Object detection and classification is an indispensable technology for interacting with the environment. In this way, the main problem is finding a robust algorithm to decrease time and computation. However, finding the object in all partsof the image, is time consuming with increment on computation and the probability of false detection. In addition, for fast moving agents, time and computational capacities on detection and classification are limited. The substitution one of the human's power with a machine, require to be mentioned that any ability of humans' uses prior knowledge to learn how can detect and classify the huge amount of information. For instance, although with new technology, machines reach the humans' visual capacity in accuracy and precision, finding the less computational and time-consuming algorithm would need more efforts by humans both in hardware and software.Utilising of deep learning, especially, Convolutional Neural Network (CNN) for automated object detection systems, has robust performance but required also high inference speed and memory. In the presented master thesis, an algorithm has been proposed that recognizes Path Precision Approach Indicator (PAPI) lights by image processing using camera geometry that uses the precise of aircraft and lights poses. The use of light position allows us to determine of a Region-Of-Interest (ROI) in an image to reduce the computational cost and false detection. In addition, by using the proper classic algorithm followed by deep learning precision of the prediction will be surveyed on class of lights. In this thesis valuation and selection of many algorithms based on the papers or experiments that have been considered on traffic lights detection and classification in autonomous vehicle driving.