Design of Multi-objective Recognition System for Road Scenes with Time and Space Fusion 2019-01-1047
Abstract：The identification technology of the road environment is an important part of the intelligent transportation system. By using visual technology to judge the road conditions when the vehicle is driving, it can detect the potential danger of the road and improve the safety of the vehicle. We conduct research on multi-target recognition technology in road scenes, propose a lightweight single-frame multi-target detection network, combine spatially-supervised recurrent neural networks to integrate historical location information from the detection network output on the timeline, and correct the recognition result of the current frame image, which can effectively improve the recognition accuracy of the recognition system. In this paper, the target appearance characteristics extracted from the single-frame multi-target detection network are sent to the recursive neural network with the function of long-term and short-term memory for prediction, and extends the learning and analysis of the neural network to the space-time domain. When LSTM interprets advanced visual features, spatially supervised regression is used as an online appearance model to regress features to a particular visual element/cue position through preliminary position inference. In the time domain, LSTM learns on sequence units to position predictions restricted to a specific space. We evaluate our proposed system on the KITTI data set and a large number of real scene experiments. The system can improve accuracy of identification by up to 3.9% and reduce the error detection rate by 5.2%. The experimental results show that the accuracy of road target detection can be effectively improved by learning historical visual semantics and target position information.
Key words：target recognition; recurrent neural network; deep learning; convolutional neural network; transfer learning; advanced driver assistance systems
Wang Huan, Jianning Chi, Chengdong Wu, Yu Xiaosheng