일부 네거리나 혼잡도로에서 특정 시간대에 행인이 많고 도로가 막혀서 발생하는 교통사고가 적지 않다. 특히 인근에 학교교차로가 있어 바쁜 시간에 학생들의 교통안전을 지키는 것이 중요하다. 과거에는 교통 신호등을 설 계 했을 때 행인의 안전성을 고려하지 않고 자동차 인식과 교통 최적화에 대하여 연구 했다. 행인, 특히 학생들의 안전을 확보하는 전제에서 가능한 한 도로의 소통을 유지하는 것이 본 연구의 중점적인 연구 방향이다. 본 연구는 사람, 오토바이, 자전거, 자동차, 버스의 식별문제를 중점적으로 연구할 것이다. 조사와 비교를 통해 본 연구는 YOLO v4 네트워크로 목표물의 위치와 수량을 식별하는 것을 제시한다. YOLO v4는 작은 목표물의 식별 능력이 강하고 정밀도가 높으며 처리속도가 빠르다는 특징을 가지고 있으며, 데이터 수집 대상을 설정하여 이미지 집합을 훈련하고 테스트 한다. 움직이는 영상에서 목표물의 정확도, 실수율과 누락율에 대한 통계를 사용하여, 본 연구에서 훈련된 네트워크는 움직이는 이미지 속의 사람, 오토바이, 자전거, 자동차와 버스를 정확하게 식별 할 수 있다.
일부 네거리나 혼잡도로에서 특정 시간대에 행인이 많고 도로가 막혀서 발생하는 교통사고가 적지 않다. 특히 인근에 학교교차로가 있어 바쁜 시간에 학생들의 교통안전을 지키는 것이 중요하다. 과거에는 교통 신호등을 설 계 했을 때 행인의 안전성을 고려하지 않고 자동차 인식과 교통 최적화에 대하여 연구 했다. 행인, 특히 학생들의 안전을 확보하는 전제에서 가능한 한 도로의 소통을 유지하는 것이 본 연구의 중점적인 연구 방향이다. 본 연구는 사람, 오토바이, 자전거, 자동차, 버스의 식별문제를 중점적으로 연구할 것이다. 조사와 비교를 통해 본 연구는 YOLO v4 네트워크로 목표물의 위치와 수량을 식별하는 것을 제시한다. YOLO v4는 작은 목표물의 식별 능력이 강하고 정밀도가 높으며 처리속도가 빠르다는 특징을 가지고 있으며, 데이터 수집 대상을 설정하여 이미지 집합을 훈련하고 테스트 한다. 움직이는 영상에서 목표물의 정확도, 실수율과 누락율에 대한 통계를 사용하여, 본 연구에서 훈련된 네트워크는 움직이는 이미지 속의 사람, 오토바이, 자전거, 자동차와 버스를 정확하게 식별 할 수 있다.
In some intersections or busy traffic roads, there are more pedestrians in a specific period of time, and there are many traffic accidents caused by road congestion. Especially at the intersection where there are schools nearby, it is particularly important to protect the traffic safety of students ...
In some intersections or busy traffic roads, there are more pedestrians in a specific period of time, and there are many traffic accidents caused by road congestion. Especially at the intersection where there are schools nearby, it is particularly important to protect the traffic safety of students in busy hours. In the past, when designing traffic lights, the safety of pedestrians was seldom taken into account, and the identification of motor vehicles and traffic optimization were mostly studied. How to keep the road smooth as far as possible under the premise of ensuring the safety of pedestrians, especially students, will be the key research direction of this paper. This paper will focus on person, motorcycle, bicycle, car and bus recognition research. Through investigation and comparison, this paper proposes to use YOLO v4 network to identify the location and quantity of objects. YOLO v4 has the characteristics of strong ability of small target recognition, high precision and fast processing speed, and sets the data acquisition object to train and test the image set. Using the statistics of the accuracy rate, error rate and omission rate of the target in the video, the network trained in this paper can accurately and effectively identify persons, motorcycles, bicycles, cars and buses in the moving images.
In some intersections or busy traffic roads, there are more pedestrians in a specific period of time, and there are many traffic accidents caused by road congestion. Especially at the intersection where there are schools nearby, it is particularly important to protect the traffic safety of students in busy hours. In the past, when designing traffic lights, the safety of pedestrians was seldom taken into account, and the identification of motor vehicles and traffic optimization were mostly studied. How to keep the road smooth as far as possible under the premise of ensuring the safety of pedestrians, especially students, will be the key research direction of this paper. This paper will focus on person, motorcycle, bicycle, car and bus recognition research. Through investigation and comparison, this paper proposes to use YOLO v4 network to identify the location and quantity of objects. YOLO v4 has the characteristics of strong ability of small target recognition, high precision and fast processing speed, and sets the data acquisition object to train and test the image set. Using the statistics of the accuracy rate, error rate and omission rate of the target in the video, the network trained in this paper can accurately and effectively identify persons, motorcycles, bicycles, cars and buses in the moving images.
