A Multi-Objective Recognition Algorithm with Time and Space Fusion

Multi-target recognition technology plays an important role in the field of intelligent driving. In this paper, we propose a novel multi-target recognition algorithm with high accuracy and efficiency. We design a time series based recurrent neural network that integrates historical appearance information on the timeline, which can effectively improve the recognition accuracy. The target appearance characteristics extracted from the feature fusion network are then sent to the recursive neural network with the function of long-term and short-term memory for prediction, extending the learning and analysis of the neural network to the space-time domain. After the LSTM interprets advanced visual features, time series based regression is used as an appearance model to regress features to a particular visual element position through preliminary position inference. We evaluate our proposed algorithm on the KITTI data set and a large number of real scene experiments. Compared to other multi-target recognition methods, our algorithm provides much better accuracy. The experimental results show that the accuracy of road target algorithm can be effectively improved by learning historical visual semantics and target position information.