Cow Crossing Road Detection Using YOLO V8 and SSD

Abstract

The increase in car accidents involving stray cows on Oman's roadways has become a pressing concern, resulting in a substantial rise in fatalities and injuries. This study aims to compare the accuracy of two prominent object detection algorithms, YOLO v8 and SSD, in detecting cows crossing roads under various weather conditions in Oman's Dhofar region to improve road safety and provide valuable insights for road monitoring authorities. The methodology involves the creation of a dataset comprising 15 videos capturing cow crossings under different weather conditions. The study evaluates detection accuracy at different confidence levels (25%, 50%, 75%, and 90%) and collects results in terms of true positives (TP), false positives (FP), and false negatives (FN). The experimental results reveal that YOLO v8 consistently outperforms SSD in all weather conditions. In clear daytime weather, YOLO v8 achieves an average precision of 96%, while SSD achieves 59%. In foggy conditions, YOLO v8 maintains a precision of 64% compared to SSD's 18%. In nighttime scenarios, YOLO v8 excels with a precision of 94%, while SSD lags at 5%. Overall, YOLO v8 attains an impressive mean average precision of 84.67% across all conditions, while SSD achieves 27.33%. These findings underscore the significance of selecting the right object detection model for specific weather conditions to enhance road safety.