Sensor fusion of laser and vision in active pedestrian detection

• Autor(a) principal: Azevedo, Rui Filipe Cabral de
• Data de Publicação: 2014
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10773/14414
• topic: Engenharia mecânica; Peões - Detecção remota; Tráfego urbano

This work explores a technique of sensor fusion that aims to equip vehicles with pedestrian fast detection mechanisms in exterior environments. This method restricts image areas of search based on indicators obtained by another sensor (LIDAR). This technique is based on the idea that when having a registration among the involved sensors, one "fast" sensor, but inaccurate, that can indicate regions where potential pedestrian are located on the image, and another sensor, "slower" but more robust that is used to confirm detection more accurately. So, an algorithm was created to merge two algorithms, a LIDAR-based tracking and a vision-based detection algorithm; The LIDAR indicates the precise location and scale of the potential pedestrian on the image, and crop the image relative to the potential pedestrian, being processed afterwards by one pedestrian detection algorithm to validate the classification. The method is tested in two different cases and the results confirm their validity.