Convolutional Neural Network for Person Detection using YOLO Framework
M. H. Putra, Zulkalnain Mohd Yussof, Sani Irwan Md Salim, Kim Lim · Journal of Telecommunication Electronic and Computer Engineering (JTEC) · 2017
In this paper we present a real-time person detection system suitable for use in Intelligent Car or Advanced Driver Assistance System (ADAS). The system is based on modified You only Look Once (YOLO) which uses 7 convolutional neural network layers. The experimental results demonstrate that the accuracy of the person detection system is reliable for real time operation. The performance of the detection is benchmarked using the standard testing datasets from Caltech and measured using Piotr’s Matlab Toolbox. The results benchmarking is emphasizing on the performance of false positive per image (FPPI) over the miss rate. ADAS demands both relatively good detection and accuracy in order to work in real time operation. A good detection result is marked by achieving low miss rate and low FPPI. This requirement was achieved by the modified YOLO with 28.5%, 26.4% and 22.7% miss rate at 0.1 FPPI and believed to be an excellent candidate for use in ADAS.