Fast object detection based on binary deep convolution neural networks

• Published in: CAAI Transactions on Intelligence Technology Vol. 3; no. 4; pp. 191 - 197
• Main Authors: Sun, Siyang, Yin, Yingjie, Wang, Xingang, Xu, De, Wu, Wenqi, Gu, Qingyi
• Format: Journal Article
• Language: English
• Published: Beijing The Institution of Engineering and Technology 01.12.2018; John Wiley & Sons, Inc; Wiley
• Subjects: 62 times faster convolutional operations; Accuracy; Algorithms; Artificial neural networks; B6135 Optical, image and video signal processing; binary deep CNNs; binary deep convolution neural networks; binary operation; binary quantisation; C5260B Computer vision and image processing techniques; C5290 Neural computing techniques; Classification; convolution; convolution kernels; deep CNN; Energy consumption; fast object detection algorithm; faster object detection; Feature maps; Field programmable gate arrays; full-precision convolution; Mean square errors; Methods; multiscale objects; neural nets; Neural networks; object detection; object detection results; Object recognition; Proposals; rapid object detection; Research Article; System effectiveness; Telematics; binary deep convolution neural networks; full-precision convolution; object detection; fast object detection algorithm; binary operation; binary quantisation; faster object detection; rapid object detection; deep CNN; binary deep CNNs; convolution; convolution kernels; multiscale objects; neural nets; object detection results; 62 times faster convolutional operations
• ISSN: 2468-2322; 2468-6557; 2468-2322
• DOI: 10.1049/trit.2018.1026

In this study, a fast object detection algorithm based on binary deep convolution neural networks (CNNs) is proposed. Convolution kernels of different sizes are used to predict classes and bounding boxes of multi-scale objects directly in the last feature map of a deep CNN. In this way, rapid object detection with acceptable precision loss is achieved. In addition, binary quantisation for weight values and input data of each layer is used to squeeze the networks for faster object detection. Compared to full-precision convolution, the proposed binary deep CNNs for object detection results in 62 times faster convolutional operations and 32 times memory saving in theory, what's more, the proposed method is easy to be implemented in embedded computing systems because of the binary operation for convolution and low memory requirement. Experimental results on Pascal VOC2007 validate the effectiveness of the authors’ proposed method.