An accurate volume estimation on single view object images by deep learning based depth map analysis and 3D reconstruction

The volume estimation of a rigid object from a single view object image is the important need in numerous automated vision based systems. The volume estimation on multiple view images are simple to estimate. But volume estimation on a single view object image is a difficult process and has significant importance in volume estimation. This work presents effective object volume estimation in both regular and irregular single view object images. Initially, the single view input images are pre-processed with Mean-median filtering. Afterwards, edge features are extracted by utilizing the Gaussian edge based laplacian operator and key points are extracted using the Scale invariant feature transform (SIFT) feature. The extracted features are considered for the shape analysis of the objects. Subsequently, VGG-ResNet framework is utilized for depth analysis based on the extracted features. The point clouds generation for the volume estimation is attained through the extracted features. Finally, the volume estimation on single view object is effectively attained through the hybrid 3 dimensional U-Net and graph neural network (Hybrid 3DU-GNet). This framework provides the 3D geometric creation for the accurate volume estimation. This provides the significant improvement on volume estimation. The presented methodology effectively estimates the volume on both regular and irregular single view object images. The presented approach is implemented in the working platform of MATLAB. The experimental results of the presented work is analysed with the different existing approaches and proved the significant improvement in performance metrics. The performance metrics are Accuracy (98.59%), precision (98.21%), recall (97.09%), computational time (3.2 seconds), R-squared (98.2%), (Mean absolute percentage error) MAPE (6.1%), and (Root mean squared error) RMSE (0.93).