PMFHNet: parallel Maxout feed harmonic network for road anomaly object detection in ADAS under low light conditions

This research proposes a Parallel Maxout Feed Harmonic Net (PMFHNet) for road anomaly object detection in an Advanced Driving Assistance System (ADAS) under low-light conditions. Initially, the input video is used to extract the frames. Following this, Contrast Limited Adaptive Histogram Equalization (CLAHE) is applied to enhance low-light images. Further, object segmentation is conducted employing RefineNet. After that, feature extraction is accomplished. Thereafter, road anomaly object detection is performed using PMFHNet. Finally, control action is done based on navigation commands to alert the driver about potential road hazards, obstacles, or irregularities detected on the road. Moreover, the devised scheme accomplished normalised Mean Squared Error (MSE), normalised Root Mean Squared Error (RMSE), normalised Mean Absolute Error (MAE), and R-squared (R2) as 0.089, 0.299, 0.103, and 0.091, respectively.