Investigation Of Efficient Depth Extraction Techniques For 3D Reconstruction Of Complex Environments

Stereo Vision based depth estimation has increasingly become an area of active research due to its enormous importance in applications such as industrial automation,Advanced Driver Assistance Systems,Robot Navigation and Scene Surveillance and Monitoring.Stereo Vision approach of extracting depth information is usually a complex process due to issues such as lighting,reflections,camera distortion and alignment,image complexity,pixel noise etc.This work investigates efficient techniques to estimate accurate depth(distance)information of 3D points from stereo pair images.Depth information is critical for automatic reconstruction and analysis of surrounding 3D environment.The work took special interest in finding the relation between image points and corresponding camera line of sight,relative pose of stereo cameras with different view angles,corresponding points matching and disparity map calculation which are key operations required for a high performance stereo vision system for depth(distance)estimation.The research further examined how edge detection methods can be uniquely combined with unreliable blocking matching algorithm to obtain a more reliable matching corresponding features(points)and an accurate disparity data.Additionally,a comparison study of a propose Edge-based Sum of Absolute Difference(SAD)technique,traditional SAD Block Matching(BM)algorithm and Semi-Global Matching(SGM)algorithm in terms of finding accurate corresponding points matching and fast computation of disparity map data were examined and discussed.The work is categorized into two sections comprising of hardware section and software section.The hardware shows details of the setup of experimental platform used to obtain input data.The software section reveals the operations performed on the obtained input data from the stereo cameras to extract accurate depth information.Implementation of the work was performed in MATLAB and the performances of the various investigated techniques were evaluated based on accuracy and efficiency(computation speed).The proposed edge-based technique was found to be efficient and also produced high accurate and reliable results.