Convolutional Neural Network for Person and Car Detection using YOLO Framework

Abstract

In this paper we present a real-time person and car detection system suitable for use in Intelligent Car or Advanced Driver Assistance System (ADAS). The system is based on modified YOLO which uses 7 convolutional neural network layers. The grid cells of the system are varied to evaluate its effectiveness and ability in detecting small size persons and cars in real world images. The experimental results demonstrate that even with 7 convolutional layers, the system is able to provide good detection accuracy and real time operation. Although the mAP scores show reduction in accuracy, the visual qualitative evaluation using real world images indicate the 7 layer YOLO with 11x11 grid cells can correctly and easily detects small size persons and cars. This makes the reduced complexity YOLO a suitable candidate for use in ADAS which demands both relatively good detection accuracy and real time operation.