| Summary: | Real-time position sensing has a wide range of applications in motion control systems, parts inspection
and general metrology. Vision-based position sensing systems have significant advantages over other
sensing methods, including large measurement volume, non-contact sensing, and simultaneous mea-
surement in multiple degrees-of-freedom (DOF). Existing vision-based position sensing solutions are
limited by low sampling frequency and low position accuracy. This thesis presents the theory, design,
implementation and calibration of a new high-speed stereo-vision camera system for real-time position
sensing based on CMOS image sensors.
By reading small regions around each target image rather than the full frame data of the sensor,
the frame rate and image processing speed are vastly increased. A high speed camera interface is
designed based on Camera Link technology, which allows a maximum continuous data throughput
of 2.3Gbps. In addition, this stereo-vision system also includes fixed pattern noise (FPN) correction,
threshold processing, and sub-pixel target position interpolation.
In order to achieve high position accuracy, this system is calibrated to determine its model parame-
ters. The primary error sources in this system include target image noise, mechanical installation error
and lens distortion. The image sensor is characterized, and its FPN data is extracted, by experiment.
The mechanical installation error and lens distortion parameters are identified through camera cali-
bration. The proposed camera calibration method uses the 3D position reconstruction error as its cost
function in the iterative optimization. The optimization of linear and nonlinear parameters is decoupled.
By these means, better estimation of model parameters is achieved. To verify the performance of the
proposed calibration method, it is compared with a traditional single camera calibration method in sim-
ulation and experiment. The results show that the proposed calibration method gives better parameter
estimation than the traditional single camera calibration method.
The experimental results indicate that the prototype system is capable of measuring 8 targets in 3-
DOF at a sampling frequency of 8kHz. Comparison with a coordinate measurement machine (CMM)
shows that the prototype system achieves a 3D position accuracy of 18μm (RMS) over a range of
400mm by 400mm by 15mm, with a resolution of 2μm. === Applied Science, Faculty of === Mechanical Engineering, Department of === Graduate
|
|---|
