— Conventional object detection techniques are based on shallow, trainable structures and handmade characteristics. Building intricate ensembles that integrate several low-level image features with high-level information from scene classifiers and object detector is able to stabilize their accuracy. More potent tools that can learn high-level, semantic and deeper features are being introduced as a result of the rapid advancement of deep learning to solve issues with conventional architectures. The paper provides an innovative method for identifying multiple objects using a publicly accessible image dataset. The dataset includes farexposed views, such as those taken in bright sunshine, and the intrinsic characteristics are not particularly trustworthy for training, making it challenging to construct detection in it. As a result, we suggest adopting a YOLOv3 model that has already been pretrained for training, increasing its basic accuracy using various regularization and loss techniques. We also suggest a solution for numerous object detection problems, particularly in real time, based on our findings. Many applications, including autonomous vehicles and sophisticated systems for driver assistance, are targeted by these strategies.