Autonomous vehicles have a necessity of object detection algorithm for safe driving and for precise detection of objects with a definite accuracy in a time bound scenario. This work puts forward an object detection model called YOLOv7-Ghost which reduces the number of floating-point operations by 49.21% and the resultant model size by 140.35% in order to increase the inference speed. Unlike YOLOv4-5D here ghost convolution with YOLOv7 is proposed. Ghost convolution and 2D convolution differ by the method in which they are convolved. Ghost convolution splits the given image into two parts for which the respective kernels are applied unlike 2D convolution where the complete image is used in kernel for convolution. Hence Ghost convolution is theoretically considered to be faster than 2D convolution methods. The floating-point operations required for ghost convolution are lesser in number as compared to 2D convolution instead making use of linear operations to generate redundant feature maps.