Automatic False Positive Canceling for Indoor Human Detection
Refereed conference paper presented and published in conference proceedings


Times Cited
Altmetrics Information
.

Other information
AbstractHumans are the most common mobile obstacles for an office robot. A robust human detector usually plays important roles in robots' perception phase. Especially, Tasks like dynamic path planning and obstacle avoidance have to employ such a detector to help make real-time decisions. Thus, an efficient human detector is meaningful for office robots. However, while deep learning based detectors are not always available in embedded applications, shallow model based ones suffer much from high false positive rates. In this paper, we present a novel approach that can automatically canceling most false positives from shallow detectors. This method regards the detection result as candidates. Then, an algorithm called rival penalized competitive learning is applied to evaluate the inner geometry relationship of each candidate. The one with low evaluation scores, which indicates possible false positives, will be discarded. Experiments show that this approach has the ability to decrease false positive rate of detectors running in office environment.
All Author(s) ListZhu D, Meng MQH
Name of Conference2016 IEEE International Conference on Information and Automation (ICIA)
Start Date of Conference01/08/2016
End Date of Conference03/08/2016
Place of ConferenceNingbo, China
Country/Region of ConferenceChina
Proceedings Title2016 IEEE International Conference on Information and Automation (ICIA)
Title of Publication2016 IEEE INTERNATIONAL CONFERENCE ON INFORMATION AND AUTOMATION (ICIA)
Year2016
Month8
Pages381 - 384
ISBN978-1-5090-4103-9
eISBN978-1-5090-4102-2
LanguagesEnglish-United Kingdom
Web of Science Subject CategoriesAutomation & Control Systems;Computer Science, Information Systems;Automation & Control Systems;Computer Science

Last updated on 2020-23-11 at 01:37