Front-view Car Detection and Counting with Occlusion in Dense Traffic Flow

Van Huy Pham,이병룡 제어·로봇·시스템학회 2015 International Journal of Control, Automation, and Vol.13 No.5

In dense traffic flow, car occlusion is usually one of the great challenges of vehicle detection and tracking in traffic monitoring systems. Current methods of car hypothesis such as symmetry or shadow based method work only with non-occluded cars. In this paper, we proposed an approach to car detection and counting using a new method of car hypothesis based on car windshield appearance which is the most feasible cue to hypothesize cars in occlusion situations. In hypothesis stage, Hough transformation is used to detect trapezoid-like regions where a car’s windshield could be located, and then candidate car regions are estimated by the windshield region and its size. In verification stage, HOG descriptor and a well-collected dataset are used to train a linear SVM classifier for detecting cars at a high accuracy rate. Then, a tracking process based on Kalman filter is used to track the movement of detected cars in consecutive frames of traffic videos, followed by rule-based reasoning for counting decision. Experimental results on real traffic videos showed that the system is able to detect, track and count multiple cars including occlusion in dense traffic flow in real-time.

시각 장애인의 안전 횡단을 위한 도로 위의 차량 검출 알고리즘 제안

이옥민(Okmin Lee),원인수(Insu Won),이상민(Sangmin Lee),권장우(Jangwoo Kwon) 한국장애인재활협회 2016 재활복지 Vol.20 No.2

본 연구에서는 시각장애인의 보행안정성 향상을 위해 도로 위를 이동하는 자동차의 영상만을 입력 받아 자동차를 검출하는 방법을 제안한다. 입력 영상은 제약 조건이 있다. 도로 위에서 아래 방향을 비스듬히 내려 보는 고정된 시야를 가져야한다는 점이다. 주어진 영상 중 도로 영역만을 이용하기 위해 동적인 관심영역을 검출해 적용한다. 동적으로 관심영역을 검출하기 위해 캐니엣지 탐지를 이용한 후 허프 변환을 응용해 분할 허프 변환 방법으로 구현하였다. 관심영역 내에서 차량 검출을 위해 모션 히스토리 이미지(Motion History Image) 추출 방법, SIFT(Scale-Invariant Feature Transform) 알고리즘, 히스토그램 분석 등 다양한 영상처리 방법을 적용한 실험결과, 평가 및 한계점을 서론에 제시했다. 이를 해결하기 위해서 GMM(Gaussian Mixture Model)을 응용해 단순한 물체 탐지가 아닌, 차량 검출에 최적화하는 방법을 제시한다. 본 연구에서는 GMM 알고리 즘을 응용한 차량 검출 GMM(Vehicle Detection-GMM) 알고리즘을 이용해 차량 검출 시스템을 구현했다. 실험 결과 정확률, 재현율, F1 측정값이 GMM은 각각 15%, 60%, 24%인데 비해 차량검출 GMM은 각각 75%, 60%, 67%로 그 성능을 입증할 수 있었다. 그 결과를 바탕으로 제안한 알고리즘이 기존의 알고리즘보다 우수함을 입증하였다. In this paper we propose car detection system using only video image as the input image that is composed of car moving on the road. It is for the people who is visually handicapped. There is a constraint condition that is the input image has downward obliquely from road and fixed view. To use only region of road, we use dynamic ROI detection using Canny edge detection and separated type hough transform based on Hough transform. To do car detection in ROI, we analyze and evaluate confines of motion history image(MHI) method, SIFT(Scale-Invariant Feature Transform) algorithm and histogram analysis in the introduction. These method has the critical point. So, we proposed to use VD-GMM(Vehicle Detection-GMM) based on GMM(Gaussian Mixture Model) for the car detection system. GMM is just moving object detection but including several geometric detection method, it could detect the car. In the experiment, we measure precision, recall and F1 rate for GMM and VD-GMM. GMM has 15%, 60%, 24% and VD-GMM has 75%, 60%, 67% for each value. Using this result, we could prove VD-GMM is better than GMM for the car detection in the conclusion.

