Low-Cost Computer Vision Based Real-Time 3D Localization of Object for Robotic Applications

In this paper, we propose low-cost robust 3D vision for robotics. This system plays an important role for robot to automatically detect and locate 3D coordinates of an object in space at real-time. In manufacturing this is very useful to automate repeated object pick and place operations. Initially we improvised 2D object detection. Later, we estimated object depth using Stereo Vision techniques. Overall we combine stereo vision techniques, object detection techniques, geometric calibration techniques, inverse kinematics into one model avoiding redundant computations and greatly simplifying underlying equations to achieve real-time speed and good-accuracy. We reduce redundant computations by replacing standard stereo matching with our own object detection model. We accurately localize detected objects by performing calibrations, improvising real world 2D coordinate equations and improvising depth estimation equation. This single framework is practically implemented and developed using open source platforms. From experimental results, we have observed the proposed model shows superior performance (Above 99% object detection rates, Accuracy upto 8mm, Only 23% CPU utilization, Only ∼155MB of RAM consumption) ∼on an average and is cost-effective (Under $20). This low-cost model can be useful for industrial purpose, small businesses, STEM education or for training and skill development.