Project

Stereo vision for real-time obstacle detection

Project (M.S., Computer Engineering)--California State University, Sacramento, 2019.

Real-time obstacle detection is one of the critical problems in autonomous vehicles. Due to the variety of environments that autonomous vehicles encounter, a more robust method is needed to address detecting objects in a vehicle’s path. One solution to this problem is utilizing a stereo camera system to detect an obstacle in the path of a vehicle. A stereo camera system provides a versatile solution to the problem of object detection, because it offers greater field of vision and range of detection than other sensors. For stereo cameras to find the distance between objects and the cameras, it will need to solve the correspondence problem. The correspondence problem refers to matching points from one image to another. Solving this problem allows for the 3-dimensional reconstruction of a scene, which can provide the distance an object is from the vehicle. For this project, an application was developed that generated a depth map that provided necessary information in a reasonable amount of time for a system to follow a collision avoidance protocol. The first step was to obtain the parameters of the stereo camera system. A depth map was then generated using the block matching algorithm to solve the correspondence problem. Finally, analysis of the depth map with different parameters for the block matching algorithm was performed to evaluate the best configuration for the system.

Real-time obstacle detection is one of the critical problems in autonomous vehicles. Due to the variety of environments that autonomous vehicles encounter, a more robust method is needed to address detecting objects in a vehicle’s path. One solution to this problem is utilizing a stereo camera system to detect an obstacle in the path of a vehicle. A stereo camera system provides a versatile solution to the problem of object detection, because it offers greater field of vision and range of detection than other sensors. For stereo cameras to find the distance between objects and the cameras, it will need to solve the correspondence problem. The correspondence problem refers to matching points from one image to another. Solving this problem allows for the 3-dimensional reconstruction of a scene, which can provide the distance an object is from the vehicle. For this project, an application was developed that generated a depth map that provided necessary information in a reasonable amount of time for a system to follow a collision avoidance protocol. The first step was to obtain the parameters of the stereo camera system. A depth map was then generated using the block matching algorithm to solve the correspondence problem. Finally, analysis of the depth map with different parameters for the block matching algorithm was performed to evaluate the best configuration for the system.

Relationships

Items