Wildlife-vehicle conflict poses injury/mortality risks to both drivers and wildlife. Roadside Animal Detection Systems (RADS) are used to detect animals near or on roads and warn drivers to reduce speed and increase attentiveness, but many contemporary technologies are in the early stages of development and require more testing. Animal detection systems include RADAR, infrared imaging, and advanced Light Detection and Range (LiDAR) sensors, buried cables, and break-the-beam systems. Although optical cameras/video systems provide validation information for other sensor types, they are sensors in their own right. We describe the Rapid Animal Detection and Identification System (RADIS), a new addition to RADS, to detect wild or domestic animals approaching animal-vehicle conflict zones (including roads and highways) and alert drivers through intelligent road signs. We developed and tested RADIS in a real-world setting, in collaboration with Nevada DOT and the University of Nevada Reno. The study area is home to wild horses that roam freely through a rural industrial area, where collisions with horses are common. RADIS is built around a low-power, edge computing module with a graphic processing unit (GPU). The system combines an optical video camera with an artificial-intelligence-based computer vision model (YOLO-v5) that can enable real-time detection and classification of moving objects including animals. Detecting animals in moving imagery presents an opportunity to map an animal’s trajectory and to determine if it is a driver-safety concern. The system uses this early intelligence to trigger a radio frequency signal and activate RF enabled road signs. The AI model was pre-trained with 200,000 original images from the COCO dataset across 80 categories in addition to 2,000 images of wild horses obtained from the study area — these images were not used to test the model. We remotely communicated with the deployed instance of RADIS, allowing real-time monitoring of the system and updating of the classification model, as needed. We tested the model using still images and video streams from the study area. The mean average precision score (mAP@0.5) was 0.907 across cars, trucks and horses on the real world data which is higher than other published systems. We achieved a recall rate of 97% and classification accuracy of 92% for individual still images. For live-streaming video, the detection recall and classification accuracy were both 99%, making it one of the most accurate RADS reported in the literature. The combination of rapid and accurate animal classification with remotely-adjustable hardware and software systems provides a highly-controllable and adjustable animal vehicle conflict detection system in real-world settings. This paired with active displays for driver warning create a new modality of roadkill prevention that could be as effective as existing passive solutions such as fences and crossing structures.