Investigation of Animal Detection for Traffic Accident Mitigation, 10-9193Printer Friendly Version
Inclusive Dates: 06/14/00 - 10/14/00
Background - Animal-vehicle crashes are a significant problem on the nation's roadways, accounting for approximately 726,000 vehicle collisions annually. Hundreds of millions of dollars are spent each year for crash-prevention measures and post-crash repair and cleanup. Intelligent Transportation Systems (ITS) initiatives have focused on advanced technology solutions to transportation system problems. The limited technologies available for roadside animal detection are short-range, limited in coverage, or provide low background contrast. This quick-look project investigated the feasibility of an animal-detection and driver-warning system using infrared imaging and automated image processing. The subject approach was promoted as the most sensitive and cost-effective technology to a multistate, pooled-fund project seeking deployment and evaluation of animal-detection systems.
Approach - Longwave infrared (LWIR) imaging and thermal image processing are proposed as the most robust solution for animal detection along roadways. The technology offers the ability to 'see' objects as an image formed by infrared emissions related to the surface temperature of a subject. Significant image contrast between a person or animal and a complex background scene comprised of nonheated surfaces is obtained. Temporal image processing is used to construct an image of the background scene and detect objects that move within the scene. Range calibration allows object sizes to be estimated, thus permitting size-based filtering. Speed and shape measurements can be used to distinguish between vehicles and animals. Long-range detection requires the use of long focal lengths, which are very expensive for refractive optics. The proposed solution uses reflective optics. A combination of short and long focal length optics provides efficient coverage of foreground and distant regions of a monitored roadway. The position(s) of animals detected by the image-processing system would be relayed by wireless link to variable message signs to alert drivers.
Accomplishments - Critical parameters impacting the feasibility of the animal-detection system were investigated. A design configuration tool was developed that supported the selection of optical design parameters, given a monitored zone definition. Long-range optical systems were assembled from commercial, off-the-shelf components and mated to a LWIR camera. Infrared video of deer was recorded using several optical configurations, providing a range of spatial resolutions. Video sequences also captured a variety of group sizes and activity levels, including sedentary, grazing, and active deer movement near and across roadways. Animal-detection and discrimination image-processing algorithms were developed and used to process the recorded video sequences. Algorithmic approaches included: temporal processing, motion detection, target persistence, target signature analysis (shape, size, contrast, intensity distribution), and range compensation.