基于毫米波雷达和视觉的车辆检测识别方法
With the increase in the number of vehicles equipped with anti-collision radar, the
probability of being affected by mutual interference for the car anti-collision radar on the road is
getting higher and higher. In order to keep the reliability of anti-collision radar, the
anti-interference technology plays a more and more important role. Referring to the
agility-frequency method of traditional radar, this paper improves the original waveform by
randomly choosing a group of combination from two kinds of modulation waveforms with a
fixed number of constant frequency and triangular frequency, and considers the combined
waveform as the improved radar transmitting one. The slope of constant frequency and triangular
frequency are randomly periodic changed. Simulation results show that the proposed method can
not only reduce the false alarm rate caused by mutual interference, but also improve the ability of
Although the radar can effectively detect the distance and velocity information of the
obstacles in front of the vehicles , it can not identify the obstacles while the machine vision can
compensate for it. In recent years, the research and development of Deep Learning has improved
the accuracy of image recognition greatly, and the machine vision has made great achievements
in the application of target recognition. Although machine vision has many advantages, it is lack
of real-time performance and vulnerable to the environment features, such as weather, ray of
light and other factors. Thus, a fusion method is proposed in this paper, that is, the radar obtains
the coordinates of the radar system in front of the vehicle, and then convert it into the pixel
coordinates of the corresponding image. Finally the convolution neural network model is used to
identify the vehicles around the pixel coordinates. The computation quantity of this fusion
method is greatly reduced compared to traditional image recognition algorithm using sliding
window method. The main research contents are as follows:
1. Target detection and waveform improvement of millimeter wave radar. Improve the
existing radar waveforms to eliminate the influence of cross-interference and realize the
multi-target detection Simulation experiments are carried out to verify the effectiveness of the
2. Research on vehicle recognition algorithm based on convolution neural network. Collect
the image samples firstly, and then the samples are preprocessed by gray level and
standardization. Finally conduct the vehicle identification by using the convolution neural
network. The study includes the construction of a convolution neural network with two
convolution layers, two pool layers, one full connection layer and one output layer as well as a
vehicle identification training set. Rectified Linear Units(RLU) are used as the activation
function for both convolution layer and full connected layer in the convolutional neural network
and the Softmax function is used as the output function. Use TensorFlow open source based
deep learning framework to train the convolution neural network model, and build the two
classification model used for vehicle recognition finally.
3. Analysis and design of radar and visual fusion based method. The principle of calibration
rules for anti-collision radar and machine vision is analyzed, and the conversion equations on the
spacing coordinates and the sampling period on the time window are given. Anti-collision radar
provides depth information to assist machine vision for vehicle recognition. Firstly, radar
provides the candidate region of target and then its coordinates in the radar system is converted
into the corresponding image pixel one, so that the trained convolutional neural network model
can be used for recognizing the vehicle in the candidate region. Since the target search area is
reduced in the image, the calculation amount of the image processing is also relatively reduced.
Key words: Millimeter-Wave Radar, Machine Vision, Convolution Neural Network, Algorithm
【后台留言，索取完整资料】
