Real-Time Multi-objects Detection Using YOLOv7 for Advanced Driving Assistant Systems

Accurate and efficient multi-object localization and categorization is one of the key needs for applications of robotic vision, intelligent military surveillance systems, security, and ADAS. It is a significant and complex issue with computer vision that has received a lot of attention. Due to the development of self-driving cars, smart video monitoring, face recognition, and several people tracking services, there is a great demand for quick and precise object detection systems. These methods locate each object by drawing bounding box around it in addition to identify and classify every object in an image or video frame. One of the basic needs for driverless vehicles and many modern driving aid technologies is the ability to recognize and interpret all stationary and moving objects surrounding a vehicle under varying driving and weather circumstances. Convolutional neural network (CNN) technology can provide safety in modern vehicles. This paper analyzes the recent deep learning-based object detection methods, challenges and presents an improved general framework for real-time multi-object detections in ADAS-based on YLOv7 and CNN. We compared the YOLOv7 approach with earlier versions in the YOLO family, and it was found that YOLOv7 outperforms all earlier real-time object detectors to obtain a reliable speed between 5 and 160 frames per second and improved accuracy.