创新背景

随着摄像机陷阱数据库的广泛部署和野外研究中长期视频档案的使用，利用人类研究人员捕捉动物行为的数量正变得越来越难以规模化。

创新过程

新的计算机模型使用了来自西非的两个野生黑猩猩种群的视频进行训练，这些黑猩猩来自几内亚的博苏和几内亚比绍的坎坦海斯国家公园，以捕捉到几个行为:敲坚果、吃东西和击鼓。该工具是首个使用音频和视频自动识别野生灵长类动物行为的工具，它建立在之前的工作基础上，之前的工作开发了一种使用人脸识别跟踪和识别个体的自动方法。

牛津大学工程系的研究员兼博士生马克斯·贝恩(Max Bain)表示，新的模型可以应用于黑猩猩在自然栖息地的数千小时的视频记录，这些视频记录来自相机陷阱或档案镜头。研究使用深度学习的方法，这种方法可以同时接收视频中的音频和视觉信息流，这在野生环境中至关重要，因为动物的行为可能只听到但看不到(比如在树后敲坚果)或听不到的动作。

对于黑猩猩等具有显著行为复杂性的物种来说，使用人工智能对视频数据集进行跨站点比较，提供了一个机会：以以前不可能的规模和深度捕捉群体之间的微妙变化和随着时间的行为进化，将以往的个体识别方法与视听行为识别相结合，以更丰富地了解动物生活的复杂性。
监测野生物种实地研究的大量数据是应对生物多样性危机的保护工作的一个关键组成部分。使用人工智能进行行为识别有可能捕捉到新的行为指标，从而更准确地衡量受威胁人口的生存能力。

创新价值

新方法将使研究人员和野生动物保护主义者能够大大减少在视频片段中分析动物行为所花费的时间和资源，帮助我们检查行为的社会和生态驱动因素，并监测这些群落是如何应对人类活动导致的气候变化和栖息地退化造成的环境压力的。

创新关键点

该工具是首个使用音频和视频自动识别野生灵长类动物行为的工具。

Using artificial intelligence to identify primate behavior in the wild

The new computer model was trained using videos from two wild chimpanzee populations in West Africa, from Bosu in Guinea and Cantanhuis National Park in Guinea-Bissau, to capture several behaviors: nut-cracking, eating and drumming. The tool is the first to automatically identify wild primate behavior using audio and video, and it builds on previous work that developed an automated way to track and identify individuals using face recognition.
Max Bain, a researcher and doctoral student in Oxford's Department of engineering, said the new model could be applied to thousands of hours of video recordings of chimpanzees in their natural habitats, from camera traps or archival footage. Research using deep learning methods that can simultaneously receive audio and visual streams of information from video is crucial in the wild, where animal behavior may be audible but not visible (such as cracking nuts behind a tree) or inaudible movements.
For the chimpanzee has significant behavior such as complexity of species, using artificial intelligence to the video data set are compared, and across sites provides an opportunity: to impossible before the subtle changes between the groups of scale and depth of the capture and behavior evolved over time, individuals will ever recognition method combining with audio-visual behavior recognition, with a richer understanding of the complexity of animal life.
Monitoring large amounts of data from field studies of wild species is a key component of conservation efforts to address the biodiversity crisis. Behavioral recognition using AI has the potential to capture new behavioral indicators that can more accurately measure the viability of threatened populations.