Comparison of Key Frames of Video Sequences Using the Izhikevich Spiking Neuron

One of the important problems in the technical vision of robots and in the tasks of local navigation and global mapping is recognition of robot location, both locally and in a global map. This problem is solved by standard approaches, such as flat possibility maps, three-dimensional maps with LIDAR sensors and stereo pairs, but all them are not always easy and convenient. A method combining the latest approaches to image segmentation without training, image processing tools, and a modified Izhikevich spiking neuron is proposed. The algorithm compares frames of the test video sequence with key frames of the train video sequence shot in the same area as part of initialization and configuration process of the system. The work was carried out under an assumption that all the main areas and positions where camera can be located in work time were captured and processed at the training stage. Object classification mechanism description is deliberately omitted in this work. Final result of the work is an algorithm that matches any keyframe from the train sequence with the best matching frame from the test sequence. By generalization to an arbitrarily large number of key frames in the observed area, an algorithm for global localization in an arbitrarily large area can be obtained.