A Visual Approach Towards Wildlife Surveillance in Malaysia

Road development has caused increased wildlife-vehicle collision (WCV) in Malaysia. WCV does not only pose a high risk to road safety but also become a major threat to the survival of precious wildlife in the country. Every year, hundreds of endangered species such as the Malayan Tapir are killed on the road. The wild animals set foot on man areas because their movement corridor has been blocked by roads or highways. Besides, the wild animals are forced to come out from the forests to hunt for food due to excessive deforestation that destroy their natural habitats. In this study, wildlife surveillance system based in visual input analysis is presented. Footages from CCTV cameras placed near the roadways are used to perform surveillance monitoring. A data-driven object detection method based on deep transfer learning is used to automatically monitor and detect the occurrence of wildlife in the scene under observation. Once the sign of wildlife is detected, a warning will be sent to the corresponding conservation or road infrastructure departments to take immediate action. Empirical results show that the proposed method achieve a mean average precision (mAP50) of 95.6% and an average recall of 94% in detecting wildlife. This suggests that proposed method offers a promising solution to help to reduce the cases of WCV in the country through visual surveillance.