After selecting the target detection algorithm, the next main task is to configure the software, build a deep learning platform, and then configure the parameters of YOLO v4 network according to the identification content of this paper, and finally train the YOLO v4 network.
The main research content of this paper includes the recognition of person, motorcycle, bicycle, car and bus, so it is necessary to establish a dataset for YOLO v4 training. In this paper, the main ideas of establishing dataset are as follows: identify the object, collect the dataset, label the dataset, and construct the PASCAL VOC2012 dataset available for YOLO v4.
The main research content of this paper includes the recognition of person, motorcycle, bicycle, car and bus, so it is necessary to establish a dataset for YOLO v4 training. In this paper, the main ideas of establishing dataset are as follows: identify the object, collect the dataset, label the dataset, and construct the PASCAL VOC2012 dataset available for YOLO v4.
대상 데이터
In this video, 22325 boxes were identified, which can accurately judge the area where travelers and electric vehicles are located. At the same time, the processing speed reaches about 30fps, which is not different from the real-time (30fps).
Therefore, when selecting pictures, in addition to the pictures provided by VOC2012, the images of UA-DETAC (a challenging real-world multi-object detection and multi-object tracking benchmark) dataset are also added, as shown in Fig 1. The UA-DETRAC benchmark dataset consists of 100 challenging videos captured from real-world traffic scenes (over 140, 000 frames with rich annotations, including illumination, vehicle type, occlusion, truncation ratio, and vehicle bounding boxes) for multi-object detection and tracking. After sorting out, the testing dataset collected about 5100 images, and the training dataset collected about 11000 images.
The targets of this training test are person, motorcycle, bicycle, car and bus, because the detection of fewer targets, so the speed is relatively fast. In the figure, "40000" refers to the number of iterations of the current training; in the figure, "0.
성능/효과
After four days of training, the average loss of training results is less than 0.060730 avg on the deep learning experimental platform built in this paper, and the training results have been satisfactory. Due to too many data results, this paper only shows the final training results, and the recognition results are shown in Fig 4.
The first task of this paper is to select a target detection algorithm with high accuracy and good real-time performance. By comparing and selecting the target detection algorithms based on CNN, the YOLO v4 target detection algorithm with high accuracy, good real-time performance and good detection effect on small targets is finally selected.
In this paper, based on the network of Yolo V4, the VOC dataset suitable for YOLO v4 is constructed, and the ideal recognition effect is achieved. In the selection of data sets, in addition to some of the pictures in the VOC data set, we also use some pictures from the ua-detac data set.
In this paper, firstly, the fixed image of electric motorcycle and pedestrian are identified. The recognition results show that the trained YOLO v4 network has high recognition accuracy, which can be applied in the field of unmanned driving and traffic target recognition.
With the continuous development of target detection algorithm based on deep learning, the recognition accuracy and recognition speed have been greatly improved. Therefore, it is feasible to use the target detection algorithm based on deep learning to recognize pedestrians, motorcycles and vehicles.
후속연구
With the continuous development of target detection algorithm based on deep learning, the recognition accuracy and recognition speed have been greatly improved. Therefore, it is feasible to use the target detection algorithm based on deep learning to recognize pedestrians, motorcycles and vehicles.
참고문헌 (8)
Ensanian A. Discovering the Motorcycle: The History. The Culture. The Machines[M]. Hillcrest Publishing Group, 2016:35-50
ZHANG W. Ataraxia. Emotional care & driverless car service[J]. 2018.
Yi Z, Yongliang S. An improved tiny-yolov3 pedestrian detection algorithm[J]. Optik, 2019, 183:17-23. DOI:10.1016/j.ijleo.2019.02.038
Silver D, Huang A, Maddison C J, et al. Mastering the game of Go with deep neural networks and tree search[J]. nature, 2016, 529(7587): 484-489. DOI: https://doi.org/10.1038/nature16961
Liu S, Huang D, Wang Y. Pay Attention to Them: Deep Reinforcement Learning-Based Cascade Object Detection[J]. IEEE Transactions on Neural Networks and Learning Systems, 2019.
Espinosa J E, Velastin S A, Branch J W. Detection of Motorcycles in Urban Traffic Using Video Analysis: A Review[J]. IEEE Transactions on Intelligent Transportation Systems, 2020.
Behailu H. A Real-time Obstracle Detection And Interframe Information Storage System From Motion Pictures For Assisting Blind People[D], 2020.
Y. S. Im, E. Y. Kang, "MPEG-2 Video Watermarking in Quantized DCT Domain," The Journal of The Institute of Internet, Broadcasting and Communication(JIIBC), Vol. 11, No. 1, pp. 81-86, 2011.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.