This paper proposes an indoor positioning method based on Bluetooth Low Energy signals by Convolution Neural Network-Long Short-Term Memory (CNN-LSTM). The proposed method determines a receiver location based on distances from adjacent transmitters. The CNN-LSTM model estimates the distance from each transmitter using continuous signal strengths. To train and validate the model, the signal strengths are collected in several locations within various indoor environments. The positioning technique is adaptively selected based on the highest signal strength to avoid the interfering problem due to an excessively strong signal. If the signal strength exceeds a certain threshold, the location is determined using the proximity technique, which utilizes only the strongest signal instead of triangulation. In the experimental results, the proposed method demonstrated an average error of about 2.90 m, which is 34.2% better than a triangulation-based positioning method that does not utilize neural networks.

The demand for non-contact object volume estimation has been increasing across various industries. Although image-based volume estimation methods have been studied, they are generally limited to measuring simple shapes and are not suitable for irregularly shaped objects. We propose a volume estimation method for a single object using color and depth images through a deep learning network. For the color image, the object region is detected using a pre-trained segmentation model. For the depth image, the heights corresponding to pixels in the object region are calculated as differences between the pixel values and a background depth image that has captured a floor without any objects. Then, the heights are divided into multiple 2 × 2 pixel units. The element volumes corresponding to the 2 × 2 pixel units are estimated by a deep learning network which takes heights in a 2 × 2 pixel unit as inputs. This network estimates the corresponding volume to this unit through multiple fully connected layers. All estimated element volumes are summed to obtain the object volume. The simulation results show that the average volume estimation error is about 2.37%.

We propose a feature enhancement method for depth images that applies convolutions over residuals and parameters obtained from dominant plane. The proposed method can obtain the initial features of depth images that are less sensitive to surface orientation and more representative of intrinsic geometric properties. Specifically, the features are obtained through the plane-based convolution that performs operation on residuals with respect to the dominant plane within a local patch of the depth image. For each patch, a dominant plane is fitted to the corresponding depth pixel values using a least-squares method. Then, convolutional operations are performed on plane residuals computed between the original depth values and the corresponding depth values on the dominant plane. In addition, standard convolution is applied to the dominant plane parameters to capture local variations and spatial consistency of surface orientation. A plane-based convolution module incorporating these convolutions is attached to the initial layer of the existing feature extractor in parallel to supplementarily obtain surface geometric features. Experiment results demonstrate that the proposed method consistently achieves performance gains on both segmentation and classification tasks.