Obstacle Detection Using a Computer Vision System for Assisting Blind People

Diplomska naloga predstavlja teoretični in praktični vidik razvoja sistema, ki deluje s pomočjo metod računalniškega vida, predvsem detektiranja objektov, in pomaga slepim uporabnikom tako, da jim rezultate detekcije ovir sporoča kot zvočne signale. Najprej sistem dobi podatke iz zaporedja oblakov 3D točk, dobljenih s stereo kamero, kar omogoča upoštevanje dimenzije globine. Podatke oblaka 3D točk sistem potem uporabi kot vhode v algoritem za detekcijo. Sistem detektira objekte, ki so pomembni za uporabnika, torej pešce, vozila, robnike in stopnice ter druge ovire na poti. Rezultate sistem pretvori v zvočne signale, ki z različnimi zvoki jasno in dovolj hitro sporočijo uporabniku, kako blizu so ovire. Končni rezultat izboljša izkušnjo slepih ljudi ob hoji v naselju tako, da v realnem času uporabniku predvaja ustrezna zvočna opozorila in zmanjša možnost, da bi se uporabnik zaletel ali spotaknil ob oviro, poleg tega pa opozorila dajejo uporabniku več informaciji o njegovi okolici.The diploma thesis consists of a theoretical as well as a practical aspect of development of a device, which is based on computer vision methods, especially object detection, and is intended to help blind users by transforming the results of obstacle detection data into audio data, which is then played back to the users. The first step is to obtain a sequence of 3D point clouds from a stereo camera, as the camera allows the recording of depth. Next, the point cloud data is used as input data for object detection algorithms. It is imperative for the system to detect the objects of importance to the user, such as pedestrians, vehicles, curbs, stairs and other obstacles on the user\u27s path. The results are transformed into an audio signal, which clearly and in real-time notifies the user about the distance of approaching objects via various sounds. The final result enhances the blind user\u27s experience when walking in a city environment by playing warning sounds in real-time and thereby reducing the danger of accidents, such as colliding with and losing balance over obstacles. The system also gives more contextual information of the user\u27s surroundings