Título: Combining Ultrasonic Signals And Multi-Colored Images To Perform Object Tracking And Recognition In Low-Cost Robotic Platforms
Autores: Menezes, Danilo H. F.; Mendonca, Thiago D.; Neto, Wolney M.; Macedo, Hendrik T.; Matos, Leonardo N.
Resumo: Robotics research focuses on a broad range of interdisciplinary aspects. Regarding robot-driven application development, different purposes and complexities may be considered. State-of-the-art platforms are usually adopted for developing non-trivial tasks, however, their high costs occasionally inhibit robotics application for education and research purposes. Some tasks, such as pattern recognition, are usually designed without considering low-cost requirements. In order to fully explore the capabilities of low-cost platforms, this article presents an empirical analysis of object tracking and recognition accuracy, non-trivial and essential tasks for Robotics. This task is performed by an autonomous robot equipped with camera and ultrasonic sensor. Three experimental scenarios are defined for further observation and comparison. Object tracking and representation acquirement are achieved in these scenarios only by camera, only by ultrasonic sensor and by combining both, respectively. 10-fold cross validation has been carried on a MLP neural network with different learning rates. Image-based recognition got an average f-measure above 0.9 and an area under ROC curve above 0.95, which proved to be better than ultrasonic-based recognition with a f-measure around 0.8 and area under ROC curve around 0.85. Experiments have also validated low-cost platforms adoption for object tracking and recognition.
Palavras-chave: Low Cost Robotic Platforms; Object tracking; Object Recognition; Multi-layer Perceptron
Código DOI: 10.21528/CBIC2011-22.4
Artigo em pdf: st_22.4.pdf
Arquivo BibTex: st_22.4.bib