Semantically integrating laser and vision

Perception systems are now a reality with the deployment of camera-based object detection and lane departure systems in top-of-the-line vehicles. Although these systems are intended to aid the driver in hazardous situations, much has yet to be done in order to make them completely reliable in several circumstances. In this regard, multi-sensor architectures may bring complementarity and redundancy, making perception systems more robust. The aim of this work is to contribute with a novel perception system based on laser scanner and vision for object detection applied to urban scenarios. Although our system can be adapted to recognize any outdoor object, our method's proof of concept is performed on pedestrians, since they represent a challenge because of the variety of their poses, positions, forms, sizes, and colors. We hope that our work acts as a starting point for others to explore different multisensor fusion methods.