Vehicle Detection Algorithm of Small Target with Feature Fusion
Journal Title: 河南科技大学学报（自然科学版） - Year 2019, Vol 40, Issue 2
Abstract
In the current target detection based on deep convolution neural network, the high-dimensional features lost small-area feature information and target position information, which led to a low recognition rate for small targets. The detection method based on feature layer fusion was proposed. The high-dimensional feature map and the low-dimensional feature map were transformed into the same size through the image interpolation. Each feature map was effectively fused by setting a network self-learning parameter, so that the feature map finally detection had rich semantic information and as much target feature information as possible.Using the method, a simple convolutional neural network model was built to detect the long-distance vehicles in the road scene. The test on the KITTI dataset shows that the model's recall rates are increased by 5. 9%and 14. 6% respectively compared with the mainstream Faster-RCNN and SSD inspection framework.
Authors and Affiliations
Pengfei LUO, Ming LI
Keywords
Related Articles
- EP ID EP422474
- DOI 10.15926/j.cnki.issn1672-6871.2019.02.008
- Views 107
- Downloads 0
Energy Estimate for Coupled Nonlinear Schrodinger Equations with Neumann Boundary
The coupled nonlinear Schrodinger equations with Neumann boundary were considered in order to estimate the energy of the equations. Firstly,the existence of its solution by using the transform between concrete equations...
Design of High-precision Phase Measuring Instrument Based on All Phase FFT Method
To overcome low precision and complex structure of traditional phase measuring instrument, a highprecision phase measuring instrument based on all phase fast Fourier transform algorithm (FFT) was proposed.The STM32F407VG...
Verification of Type A Evaluation Methods of Uncertainty Based on Monte Carlo Method
To solve the reliability problem of uncertainty evaluation method based on objective probability, by using the advantages of Monte Carlo method (MCM) suitable for computer simulation, the accuracy of three kinds of type...
Design of Adjustable Ethernet Serial Server Based on ARM+FPGA
For the problems that the number of Ethernet serial device server's serial ports was fixed and was not be increased or decreased in real time, the Ethernet serial device server based on advanced risc machines ( ARM) and...
Track Control of Uncertain Nonlinear System with Time-varying Delay
For the track control problem of a class of uncertain nonlinear system with time-varying delay,the extended state observer was utilized to estimate the unknown term with time-varying delay. The nonlinear tracking differe...