Object Detection for Autonomous Vehicles Using Deep Learning Algorithm

Self-driving cars is recently gaining an increasing interest from the people across the globe. Over 33,000 Americans are killed in car accidents every year and lots of those accidents are often avoided by implementing the autonomous vehicle detection. Different ways are developed to manage and detect the road obstacles with the help of the techniques like machine learning and artificial intelligence. To resolve the issues associated with the existing vehicle detection like vehicle type recognition, low detection accuracy, and slow speed, many algorithms like the fast and faster region-based convolutional neural networks (RCNNs) are implemented but those were not supportive in real time because of the speed at which they compute and its two-step architecture with the faster RCNN, which is the enhanced version of RCNNs that runs at a speed of 7 frames per second. As it is observed that the CNN family has two steps (object detection and classification), which can reduce the response time in real time with good accuracy and high image resolution. So, the vehicle detection model like YOLOv2 and YOLOv3 is taken into consideration in this paper as they are very useful in real-time detection with a comparatively higher frame rate. As YOLO family of algorithms will mostly use the single step detection and classification. YOLO has an FPS rate of 45 which is pretty good in the real-time scenarios. We had an average of 90.4 using the taken algorithm for each image in this paper with a lower resolution image alone.