Pedestrian detection method based on binarized convolutional neural network
A convolutional neural network and pedestrian detection technology, applied in the field of electric digital data calculation and calculation, can solve the problems of difficulty in implementation, unsuitable full-precision neural network, poor flexibility, etc., and achieves short memory access time, good application prospects, and memory occupation. less effect
Inactive Publication Date: 2018-07-20
SOUTHEAST UNIV +1
View PDF2 Cites 14 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
These classification algorithms are slow, inflexible, and have limitations in application scenarios
[0004] The full-precision convolutional neural network has a large number of learning parameters, which can fit complex nonlinear problems such as correct pedestrian detection frames. However, due to the slow calculation speed on the PC, the cost of using the GPU is too high and the power is too high. Large, not very suitable for implementation on PC
However, embedded systems have limited computing power and cannot carry high-power GPUs, making it more difficult to implement pedestrian detection on them.
The present invention aims to propose a pedestrian detection method based on a binary neural network to solve the problem that the full-precision neural network is not suitable for running on a PC and is difficult to implement in an embedded system
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0025] The technical solution of the invention will be described in detail below in conjunction with the accompanying drawings.
[0026] Such as Figure 7 As shown, the obtained color picture or video frame is sampled to a size of 416*416*3 through bilinear interpolation, input into the trained binary convolutional neural network, and a series of prediction frames are obtained, and then non-maximum simulation is performed Operation, get the final detection frame containing the pedestrian and display it.
[0027] The binarization method of the convolutional layer of the binarized convolutional neural network and its input is as follows: figure 1 As shown, first find the average of the absolute value of the weight parameter of each convolution kernel group in the convolutional layer of this layer as the weight parameter α of this layer, and then binarize each weight parameter, that is, if the weight parameter is greater than 0, then Binarize to 1, and if the weight parameter i...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention, which belongs to the technical field of electrical digital data calculation for pedestrian detection, discloses a pedestrian detection method based on a binarized convolutional neural network. Binaryzation is carried out on input data of a convolutional layer and a weight parameter of a convolution kernel group to obtain an unsigned integer array; and convolution calculation is realized by the XNOR bit operation and the BITCOUNT operation of the integer array. Therefore, the computing load is reduced substantially; less memory is occupied; the access memory time is shortened; and the convolution speed is increased. The pedestrian detection method having broad application prospects can be realized at a PC and can be applied to an embedded system.
Description
technical field [0001] The invention discloses a pedestrian detection method based on a binarized convolutional neural network, and belongs to the technical field of calculation and calculation of electrical digital data for pedestrian detection. Background technique [0002] The application of deep learning in computer vision can be said to be a milestone in the development of computer vision. Various novel methods emerge in an endless stream, changing the traditional cumbersome process of first manually designing features and then extracting artificial features and then performing classification and recognition processing. In terms of accuracy, it exceeds the traditional method. [0003] As an important branch of computer vision, pedestrian detection has been widely used in many fields such as automobile assisted driving, field search and rescue, advanced human-computer interaction, and intelligent monitoring. Traditional pedestrian detection is mostly based on feature ex...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Applications(China)
IPC IPC(8): G06K9/00G06K9/38G06N3/04G06N3/08
CPCG06N3/08G06V40/103G06V10/28G06N3/045
Inventor 陆生礼杨海平庞伟戎海龙韩志李硕
Owner SOUTHEAST UNIV