We found a match
Your institution may have rights to this item. Sign in to continue.
- Title
Learning spatial-temporal representation for smoke vehicle detection.
- Authors
Cao, Yichao; Lu, Xiaobo
- Abstract
Vehicle exhaust emissions are notorious for being unhealthy both for humans and the environment. Smoke vehicle, emitting excess levels of visible black smoke, is representative heavy pollution vehicle. It is a challenging task to recognize smoke vehicles from traffic surveillance due to the large variance of smoke color, texture, and interference. To solve this problem, this paper proposes smoke vehicle detection methods by learning spatial-temporal representation from image sequences. Firstly, motion detection algorithm is used to obtain the rear section of vehicle that need to be identified. Then, space information of each suspected frame is captured by Inception V3 convolutional neural network (CNN), and a temporal Multi-Layer Perception (MLP) or Long Short Term Memory network (LSTM) is used to effectively train the smoke vehicle model. The first method attempts to jointly model spatial-temporal clues for smoke vehicle detection in the video by fully-connected layers. The second method aims to learn temporal dependencies between video frames with LSTM. LSTM networks could combine image information in video over a longer period of time. Experimental results on our dataset have shown that the LSTM-based model achieve a highly accuracy of 97.6875%, and there is 9.25% improvement over the single frame model.
- Subjects
SMOKE; LONG-term memory; SHORT-term memory; TRAFFIC monitoring; VEHICLE models
- Publication
Multimedia Tools & Applications, 2019, Vol 78, Issue 19, p27871
- ISSN
1380-7501
- Publication type
Article
- DOI
10.1007/s11042-019-07926-1