Intel/heatmap-generation

Heatmap Generation

Property	Value
Category	Object Detection + Speed Heatmap Aggregation
Base Model	YOLO26 (Ultralytics, default) or DETR (Facebook, alternative)
Source Framework	PyTorch (Ultralytics, Hugging Face Transformers)
Supported Precisions	FP32, FP16, INT8 (mixed-precision)
Inference Engine	OpenVINO
Hardware	CPU, GPU, NPU
Detected Class(es)	All 80 COCO classes (heatmap colored by object speed)

---

Overview

Heatmap Generation is a Metro Analytics use case that detects objects across video frames and colors each region of the scene by how fast traffic moves through it. It is built on YOLO26, a state-of-the-art real-time object detector, quantized to INT8 for efficient inference on Intel hardware.

Each detection's per-frame displacement is used as a speed estimate, deposited over the object's footprint and averaged per location with Gaussian smoothing into a color-coded overlay.

The overlay uses the following color scheme:

• Red -- fast-moving traffic.
• Yellow / green -- medium speed.
• Blue -- slow-moving or stationary traffic.

Typical Metro deployments include:

• Traffic Speed Mapping -- highlight fast corridors and slow/congested lanes.
• Congestion Detection -- surface persistently slow (blue) areas for safety planning.
• Pedestrian Flow Analysis -- compare fast throughways against lingering areas.
• Incident Spotting -- flag unusually fast or stalled movement.

Available variants: yolo26n, yolo26s, yolo26m, yolo26l. Smaller variants (yolo26n, yolo26s) are recommended for high-FPS edge deployment; larger variants improve recall for small or distant objects.

Model Selection

DETR (DEtection TRansformer) is available as an alternative to YOLO26. DETR is a transformer-based, NMS-free detector trained on COCO and released under the Apache-2.0 license. The export script embeds DETR normalization, box decoding, and the class remap into the OpenVINO IR, so the DETR model produces the same output layout and class ids as YOLO26. The same OpenVINO and DLStreamer samples run either model; select one with the MODEL constant at the top of each sample.

Model	Variants	Input Size	Output	License
YOLO26 (default)	yolo26n, yolo26s, yolo26m, yolo26l	640 x 640	[1, 300, 6]	AGPL-3.0
DETR	detr-resnet-50	800 x 800	[1, 100, 6]	Apache-2.0

Choose YOLO26 for maximum throughput on edge devices. Choose DETR when a permissively licensed transformer detector is preferred; it is heavier and runs at a lower frame rate.

---

Prerequisites

Target stack: Intel DLStreamer 2026.3.0 RC1 with the matching OpenVINO GenAI bundle 2026.5.0.0.dev20261030 (OpenVINO Runtime 2026.5.0.dev20261030). Use the native RC1 setup before running these samples; this is a development stack, not a stable release. See the device validation matrix for measured CPU, GPU, and NPU results.

• Python 3.11+
• Install OpenVINO (latest version)
• Install Intel DLStreamer (latest version)

Create and activate a Python virtual environment before running the scripts:

python3 -m venv .venv --system-site-packages
source .venv/bin/activate

Note: The --system-site-packages flag is required so the virtual environment can access the system-installed OpenVINO and DLStreamer Python packages.

---

Getting Started

Download and Quantize Model

Run the provided script to download, export to OpenVINO IR, and optionally quantize:

chmod +x export_and_quantize.sh
./export_and_quantize.sh

This exports the default yolo26n model in FP16 precision.

Optional: Select a Different Model, Variant, or Precision

./export_and_quantize.sh yolo26n FP32          # full-precision
./export_and_quantize.sh yolo26n INT8          # quantized
./export_and_quantize.sh yolo26s               # larger variant, default FP16
./export_and_quantize.sh detr-resnet-50        # DETR alternative, default FP16
./export_and_quantize.sh detr-resnet-50 INT8   # DETR quantized

The first argument selects the model: any YOLO26 variant (yolo26n, yolo26s, yolo26m, yolo26l) or DETR (detr-resnet-50). The script performs the following steps:

1. Installs dependencies (openvino, ultralytics for YOLO26 or transformers, torch for DETR; adds nncf for INT8).
2. Downloads a sample test image (test.jpg) and a sample test video (test_video.mp4).
3. Downloads the PyTorch weights and exports to OpenVINO IR.
4. (INT8 only) Quantizes the model using NNCF post-training quantization.

Output files (replace yolo26n with the selected model, for example detr-resnet-50):

• yolo26n_openvino_model/ -- FP32 or FP16 OpenVINO IR model directory.
• yolo26n_heatmap_int8.xml / .bin -- INT8 quantized model (only when INT8 is selected).

