Deep learning based pedestrian prediction for mobile robot navigation

• Main Author: Chen, Weipeng
• Other Authors: Teoh Eam Khwang
• Format: Final Year Project
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Electrical and electronic engineering
• Online Access: http://hdl.handle.net/10356/77416
• Institution: Nanyang Technological University

What is artificial intelligence? In simple terms, it gives computer ability to think and a decision like the human by learning the pattern through an artificial neural network. In the modern day, thousands of models have developed for object detection, speech recognition, and classification. However, for mobile robots to work autonomously or cooperate with humans in daily workspaces. They should not only detect the object and also react base on the detected object. The surrounding people may influence human navigation behavior. There are several approaches present over the year to understand people behavior, motion, and posture such as the recurrent neural network. It can study long sequence data over time and use the data to predict the next move of a people. This project targeted towards developing an algorithm and model to detect the pedestrian and predict the next move. The project will be divided into three parts. First is detection, by using MobileNet Single Shot Multi-Boxes detector (SSD) to detect the pedestrian and store the past pedestrian coordinates. MobileNet SSD is the network which can compute high accurate results with the limited number of computation power. Second is for id tracking. An algorithm is essential to obtain the correction prediction results for the correct pedestrian. Intersection over Union (IoU) is implemented for id tracking. By just finding the largest area of overlapped to map the detection object in the next frame. Last, to use Long Short-Term Memory (LSTM) and Kalman filter to predict the future trajectory of the pedestrian