Deep learning approach on detection and localization using stereo vision camera

Mohd Rahim, Farah Liyana (2021) Deep learning approach on detection and localization using stereo vision camera. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Deep learning approach on detection and localization using stereo vision camera.pdf - Submitted Version Download (2MB)

Abstract

This research presents the detection and classification of chilli-padi based on its maturity and position using the Deep Learning (DL) technique and stereo vision camera. This research is beneficial for farmers to aid them in chilli padi picking. This research is also significant for robotic vision picking activity to pick the chillis' automatically. The aim could be achieved using the RGB and depth map images as the input of this project. For this project, about 35 chilli-padi two-dimensional photos were captured using a stereo camera from multiple angles. These projects were divided into 70-30 percent training, validation, and testing ratios. The method that we used in this project was using Faster Region-based CNN (Faster R-CNN) as a deep learning model for train the input and output layers. The outcome was determined by the deep neural network model's performance measured by mean average precision (MAP). The average precision score for RGB, the depth map, and the RGB + depth map is 0.2, 0.01, 0.6. The best value of AP was the RGB + depth map image. The range of average precision is from 0 to 1, closes with 1 being the best result.

Actions (login required)

View Item