Pashu Vision: An MI-Based Automated Cattle Body Measurement and Evaluation System Using Single-Image Keypoint Detection

Authors:

Asst. Prof. Vishnu Prashd Verma, Piyush Sahu Aakash, Katch Ankita Banjare, Nishant Kumar

Abstract

Accurate and standardized assessment of cattle body conformation is essential for dairy improvement programs, especially under the Rashtriya Gokul Mission (RGM). Existing manual Animal Type Classification (ATC) processes rely heavily on trained personnel and are prone to subjectivity, fatigue, misjudgment, and inconsistent measurements. Recent research has explored deep-learning-based cattle pose estimation, body size prediction, and digital livestock management; however, most solutions require multi- view images, specialized hardware, controlled environments, or provide limited trait extraction.

This study proposes Pashu Vision, an end-to-end AI system for automated cattle body measurement and disease detection using a single mobile image captured in field conditions. The system integrates: (a) 34 anatomical keypoints for holistic cow body mapping, (b) pose-based geometry models for extracting body length, height, hip width, chest depth, and other ATC- critical traits, (c) machine-learning scoring for trait evaluation, and (d) a multi-class disease classification module for detecting major cattle diseases including mastitis, lumpy skin disease, foot-and-mouth disease, and lameness.

In addition to body measurement and ATC scoring, the proposed system is extended to perform major cattle disease classification using image-based machine learning models. The disease module focuses on common and high- impact diseases affecting cows and buffaloes in Indian field conditions, enabling early-stage health assessment using the same captured image.

Unlike previous works that use limited keypoints (10-22), require 3D sensors, or rely on enclosure-based imaging, our proposed system supports Indian indigenous breeds, handles uncontrolled backgrounds, and operates directly on smartphones. The proposed solution improves upon existing literature by providing a field deployable, scalable, low-cost alternative to manual ATC scoring and veterinary health assessment, enabling automated measurements, reduced subjectivity, integration with BPA, and suitability for PT/PS programs under RGM.

The proposed system offers a low-cost, scalable, and objective alternative to manual ATC scoring and disease screening, addressing critical limitations in current research regarding keypoint coverage, measurement completeness, breed diversity, health monitoring, and real-world deployability.By enabling standardized, automated, and field- ready cattle evaluation with integrated health assessment, Pashu Vision holds significant potential to enhance progeny testing, pedigree selection, disease management, and overall genetic improvement efforts within India's dairysector.