chaselolley.github.io/project_grab.html at main · chaselolley/chaselolley.github.io · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
<!DOCTYPE html>
<html lang="en">
<head>
    <meta charset="UTF-8">
    <meta name="viewport" content="width=device-width, initial-scale=1.0">
    <title>GRAB - Chase Lolley</title>
    <style>
        /* Reuse the same CSS variables for consistency */
        :root {
            --bg-color: #121212;
            --text-color: #e0e0e0;
            --accent-color: #64ffda;
            --card-bg: #1e1e1e;
            --font-main: 'Segoe UI', Roboto, Helvetica, Arial, sans-serif;
        }
        body { background-color: var(--bg-color); color: var(--text-color); font-family: var(--font-main); margin: 0; padding: 0; line-height: 1.6; }
        a { color: var(--accent-color); text-decoration: none; }
        a:hover { text-decoration: underline; }

        /* Layout for the detail page */
        .container { max-width: 800px; margin: 0 auto; padding: 40px 20px; }

        /* Back Button Style */
        .back-nav { margin-bottom: 40px; }
        .back-nav a { font-size: 0.9rem; color: #888; }
        .back-nav a:hover { color: var(--accent-color); }

        /* Typography */
        h1 { font-size: 2rem; color: #fff; margin-bottom: 10px; }
        h3 { color: #fff; margin-top: 30px; border-bottom: 1px solid #333; padding-bottom: 5px; }
        ul { margin-bottom: 20px; }
        li { margin-bottom: 10px; }
        img { border-radius: 4px; border: 1px solid #333; }

        .meta-grid { display: grid; grid-template-columns: repeat(auto-fit, minmax(200px, 1fr)); gap: 15px; background: #1e1e1e; padding: 20px; border-radius: 8px; margin-bottom: 40px; font-size: 0.9rem; }
    </style>
</head>
<body>

<div class="container">

    <div class="back-nav">
        <a href="index.html">&larr; Back to Portfolio</a>
    </div>

    <h1>Generalizable Recognition of Activity in Breeders (GRAB)</h1>

    <div class="meta-grid">
        <div><strong>Affiliation:</strong> GTRI - Intelligent Sustainable Tech</div>
        <div><strong>Timeframe:</strong> August 2025 - Present</div>
        <div><strong>Status:</strong> Ongoing</div>
        <div><strong>Links:</strong> <a href="#" target="_blank">Tracklet Output (Video)</a></div>
    </div>

    <h3>Project Overview</h3>
    <p>
        Internal research project focused on building a robust, end-to-end computer vision pipeline for recognizing poultry behaviors from barn video. The system is designed to generalize across barns, lighting conditions, and camera viewpoints, supporting long-term goals in precision agriculture and animal welfare monitoring.
    </p>
    <p><em>Note: Descriptions and visuals have been intentionally abstracted to respect confidentiality.</em></p>

    <h3>System and Technical Context</h3>
    <img src="images/grab_pipeline.png" alt="GRAB pipeline" style="width: 100%; margin-bottom: 5px;">
    <p style="font-size: 0.85rem; color: #888; text-align: center;"><em>Figure: End-to-end perception pipeline.</em></p>

    <p>The system operates in real agricultural environments characterized by dense scenes, heavy occlusion, and variable lighting. Raw video is processed through a multi-stage perception pipeline that isolates individual animals and classifies behaviors over extended time horizons.</p>

    <h3>Responsibilities and Contributions</h3>

    <strong>Perception Pipeline Design</strong>
    <ul>
        <li>Designed an end-to-end poultry activity recognition pipeline spanning segmentation, tracking, and temporal behavior classification.</li>
    </ul>

    <strong>Segmentation and Data Quality</strong>
    <ul>
        <li>Implemented a segmentation-based preprocessing stage using vision foundation models to isolate individual animals.</li>
        <li>Developed a custom binary classifier to filter segmentation outputs.</li>
    </ul>

    <img src="images/grab_conf_mat.png" alt="Confusion Matrix" style="width: 100%; max-width: 500px; display:block; margin: 20px auto;">

    <strong>Temporal Modeling</strong>
    <ul>
        <li>Built a tracklet generation pipeline that aggregates frame-level segmentations into temporally consistent clips.</li>
        <li>Evaluated transformer-based architectures for long-range temporal reasoning.</li>
    </ul>

    <h3>Results and Evidence</h3>
    <ul>
        <li>Demonstrated reliable end-to-end activity recognition from raw video under real-world barn conditions.</li>
        <li>Validated model robustness across multiple environments and acquisition conditions.</li>
    </ul>

    <h3>Tools Used</h3>
    <p>Python, PyTorch, ROS, OpenCV, Transformer Architectures.</p>

    <div style="margin-top: 50px; border-top: 1px solid #333; padding-top: 20px; text-align: center;">
        <a href="index.html">Back to Home</a>
    </div>

</div>

</body>
</html>