic Reinforcement Learning
AAMAS '17: Proceedings of the 16th Conference on Autonomous Agents and MultiAgent SystemsRecent advancements in reinforcement learning confirm that reinforcement learning techniques can solve large scale problems leading to high quality autonomous decision making. It is a matter of time until we will see large scale applications of ...
Handling camera movement constraints in reinforcement learning based active object recognition
DAGM'06: Proceedings of the 28th conference on Pattern RecognitionIn real world scenes, objects to be classified are usually not visible from every direction, since they are almost always positioned on some kind of opaque plane. When moving a camera selectively around those objects for classifying them in an active ...
Air-to-Ground Active Object Tracking via Reinforcement Learning
Artificial Neural Networks and Machine Learning – ICANN 2023AbstractOver the years, active object tracking has emerged as a prominent topic in object tracking. However, most of these methods are unsuitable for tracking ground objects in high-altitude environments. Therefore, the paper proposes an air-to-ground ...
Comments