주차장 환경에서의 차량 사고 검출

정우진(Woo Jin Jeong),이종민(Jong Min Lee),박기태(Ki Tae Park),문영식(Young Shik Moon) 대한전자공학회 2015 전자공학회논문지 Vol.52 No.3

본 논문에서는 주차장 환경에서의 차량 사고 감지방법을 제안한다. 제안하는 방법은 차량 검출, 차량 추적, 사고 감지의 3단계로 구성된다. 차량 검출 단계에서는 픽셀 기반의 전경 검출 방법과 모션맵을 이용하여 차량을 검출하고, 차량 추적 단계에서는 검출된 차량 정보를 바탕으로 차량의 이동을 추적한다. 마지막 단계인 차량 사고 감지 단계에서는 차량의 이동 방향에 맞추어 사고 감지 영역을 지정하고 사고 감지 영역에서 발생하는 움직임의 변화량을 분석하여 차량 사고를 감지한다. 실험을 통해 제안하는 방법은 주차장 환경에서 발생하는 차량 사고를 효과적으로 검출함을 보였다. We propose a detecting method for a car accident in parking lots. The proposed method consists of 3 parts : car detection, car tracking, and accident detection. In the car detection part, we detect the car using the pixel based foreground extraction method and the motion map. From the result of the car detection, the moving car is tracked. In the accident detection part, we set the accident detecting region in front of car, and then the car accident is detected using the difference of the motion. Experimental results show that the proposed method effectively detects the car accident in the parking lots.

자동차의 자기 주행차선 검출을 위한 시각 센싱

김동욱,도용태 한국센서학회 2018 센서학회지 Vol.27 No.2

Detecting the ego-lane of a vehicle (the lane on which the vehicle is currently running) is one of the basic techniques for a smart car. Vision sensing is a widely-used method for the ego-lane detection. Existing studies usually find road lane lines by detecting edge pixels in the image from a vehicle camera, and then connecting the edge pixels using Hough Transform. However, this approach takes ratherlong processing time, and too many straight lines are often detected resulting in false detections in various road conditions. In this paper,we find the lane lines by scanning only a limited number of horizontal lines within a small image region of interest. The horizontal imageline scan replaces the edge detection process of existing methods. Automatic thresholding and spatiotemporal filtering procedures arealso proposed in order to make our method reliable. In the experiments using real road images of different conditions, the proposed method resulted in high success rate.

Vehicle Detection in Aerial Images Based on Hyper Feature Map in Deep Convolutional Network

( Jiaquan Shen ),( Ningzhong Liu ),( Han Sun ),( Xiaoli Tao ),( Qiangyi Li ) 한국인터넷정보학회 2019 KSII Transactions on Internet and Information Syst Vol.13 No.4

Vehicle detection based on aerial images is an interesting and challenging research topic. Most of the traditional vehicle detection methods are based on the sliding window search algorithm, but these methods are not sufficient for the extraction of object features, and accompanied with heavy computational costs. Recent studies have shown that convolutional neural network algorithm has made a significant progress in computer vision, especially Faster R-CNN. However, this algorithm mainly detects objects in natural scenes, it is not suitable for detecting small object in aerial view. In this paper, an accurate and effective vehicle detection algorithm based on Faster R-CNN is proposed. Our method fuse a hyperactive feature map network with Eltwise model and Concat model, which is more conducive to the extraction of small object features. Moreover, setting suitable anchor boxes based on the size of the object is used in our model, which also effectively improves the performance of the detection. We evaluate the detection performance of our method on the Munich dataset and our collected dataset, with improvements in accuracy and effectivity compared with other methods. Our model achieves 82.2% in recall rate and 90.2% accuracy rate on Munich dataset, which has increased by 2.5 and 1.3 percentage points respectively over the state-of-the-art methods.

YOLO v8을 활용한 컴퓨터 비전 기반 교통사고 탐지

이충권,노미진,Chacha Andrea Marwa,문상일,김양석,신재호 한국산업정보학회 2024 한국산업정보학회논문지 Vol.29 No.1

Car accidents occur as a result of collisions between vehicles, leading to both vehicle damage and personal and material losses. This study developed a vehicle accident detection model based on 2,550 image frames extracted from car accident videos uploaded to YouTube, captured by CCTV. To preprocess the data, bounding boxes were annotated using roboflow.com, and the dataset was augmented by flipping images at various angles. The You Only Look Once version 8 (YOLOv8) model was employed for training, achieving an average accuracy of 0.954 in accident detection. The proposed model holds practical significance by facilitating prompt alarm transmission in emergency situations. Furthermore, it contributes to the research on developing an effective and efficient mechanism for vehicle accident detection, which can be utilized on devices like smartphones. Future research aims to refine the detection capabilities by integrating additional data including sound.

Fast Vehicle Detection using Correlation Filters

Sangpil Han,Min-jae Kim,Seokmok Park,Joonki Paik 대한전자공학회 2017 IEIE Transactions on Smart Processing & Computing Vol.6 No.5

Object detection is very challenging research in the computer vision community, and vehicle detection has become an important issue in various applications, such as unmanned systems and intelligent transportation systems. Most of these applications require fast and accurate vehicle detection. In recent years, it has been proven that correlation filters can find target objects fast and accurately owing to Parseval’s theorem and dense sampling. However, we think that the existing correlation filters have not used all the helpful information. Therefore, we propose a robust and fast vehicle detection method based on an improved correlation filter framework that exploits the additional information from correlation filters. The proposed vehicle detection algorithm consists of five steps: i) training the correlation filter, ii) correlation of input and the trained filter, iii) finding local maxima as vehicle candidates in the correlation output, iv) filtering the candidates by using the shape and sharpness of the maxima, and v) estimation of the location and scale of the vehicles. The proposed algorithm runs fast and accurately, so it can be applied to many other applications, such as object alignment, object detection, and object tracking. We evaluated the proposed algorithm performance by comparing it with the state-of-the-art correlation filter-based methods.

