Open AccessVehicle Collision Detection During Night Time
Author(s) -
Sudeep R K*,
J. Murali Krishnan,
Pradeep N.R,
Sajith Vijayaraghavan
Publication year - 2019
Publication title -
international journal of recent technology and engineering
Language(s) - English
Resource type - Journals
ISSN - 2277-3878
DOI - 10.35940/ijrte.c6501.098319
Subject(s) - artificial intelligence , computer science , support vector machine , feature (linguistics) , frame (networking) , histogram , histogram of oriented gradients , computer vision , pattern recognition (psychology) , local binary patterns , object detection , image (mathematics) , telecommunications , philosophy , linguistics
”Vehicle collision detection during night time” is an attempt to find an improved solution for vehicle detection problem, during night time. Night time vehicle detection is a challenging area since, the contrast between back ground and foreground is very less due to non uniform lighting, reflections, multiple light sources etc. This algorithm is a blend of night time image enhancement technique and fusion technique for fusing multiple features based on their weight. Image enhancement is an important aspect of the algorithm as it helps in improving the contrast between foreground and back ground as well as improving the lighting, so that it becomes easier to distinguish vehicle and other illuminated objects such as street lights. For improved reliability of the algorithm, set of complementary features - LBP (local binary pattern), HOG (Histogram of Oriented Gradients) and a CNN based feature - are used in the algorithm and are fused together using weighted fusion technique. For classification purpose, individual SVMs are trained for each feature and another SVM is trained for final rank based vehicle detection. Accurate region of interest in computed during detection phase, and is used for vehicle detection by trained SVM classifiers. Tail light detection is used for efficient region proposals. The algorithm is improved by leveraging the region proposals and high accuracy detection of Faster RCNN [10]. The algorithm detects multiple vehicles in single image frame and at possible multiple locations and distances within the frame.