| Abstract |
To address air pollution and reduce carbon emissions, vehicle electrification has become the most essential and feasible solution. Consequently, major economies such as the EU, Japan, and the United States have implemented policies to promote vehicle electrification. Within this trend, autonomous driving plays an indispensable role, with obstacle avoidance being a top priority to ensure driving safety. This thesis employs a dual-camera setup as the core of the stereoscopic vision system for vehicle obstacle avoidance, aiming at addressing camera-based obstacle detection solutions available on the market. To compensate for the blind spots of dual cameras, ultrasonic sensors are added. By utilizing the disparity between the dual cameras and employing the SGBM (Semi-Global Block Matching) algorithm, the system calculates depth, determining the distance between the vehicle and obstacles to facilitate effective obstacle avoidance. To ensure that visual recognition operates independently of network connectivity or a computer for processing, this study uses the Raspberry Pi 4B as the processor for the dual-camera system, along with an Arduino-controlled obstacle-avoidance vehicle commercially available. Testing was conducted in a custom-designed field to verify the effectiveness of the obstacle avoidance vehicle designed in this thesis. |