자동차 경성 사기 방지방안에 관한 연구 -고의충돌을 중심으로-

이동임 사단법인 한국안전문화학회 2022 안전문화연구 Vol.- No.16

The main culprit of car hard fraud is using social media to recruit people through a legal method of recruiting part-time jobs and engaging them in crimes. Through this, criminals intentionally cause traffic accidents targeting vehicles that violate the law and acquire insurance and settlement money from the victim driver or insurance company. As a result, the purpose of the study is to present measures to prevent insurance leakage and increase insurance premiums for auto insurance subscribers. The method of research is to identify the status and trend of insurance fraud by the Financial Supervisory Service, identify cases of auto-hard fraud based on Internet and domestic and foreign research data, and identify what measures insurers are taking to prevent damage to insurance fraud. As a result of the study, car-hard fraud is causing an accident by intentionally colliding a vehicle that violates traffic laws. Through this, the high proportion of the victim drivers' negligence and the fact that the perpetrator claims to be a negligence accident are exploited by the insurance company. Insurance companies are responding by establishing a system that can detect habitual criminals, but it is less efficient in crime prediction and detection, and it seems that there is a limit for insurance companies to respond to and detect first-time criminals. Therefore, it was derived that it is possible to prevent insurance fraud only by preparing countermeasures through a more scientific and detailed approach. In conclusion, car-hard fraud are becoming increasingly intelligent, and as a result, it is difficult to arrest criminals, so crimes are on a continuous increase. In order to eradicate such crimes, a control tower should be created under the Financial Supervisory Service to predict and arrest insurance fraud crimes. The first offender of a fraudulent crime using a car has a limit that cannot be arrested. Therefore, a detective system should be established to approach from the recruitment stage through social media, participate in crime execution, secure evidence, and request an investigation to an investigative agency. Car fraud criminals target vehicles that violate traffic laws, so if an accident occurs in violation of traffic laws, the driver is obligated to receive compliance education from the Korea Transportation Safety Authority, and The punishment regulations should be revised so that criminals can be reasonably punished.

UWB 통신을 이용한 센서 응용: 자동차 내 좌석 탑승 감지

김창진,한병근,이준영,정선우,장병준 한국전자파학회 2023 한국전자파학회논문지 Vol.34 No.4

UWB(ultra-wideband) technology has recently attracted attention owing to its ability to accurately measure distance and enable communication. Currently, applications such as sensors capable of recognizing situations are being considered, moving beyond communi- cation and distance measurement. In this study, we developed a UWB-based sensor application system that detects whether a person is seated in a car. With this system, it is possible to send an alert to the driver’s smartphone if the driver leaves the car with a child in the back seat without a separate seat detection sensor. The proposed sensor system uses changes in the channel status of the vehicle, which are recognized when the UWB communication device performs the communication function, to detect whether all seats in a car are occupied. The proposed system, using the DW1000 module, demonstrated detection performance of 93 % or more in the case of one receiver, 97 % in the case of two receivers, and 100 % in the case of four receivers for all nine scenarios where four people were seated in a car in different combinations of seats.

Detection of Pedestrians Employing a Wide-angel Camera

Ryuichi Matsuda,Joo Kooi Tan,Hyoungseop Kim,Seiji Ishikawa 제어로봇시스템학회 2011 제어로봇시스템학회 국제학술대회 논문집 Vol.2011 No.10

Recently, the number of accidents that pedestrians have law violation is in the tendency of decrease in Japan. However, accidents caused by pedestrians crossing a crosswalk or dashing into a crosswalk still have high ratio, and both accident sources account for 15% of the whole number of accidents caused by a pedestrian. Although many researches in ITS in which pedestrians are detected from in-vehicle cameras have been actively done to solve these problems, they usually employ standard cameras, and those pedestrians who exist outside of the camera view cannot be detected. In this paper, we employ a wide-angle camera which has wider view than a general camera and propose a technique for detecting pedestrians from the wide-angle image. Since, in a wide-angle camera image, every object becomes smaller, we propose a technique for detecting pedestrians employing optical flows converging to a FOE (Focus of Expansion). Experimental results show satisfactory performance of the technique.