基於即時影像辨識模型之機車盲點偵測系統

盲點是汽車、機車等交通工具行駛於道路上時，駕駛無法透過檢測盲點工具注意到的區域，甚至是會被自己車體擋住導致看不到的區域。不論是小型汽車、大型汽車或者是機車，都會有盲點的存在，並且不同車種的盲點範圍也會不一樣。其中，機車騎乘在道路上時因為騎乘視角的關係，導致機車的左右兩側以及後方皆有盲點增加駕駛行駛在道路上的危險性，並且汽車的盲點也會導致機車的不安全。為了解決機車盲點的危險性問題，本篇論文提出了一套透過即時影像辨識結合樹莓派的系統，運作流程為安裝於機車兩側的攝影鏡頭會持續偵測盲點的畫面，接著即時回傳影像至樹莓派上透過盲點辨識模型辨識出盲點是否有車輛，最後透過安裝於機車左右兩側的 LED 燈號模組警示駕駛盲點是否有其他車輛來幫助駕駛判斷盲點的危險性，不用擔心變換車道時與後方車輛的安全距離不足導致擦撞等問題，可以增加行車時的安全性。本系統具有安裝過程簡易、警示方式直覺、硬體成本低、檢測即時等優點，並且本系統測試時有分類為白天及晚上兩種場景來檢驗系統的準確率，結果顯示，不管白天或晚上準確率皆有達到 90%以上。

關鍵字

盲點偵測輔助系統；行車紀錄器； Haar-like特徵；機車

並列摘要

The blind spot is an area that a driver cannot see, even with a blind spot detection tool, oncoming cars or motorcycles, or is an area that is blocked by the body of the car.Whether it is a small vehicle, a large vehicle or a motorcycle, all have blind spots, and the blind spot area varies from vehicles to vehicles. The blind spot of a motorcycle on the road is due to the angle vision of the motorcycle, which leads to blind spots on the left and right sides of the motorcycle as well as the rear, increasing the danger of driving on the road. In order to solve the problem of blind spots, this paper proposed a system that combined Raspberry Pi with real-time image recognition. The operation process is that cameras installed on both sides of the motorcycle will continuously detect the blind spots, and then send the images back to Raspberry Pi to identify whether there is a vehicle in the blind spot through the blind spot recognition system. The LED light modules installed on the left and right sides of the locomotive warn whether there are other vehicles in the blind spot to help drivers determine the danger of the blind spot, so that they do not have to worry about changing lanes with the vehicle behind the safety distance and whether if it is not enough to cause a collision. This increases safety driving. The system has the advantages of easy installation, intuitive warnings, low hardware costs, and instant detection. Moreover, the system was tested in two scenarios, aytime and nighttime, to check the accuracy of the system.