Abstract:
본 발명에서 제안하고 있는 인공위성 영상과 랜덤포레스트 분류기 결합을 이용한 자동 하천 검출 시스템 및 방법에 따르면, 위성 영상의 다중 스펙트럴 이미지로부터 TOA(Top Of Atmosphere) 반사도 및 WI(Water Index)를 특징벡터로 추출하고, 휴리스틱 임계값이나 자율 학습 방법 대신 TOA 반사도 및 WI를 이용해 복수의 타입의 랜덤포레스트 분류기를 학습하며, 학습된 분류기를 이용해 테스트 영상으로부터 하천 영역을 검출함으로써, 보다 정확하게 자동으로 하천을 분류할 수 있다.
Abstract translation:根据本发明提出的利用卫星图像和随机森林分类器组合的自动河流检测系统和方法,TOA(Top Of Atmosphere)反射率和WI 提取水指数)作为特征向量,并且通过使用启发式阈值或自学习方法,代替TOA反射率和WI,通过使用经训练的分类从测试图像中检测河流区域学习多种类型的随机森林分类,更 您可以自动准确地对河流进行分类。 P>
Abstract:
본 발명에서 제안하고 있는, 교사-학생 랜덤 펀을 이용한 다수의 보행자 추적 방법 및 시스템에 따르면, 심층 네트워크의 한 종류인 tiny YOLO를 사용하여 보행자의 특징값을 추출하고, 추출된 특징값을 이용하여 랜덤 펀(Random Ferns)을 학습함으로써, 실시간 학습이 가능하여 보행자의 형태변화, 크기변화로 인한 오-추적을 최소화할 수 있다.
Abstract:
본 발명에서 제안하고 있는 야간 환경에서의 운전자 보조 시스템을 위한 위험 보행자 검출 방법 및 시스템에 따르면, 스케일링 비율 및 보행자 탐색 영역을 설정하고 보행자 윈도우를 검출함으로써, 처리 시간을 단축하고 신속하게 보행자를 검출할 수 있으며, 차량의 진행 방향을 고려하여 기준선을 설정하고 위험 보행자를 판단함으로써, 보다 정확하게 보행자의 위험성을 결정할 수 있다.
Abstract:
The present invention relates to a real-time object tracking method in a moving camera using a particle filter. The real-time object tracking method according to the present invention includes the steps of: initializing a target model; generating a plurality of candidate particles with a random distribution; generating an observation model; determining a particle weight; estimating the state of the target object; and extracting particles to be used in a frame which is inputted next time again.
Abstract:
The present invention relates to a method for detecting a forest fire using spatiotemporal bag-of-features (BoF) and a random forest and, more particularly, to a method for detecting a forest fire using spatiotemporal bag-of-features and a random forest comprising the steps of: (1) whenever a frame of a video sequence is inputted, detecting the difference between the input frame and a previous frame and, if the difference value exceeds a predetermined first threshold, setting the input frame to a key frame; (2) detecting a moving block from the set key frame; (3) extracting a candidate smoke block from the moving block using a smoke color model; (4) generating BoF from the detected candidate smoke block; and (5) performing learning by a random forest with respect to the generated BoF to determine whether the smoke of the candidate smoke block is real. The method proposed by the present invention can set the key frame from the video sequence, extract the candidate smoke block using the non-parametric smoke color model, extract HOG and HOF from the extracted candidate smoke block to generate BoF as spatiotemporal features from the HOG and the HOF, perform learning by the random forest with respect to the generated BoF, thereby enhancing the capability of detecting a forest fire in real time, reducing a false alarm, and accurately detecting smoke caused by the forest fire. [Reference numerals] (AA) Start; (BB) End; (S100) Divide frames forming a video sequence into a plurality of blocks, respectively; (S200) Whenever a frame of the video sequence is inputted, detect the difference between the input frame and a previous frame and, if the difference value exceeds a predetermined first threshold, set the input frame to a key frame; (S300) Detect a moving block from the set key frame; (S400) Extract a candidate smoke block from the moving block using a smoke color model; (S500) Generate bag-of features (BoF) from the detected candidate smoke block; (S600) Perform learning by a random forest with respect to the generated BoF to determine whether the smoke of the candidate smoke block is real
Abstract:
PURPOSE: A method for detecting pills inside a medicine box is provided to improve the performance and the speed for detecting pills inside the medicine box, thereby accurately certifying that a medicine is a compounded original medicine. CONSTITUTION: A method for detecting pills inside a medicine box is as follows. A pill image detection system photographs loaded medicines under the transparent bottom surface of the medicine box with a camera (S100). An image information extractor of the system translates the obtained images into binary code and removes edges of the images (S200). An image analysis unit of the system applies circular filters of different sizes to the binary code images in stages, thereby detecting images of the pills (S300). [Reference numerals] (AA) Start; (BB) Finish; (S100) Obtain images by photographing loaded medicines; (S200) Translate the obtained images into binary code and extract binary images by removing edges of the images; (S310) Generate a first image by applying a circular filter algorithm of a size of 40 X 40 to the binary images; (S320) Generate an exclusive image by XOR calculating the first image and the binary images; (S330) Generate a second image by applying a circular filter algorithm of a size of 12 X 12 to the exclusive image; (S340) Generate a third image by applying a circular filter algorithm of a size of 28 X 28 to the binary images; (S350) Detect images of pills by OR calculating the second and third images