Computer Vision

Powered by OpenCV and YOLOv5, Inigami brings production-grade computer vision to your applications through a simple JSON protocol.

Object Detection

Powered by YOLOv5 running through OpenCV's DNN module, Inigami's object detection identifies and locates objects across 80+ COCO categories. Each detection includes a class label, confidence score, and precise bounding box coordinates. The server loads the model once at startup for fast inference on subsequent requests.

Request json

1{
2  "request_id": "det-001",
3  "type": "detect_objects",
4  "image_path": "/data/street_scene.jpg"
5}

Response json

1{
2  "request_id": "det-001",
3  "result": {
4    "objects": [
5      {
6        "class": "person",
7        "confidence": 0.95,
8        "bbox": { "x": 120, "y": 80, "w": 200, "h": 350 }
9      },
10      {
11        "class": "car",
12        "confidence": 0.88,
13        "bbox": { "x": 400, "y": 200, "w": 300, "h": 180 }
14      },
15      {
16        "class": "bicycle",
17        "confidence": 0.76,
18        "bbox": { "x": 50, "y": 300, "w": 120, "h": 100 }
19      }
20    ]
21  }
22}

Depth Estimation

Generate monocular depth maps from single images using a pipeline of Gaussian blur, Sobel X/Y gradient computation, and magnitude calculation. The resulting depth map is color-mapped using OpenCV's JET colormap and saved to a temporary file. This approach provides fast, approximate depth information without requiring stereo pairs or specialized depth sensors.

Request json

1{
2  "request_id": "depth-001",
3  "type": "estimate_depth",
4  "image_path": "/data/landscape.jpg"
5}

Response json

1{
2  "request_id": "depth-001",
3  "result": {
4    "depth_map_path": "/tmp/depth_landscape_001.jpg",
5    "colormap": "JET",
6    "width": 1920,
7    "height": 1080
8  }
9}

Image Segmentation

Automatic image segmentation using K-means clustering in the LAB color space. The image is converted from RGB to LAB, reshaped for clustering, and segmented into K distinct regions. Each segment mask is extracted and saved as a separate image file, enabling downstream analysis of individual regions. The number of clusters (K) is configurable per request.

Request json

1{
2  "request_id": "seg-001",
3  "type": "segment_image",
4  "image_path": "/data/garden.jpg",
5  "num_segments": 5
6}

Response json

1{
2  "request_id": "seg-001",
3  "result": {
4    "segments": [
5      "/tmp/segment_garden_0.jpg",
6      "/tmp/segment_garden_1.jpg",
7      "/tmp/segment_garden_2.jpg",
8      "/tmp/segment_garden_3.jpg",
9      "/tmp/segment_garden_4.jpg"
10    ],
11    "num_segments": 5
12  }
13}

Keypoint Detection

Detect feature keypoints in images for applications like image matching, stitching, and motion tracking. Inigami uses OpenCV's feature detection algorithms to identify salient points and compute descriptors. Keypoints include position, scale, orientation, and response strength.

Request json

1{
2  "request_id": "kp-001",
3  "type": "detect_keypoints",
4  "image_path": "/data/building.jpg",
5  "method": "ORB",
6  "max_keypoints": 500
7}

Response json

1{
2  "request_id": "kp-001",
3  "result": {
4    "keypoints": [
5      { "x": 245, "y": 120, "size": 31.0, "angle": 78.5, "response": 0.0042 },
6      { "x": 510, "y": 340, "size": 31.0, "angle": 162.3, "response": 0.0038 }
7    ],
8    "count": 487,
9    "method": "ORB"
10  }
11}

Edge Detection

Multiple edge detection algorithms are available: Canny for robust multi-stage edge detection, Sobel for directional gradient computation, and Laplacian for second-derivative edge finding. Each method produces a single-channel edge map that highlights boundaries and contours in the source image.

Request json

1{
2  "request_id": "edge-001",
3  "type": "detect_edges",
4  "image_path": "/data/architecture.jpg",
5  "method": "canny",
6  "threshold1": 50,
7  "threshold2": 150
8}

Response json

1{
2  "request_id": "edge-001",
3  "result": {
4    "edge_map_path": "/tmp/edges_architecture_001.jpg",
5    "method": "canny",
6    "width": 1920,
7    "height": 1080
8  }
9}

Histogram Analysis

Compute color histograms for detailed image statistics. Analyze the distribution of pixel intensities across channels, perform histogram equalization for contrast enhancement, or compare histograms between images for similarity metrics.

Request json

1{
2  "request_id": "hist-001",
3  "type": "analyze_histogram",
4  "image_path": "/data/sunset.jpg",
5  "channels": ["red", "green", "blue"]
6}

Response json

1{
2  "request_id": "hist-001",
3  "result": {
4    "histograms": {
5      "red": { "mean": 142.5, "std_dev": 45.2, "bins": 256 },
6      "green": { "mean": 98.3, "std_dev": 38.7, "bins": 256 },
7      "blue": { "mean": 165.8, "std_dev": 52.1, "bins": 256 }
8    },
9    "histogram_image_path": "/tmp/histogram_sunset_001.png"
10  }
11}

Integrate Computer Vision Today

Send a JSON request, get structured results. No machine learning expertise required.

Read the Docs