Co-evolutionary Multi-Sensor Placement in computer Vision

the complesities associated with the sensor placement for the purpose of extraction of 3D coordinates of the objects have made the automatic solution of this problem rather ompractical. These complexities are due to the fact that sensor positions, attitude and parameters are needed to be optimized in a network of imaging stations constructed around the object with a view to obtaining optimum accuracu for the final values of the extracted 3D coordinates. In this paper for the problem of Automatic Multi-Sensor Placement, we propose a solution based on evolutionary algorithms using simultaneous incorporation of CGA (co-evolutionary genetic algorithm) and LTFE (long-time fitness evaluation) techniques. The CGA technique for this particular problem seems to have provided faster convergence and reliablity factors as compared with the traditional GA approaches. Regarding the nature of our problem, the first population is considered to be the vision constraints and the second population is taken to be the appropriate objective functions of the cost and measurement accuracy. Since in this problem, similar to constaint satisfication problems (CSP), several cmplex constraints are to be satisfied, the LTFE technique is also utilized so that with a minimum computational cost, more complex constaints are satisfied with more attention. The preliminary experimints with the proposed approach indicate that there are not serious limitations as regards the type of sensors, constraints and objective functions in our approach. Besides, the approach shows high flesibility when environmental conditions are changed as it incorporates the AI concepts.

Co-evolutionary Multi-Sensor Placement in computer Vision Keywords:

Machin Vision , Close-Range Photogrammetry , Sensor Placement , Optimization , Co-evolution , CSP