출입자 자동 검출을 위한 물체 볼륨 분석 알고리즘
Object Volume Analysis Algorithm for Automatic Detection of Visitors
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
건물에 출입하는 인원을 자동으로 검출하는 기술은 화재 등 재난 발생 시 요구조자의 피난을 위해 필요하다. 본 논문은 이를 위한 방안으로 출입구에 설치된 열화상 카메라를 이용하여 출입 인원의 볼륨 영역을 실시간 자동 검출하는 방안을 제안한다. 이와 같은 검출 시스템의 첫 단계는 이미지 내 정확한 인원을 예측하는 일이며 본 논문에서는 YoloV5 딥러닝 모듈을 적용하였다. 다음 단계는 본 논문에서 제안하는 열화상 물체의 이동 방향 분석 알고리즘이다. 이 알고리즘은 연속적으로 입력되는 열화상 볼륨 이미지의 순서를 분석하여 들어오는 인원과 나가는 인원을 구분하여 이동을 카운팅하였다. 제안하는 알고리즘은 이미지에서 재실자가 이동하는 주 영역을 4 개의 부분 영역으로 나누고 각 부분 영역마다 물체 볼륨 모양과 분포 변화를 추적하는 방식으로 구성하였다. 제안하는 방식은 출입구에서 발생하는 보행자의 다양한 움직임 변이에도 안정적인 검출 성능을 보였다. 이러한 방식의 재실자 분석은 건물의 출입구 별로 진행되며 전체적으로 빌딩 내 들어오는 인원과 나가는 인원을 합산하여 빌딩 내 전체 재실자를 추정하는 방식이다. 실험 결과 제안하는 물체의 동선 검출 알고리즘을 YoloV5 분석 후에 적용하였더니 실시간 고속 처리가 가능하였고 기존 중앙선 검출 방식보다 향상된 검출 성능을 확인하였다.
Abstract
A technology that automatically detects the number of people entering and exiting a building is necessary for the evacuation of victims in the event of a disaster such as a fire. As a solution for this, this paper proposes a method of automatically detecting the volume area of entry and exit personnel in real time using a thermal imaging camera installed at the entrance. The first step of such a detection system is to predict the exact number of people in an image, and the YoloV5 deep learning module was applied in this paper. The next step is the moving direction analysis algorithm of the thermal imaging object proposed in this paper. This algorithm analyzes the sequence of sequentially input thermal volume images and counts movement by distinguishing between incoming and outgoing people. The proposed algorithm divides the main area where the occupant moves in the image into 4 subareas and tracks the shape of the object volume and distribution change in each subarea. The proposed method showed stable detection performance even with various movement variations of pedestrians occurring at the entrance. The occupant analysis in this way is performed for each entrance of the building, and the total number of occupants in the building is estimated by adding the number of people entering and leaving the building as a whole. As a result of the experiment, when the proposed motion detection algorithm of the object was applied after YoloV5 analysis, real-time high-speed processing was possible and improved detection performance was confirmed compared to the existing centerline detection method.
참고문헌 (16)
- H. Ruser, "People counter based on fusion of reflected light intensities from an infrared sensor array", Informatik 2006 Conference, pp.379~383 (2006)
- C. Wang, "Deep people counting in extremely dense crowds", MM '15: Proceedings of the 23rd ACM International Conference on Multimedia, pp.1299~1302 (2015)
- A. Krizhevsky, "Imagenet classification with deep convolutional neural networks", Advances in neural information processing systems, pp.1097~1105 (2012)
- H. S. Kwon, "High Speed People Counting Using Neural Network and Cluster Analysis Based on Infrared Video", Summer Annual Conference of IEIE, 43(01), pp.2193~2197 (2020), The Institute of Electronics and Information Engineers
- Q. Zeng, "People detection in crowded scenes using hierarchical features", IEEE International Conference on Imaging Systems and Techniques(IST) (2017)
- F. Dittrich, "People counting in crowded and outdoor scenes using a hybrid multi-camera approach", arXiv : 1704. 00326v2, ~ (2017)
- Multi-view people counting system –Pedestrian representation, "J. M. Wang", Pattern Recognition and Machine Vision: In Honor and Memory of Late Professor King-Sun, 1~16 (2010)
- J. Cho, "Moving object detection based on optical flow estimation and a Gaussian mixture model for advanced driver assistance systems", Sensors, 19(14), pp.3217~ (2019)[DOI]
- K. Desai, "Cybernetics, Cognition and Machine Learning Applications. Algorithms for Intelligent Systems" (2020), Springer
- G. Jocher, "NanoCode012, ChristopherSTAN, L. Changyu, Laughing, tkianai, yxNONG, A. Hogan, Ultralytics", yolov5: v4.0 - nn.SiLU()activations, weights & biases logging, 2021
- "2015", ~
- K. Lee, "Brightness-based convolutional neural network for thermal image enhancemen", IEEE Access, 5, pp.26867~26879 (2017)
- M. Thom, "Sparse activity and sparse connectivity in supervised learning", Journal of Machine Learning Research, 14, pp.1091~1143 (2013)
- J. Redmon, "You only look once: Unified, real-time object detection", Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, pp.779~788 (2016)
- "2019", ~
- V. Nogueira Jr, "RetailNet: A deep learning approach for people counting and hot spots detection in retail stores", 32nd SIBGRAPI Conference on Graphics, Patterns and Images (2019)