Various Obstacles Detection Systems using Single Shot Multi-Box Detector (SSD) for Autonomous-Driving Vehicles

Abstract

One of the most important features of an autonomous vehicle is obstacle detection. The vehicle should be able to precisely and timely detect the presence of an obstacle to avoid a collision. This study aims to design and build an obstacle detection system to detect four types of obstacles (cars, motorcycles, people, and potholes) using the Single Shot Multi-box Detector (SSD) method and mobilenet v2 architecture. The input is video data extracted into frames and taken using a dash camera installed in the car. The dataset contains 720 images for each obstacle object. The training parameters are num_steps=20000 and batch_size=16. The result shows that the SSD method can be implemented properly for detecting and classifying obstacles in real-time. From the testing stage, the system obtains accuracy of 93.88%, 97.22%, 95.83%, and 94.44% at speeds of 10 km/h, 20 km/h hour, 30 km/hour, and 40 km/hour, respectively.

Access to Document

Other files and links

• Link to publication in Scopus
• Open Access Version Available

Fingerprint