Performance of Inter-vehicular Distance Estimation: Pose from Orthography and Triangle Similarity

This paper reports the comparison of two methods of distance measuring using a digital camera and computing resources namely: Pose from Orthographic Projection and Triangle Similarity. Both methods incorporating a computer vision algorithm to properly functioned. Specifically, this report describes the utilization of such methods for vehicular application, for example, the inter-vehicular distance estimation, hence, to improve safety driving moreover to support the developing Advanced Driver-Assistance Systems ADAS. Each method constructed in an algorithm wrote in Python running in a Raspberry Pi computer equipped with a suitable camera. To process the incoming images, an OpenCV library was tasked to conduct a classification to distinguish vehicle-like features in an image frame from the rest of the universe. Haar cascade classifier was chosen to perform the image features classification. The algorithm then annotates and marks the classified features as a candidate for a vehicle-like object. The classifier was trained by a precompiled dataset. Both methods compared for the best performance on three distance measurement: 10, 15, and 20 meters. With experiment setup, the best distance to measure was 15 meters with small error to the ground truth.