2011.04244.pdf - Real-time object detection method for...

1
Real-time object detection method for
embedded devices
ZICONG JIANG 1 , LIQUAN ZHAO 1 , SHUAIYANG LI 1 and YANFEI JIA 2
1 Key Laboratory of Modern Power System Simulation and Control and Renewable Energy Technology, Ministry of Education, Northeast Electric
Power University, jilin, jilin 132012, China
2 College of Electrical and Information Engineering, Beihua University, Jilin 132013, China;
Corresponding author: Zhao Liquan (: [email protected]; Tel.: [phone]).
This work was supported by the National Natural Science Foundation of China (61271115)
ABSTRACT The “ You only look once v4 ” (YOLOv4) is one type of object detection methods in deep
learning. YOLOv4-tiny is proposed based on YOLOv4 to simple the network structure and reduce
parameters, which makes it be suitable for developing on the mobile and embedded devices. To improve
the real-time of object detection, a fast object detection method is proposed based on YOLOv4-tiny. It
firstly uses two ResBlock-D modules in ResNet-D network instead of two CSPBlock modules in Yolov4-
tiny, which reduces the computation complexity. Secondly, it designs an auxiliary residual network block to
extract more feature information of object to reduce detection error. In the design of auxiliary network, two
consecutive 3x3 convolutions are used to obtain 5x5 receptive fields to extract global features, and channel
attention and spatial attention are also used to extract more effective information. In the end, it merges the
auxiliary network and backbone network to construct the whole network structure of improved YOLOv4-
tiny. Simulation results show that the proposed method has faster object detection than YOLOv4-tiny and
YOLOv3-tiny, and almost the same mean value of average precision as the YOLOv4-tiny. It is more
suitable for real-time object detection, especially for developing on embedded devices .
INDEX TERMS Deep learning, real-time object detection, YOLO, auxiliary residual network
I. INTRODUCTION
Object detection method based on deep learning mainly
includes two types: region proposal-based two-stage method
and regression-based one-stage method [1-2]. The typical
two- stage methods include region-based convolution neural
network (R-CNN) method [3]. Fast R-CNN [4], Faster R-
CNN [5] method, region-based fully convolutional networks
(R-FCN) method [6], light head R-CNN method and other
improve method based on convolution neural network [7-8].
Although two-stage method has higher accuracy than one-
stage method, the one-stage method has faster detection
speed than two-stage method [9-10]. The one-stage method is
more suitable for application in some conditions that require
higher real-time.
The You Only Look Once (YOLO) method [11] proposed
by Redmon, et al. is the first regression-based one stage
method. Redmon, et al. also proposed the You Only Look
Once version 2 (YOLOv2) [12] based on YOLO by deleting
fully connected layer and the last pooling layer, using anchor
boxes to predict bounding boxes and designing a new basic
network named DarkNet-19. The You Only Look Once
version 3 (YOLOv3) [13] is the last version of YOLO
method proposed by Redmon, et al. It introduces feature
pyramid network, a batter basic network named darknet-53
and binary cross-entropy loss to improve the detection
accuracy and the ability of detecting smaller object. Due to
the type of information fusion employed by YOLOv3 does
not make full use of low-level information, a weakness