Minimum Delay Motion Detection

Dong Lao and Ganesh Sundaramoorthi

King Abdullah University of Science and Technology

Abstract

We present a general framework and method for detection of an object in a video based on apparent motion. The object moves relative to background motion at some unknown time in the video, and the goal is to detect and segment the object as soon it moves in an online manner. Due to unreliability of motion between frames, more than two frames are needed to reliably detect the object. Our method is designed to detect the object(s) with minimum delay, i.e., frames after the object moves, constraining the false alarms. Experiments on a new extensive dataset for moving object detection show that our method achieves less delay for all false alarm constraints than existing state-of-the-art.

Citation

Lao, and Sundaramoorthi. "Minimum Delay Moving Object Detection." Computer Vision and Pattern Recognition (CVPR), 2017 IEEE Conference on. IEEE, 2017.

@inproceedings{lao2017minimum,
  title={Minimum Delay Moving Object Detection},
  author={Lao, Dong and Sundaramoorthi, Ganesh},
  booktitle={2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={4809--4818},
  year={2017},
  organization={IEEE}
}

Sample Results of Detection Method

Evaluation

-CHANGE TIME: Object(s) within the scene begin to move or come into view of the camera.

-CORRECT DETECTION: Detection declared after change time and F-measure larger than F-limit.

-FALSE ALARM: Detection declared before change time or F-measure smaller than F-limit.

-DELAY: Moment of correct detection minus change time

(1) False Alarm Rate - Delay

(2) Segmentation accuracy at detection time

(3) Ideal detection mechanisms

(1) False alarm v.s. delay curve shows our method achieves the least delay under same false alarm constraint.

(2) All moving object detectors are compared in terms of their average F-measure to ground truth at detection time.

(3) By detection mechanisms, we mean the test that decides the detection, e.g., the ratio test for ours and the area test for others. We show that under the case of perfect detection mechanisms for all methods, the segmentation procedure from our method leads to the best overall detection schemes compared to other approaches.

Dataset


[Dataset] [Evaluation Code]