Mobile Device Based Outdoor Navigation with On-Line Learning Neural Network: A Comparison with Convolutional Neural Network

Abstract

Outdoor navigation is challenging with its dynamic environments and huge appearance variances. Traditional autonomous navigation systems construct 3D driving scenes to recognize open and occupied voxels by using laser range scanners, which are not available on mobile devices. Existing image-based navigation methods, on the other hand, are costly in computation and thus cannot be deployed onto a mobile device. To overcome these difficulties, we present an on-line learning neural network for real-time outdoor navigation using only the computational resources available on a standard android mobile device (i.e. camera, GPS, and no cloud back-end). The network is trained to recognize the most relevant object in current navigation setting and make corresponding decisions (i.e. adjust direction, avoid obstacles, and follow GPS). The network is compared with state of the art image classifier, the Convolutional Neural Network, in various aspects (i.e. network size, number of updates, convergence speed and final performance). Comparisons show that our network requires a minimal number of updates and converges significantly faster to better performance. The network successfully navigated in regular long-duration testing in novel settings and blindfolded testing under sunny and cloudy weather conditions.