A Hybrid Approach: Image Processing Techniques and Deep Learning Method for Cow Detection and Tracking System
Cho Cho Mar, Thi Thi Zin, Ikuo Kobayashi, Yoichiro Horii
Abstract
Cow detection and tracking system plays an important role in cattle farming and diary community to reduce expenses and workload. This research presents how the conventional image processing techniques can be combined with deep learning concepts to establish cow detection and tracking system. Specifically, we first employ a Hybrid Task Cascade (HTC) instance segmentation network for cow detection. We then built the multiple objects tracking (MOT) algorithm utilizing location and appearance cues (color and CNN features) to carry out cow tracking process. To leverage the robustness of the system, we also considered the recent features from the previous tracked cow.
Topics & Concepts
Computer scienceArtificial intelligenceRobustness (evolution)Computer visionLeverage (statistics)Image segmentationTracking systemSegmentationWorkloadImage processingDeep learningPattern recognition (psychology)Image (mathematics)Kalman filterChemistryOperating systemGeneBiochemistryFood Supply Chain TraceabilityVideo Surveillance and Tracking MethodsAnimal Behavior and Welfare Studies