Towards Safer Roads: A Deep Learning Based Object Detection Technique for Vehicle Safety

In this work, we present a novel image processing and artificial intelligence-based driving assistance of vehicle system for vehicle safety and safer roads. This research introduces an innovative approach to advance object detection capabilities through the seamless integration of a custom -designed EfficientNet-based backbone into the Faster R-CNN architecture. The proposed model strategically leverages the EfficientNetB1 base model for efficient feature extraction, incorporating a sequence of Conv2D layers, batch normalization, and LeakyReLU activations to refine intricate features.After pre-training on the COCO dataset with a resnet50 backbone, our model obtains the best results on both COCO val 2017 (68.4AP) and PASCAL VOC 2012 (63.3AP). Compared to other models, our model significantly reduces its model size and pre-training data size while achieving better results.. Simultaneously, the algorithm exhibits efficacy in scenarios with considerably larger images, where objects encompass a more substantial percentage of the screen area. This dual capability underscores the algorithm's versatility, contributing to its broader applicability in object detection tasks across varying spatial contexts. Furthermore, the algorithm is specifically designed for the comprehensive detection of multiple objects within an image. Its efficacy is exemplified by consistent and successful identifications of desired objects in nearly all instances.