Sensor Fusion of Stereo Vision and Radar Systems for Vehicle Collision Avoidance

主動式車輛安全系統的開發在於提供使用者更完善的安全，宗旨為降低意外事故的發生機率，當系統發現有潛在危險的障礙物時，藉由圖形使用者介面 (graphical user interface, gui) 提醒使用者注意當前狀況，如果使用者沒有對系統的警示沒有反應的話則直接制動煞車系統反應，以避免意外事故的發生。本研究以立體視覺與雷達感測器建立一套中階感測器融合的車輛安全系統，立體視覺以兩顆攝影機組成，經由對應點匹配後可以獲得影像中的三維資訊，將深度資訊投射至上視圖進行障礙物偵測，可提供色彩與深度資訊，但深度資訊容易受外在因素影響；雷達感測器提供精確的距離和速度資訊，但無法辨識障礙物，為了獲得更完整的環境資訊，將兩種感測器進行感測器融合，以獲得更精確的資訊。兩種感測器的障礙物偵測完成後，依據感測器資訊的機率分布函數估算其可靠性，在障礙物偵測階段使用感測器融合方法結合感測器的資訊。障礙物追蹤使用卡爾曼濾波器，用於得知物體在時間序列中的相對關係並預測其運動模型。在行駛的過程中，撞擊預先警示系統會不間斷觀察是否有潛在的危險，若有障礙物可能會與車量進行碰撞，例如緊急煞車或側向突然出現，系統會提醒駕駛人注意當前狀況並判斷是否制動煞車系統防止與障礙物碰撞，同時啟動 A* 避障演算法進行路徑規劃，提供當前安全的路徑供車輛進行閃避。本研究的實驗場景為校園道路以及農場，經過實驗驗證，本系統能夠應用於校園道路的即時主動式車輛安全系統。Active vehicle safety system is aimed to enhance the safety of driving and reduce the accidents. When the dangerous situation was detected, the system would warn the user by graphical user interface to pay attention. If the user failed to response, the brake system would swing into action for preventing an accident. In this study, an intermediate-level obstacle-detection-based sensor-fusion vehicle safety system was proposed using stereo vision rig and radar. Stereo vision was composed of two cameras with color and depth information. The depth information estimated by stereo vision algorithm was projected onto the top-view. Obstacles were filtered out using blob method and 3D geometric constraints. High accuracy range information such as position and speed was provided by radar; however, there was no color information from radar. Stereo vision and radar were fused at detection-based sensor fusion in order to acquire more realistic information. After obstacle detection, Kalman filter was implemented for obstacle tracking. The motion model of obstacles was estimated between the data sequences. The pre-collision warning system was processed continuously to detect the obstacles with potential danger. When the obstacle performed sudden braking maneuver or popped out from side…etc, the obstacle avoidance system would warn the user and was prepared to take brake in preventing collision and plan a safe path for user to follow. A* algorithm was implemented to plan a safer path for user as a reference. After experimental validation, the active vehicle safety system was applied to real-time environment surveillance system in the school street