Object detection and recognition by using sensor fusion

This paper studies a sensor fusion method focusing on a collision avoidance system, capable of high accurate object detection by blending a camera and 2D LIDAR with the aid of pixel analysis. Pixel analysis is performed via single camera under a concern that the free space in front of vehicles is limited by objects on almost vertical surfaces. The considered problem is defined as that how to efficiently analyze a large number of disparity data on image for the object identification. A solution comprising of vision, rangefinder, and inertial sensors is thus proposed. To reduce its computational costs in real-world applications, we fulfill the object detection and locating process in the laser space while completing the object classification in the vision space in the environments of expressway with a speed range [60, 70] (kph) and urban roads with a speed range [10, 30] (kph). Furthermore, verification results show the superiority of the constructed mechanism in the moving/stationary object recognition with the data fusion.