Precision / Device Compatibility

Precision	CPU	GPU	NPU
FP32	Yes	Yes	No
FP16	Yes	Yes	Yes
INT8	Yes	Yes	Yes

OpenVINO Sample

The sample below runs YOLO26 inference on a video, estimates each object's speed from its per-frame displacement, and writes a speed-colored heatmap overlay (red = fast, blue = slow) to output_openvino.mp4. It also saves the final speed heatmap as heatmap.jpg. Change the device string to run on CPU, GPU, or NPU.

The full sample is in openvino_sample.py; run it with python3 openvino_sample.py or copy the code below.

# SPDX-License-Identifier: MIT
# Copyright (C) Intel Corporation
#
# Heatmap Generation: OpenVINO Runtime sample.
# See README.md for prerequisites and inputs. Run: python3 openvino_sample.py

import cv2
import numpy as np
import openvino as ov

CONF_THRESHOLD = 0.4
# Select the model: "yolo26n" (default) or "detr-resnet-50".
MODEL = "yolo26n"
INPUT_SIZE = 800 if MODEL.startswith("detr") else 640
HEATMAP_ALPHA = 0.55
MATCH_DIST = 80.0   # max px between frames to treat detections as the same object
MAX_SPEED = 20.0    # px/frame that maps to full red


def render_speed_heatmap(frame, speed_sum, count, alpha):
    """Color traffic by average speed: blue = slow, yellow = medium,
    red = fast. Only regions where vehicles were seen are tinted, so
    empty background keeps its original color."""
    avg = np.zeros_like(speed_sum)
    seen = count > 0
    avg[seen] = speed_sum[seen] / count[seen]
    avg = cv2.GaussianBlur(avg, (0, 0), sigmaX=15)
    presence = cv2.GaussianBlur(seen.astype(np.float32), (0, 0), sigmaX=15)
    norm = np.clip(avg / MAX_SPEED, 0, 1)  # 0 = slow (blue), 1 = fast (red)
    color = cv2.applyColorMap((norm * 255).astype(np.uint8), cv2.COLORMAP_JET)
    weight = (np.clip(presence, 0, 1) * alpha)[..., np.newaxis]
    overlay = frame.astype(np.float32) * (1 - weight) + color.astype(np.float32) * weight
    return overlay.astype(np.uint8), color


core = ov.Core()
model = core.read_model(f"{MODEL}_openvino_model/{MODEL}.xml")

# Change device to "GPU" or "NPU" to run on integrated GPU or NPU.
compiled = core.compile_model(model, "CPU")

cap = cv2.VideoCapture("test_video.mp4")
fps = cap.get(cv2.CAP_PROP_FPS) or 30.0
width = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH))
height = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
writer = cv2.VideoWriter(
    "output_openvino.mp4", cv2.VideoWriter_fourcc(*"mp4v"), fps, (width, height))

speed_sum = np.zeros((height, width), dtype=np.float32)
count = np.zeros((height, width), dtype=np.float32)
prev_centroids = []
heatmap_color = None
frame_idx = 0
total_dets = 0

while True:
    ok, frame = cap.read()
    if not ok:
        break
    frame_idx += 1
    h0, w0 = frame.shape[:2]
    sx, sy = w0 / INPUT_SIZE, h0 / INPUT_SIZE

    blob = cv2.resize(frame, (INPUT_SIZE, INPUT_SIZE))
    blob = cv2.cvtColor(blob, cv2.COLOR_BGR2RGB).astype(np.float32) / 255.0
    blob = blob.transpose(2, 0, 1)[np.newaxis, ...]

    output = compiled([blob])[compiled.output(0)][0]
    dets = output[output[:, 4] >= CONF_THRESHOLD]
    total_dets += len(dets)

    cur_centroids = []
    for det in dets:
        x1, y1 = int(det[0] * sx), int(det[1] * sy)
        x2, y2 = int(det[2] * sx), int(det[3] * sy)
        x1, x2 = max(0, x1), min(width, x2)
        y1, y2 = max(0, y1), min(height, y2)
        cx, cy = (x1 + x2) / 2, (y1 + y2) / 2
        cur_centroids.append((cx, cy))

        # Speed = displacement from the nearest detection in the previous frame.
        speed = 0.0
        if prev_centroids:
            d = min(np.hypot(cx - px, cy - py) for px, py in prev_centroids)
            if d <= MATCH_DIST:
                speed = d
        speed_sum[y1:y2, x1:x2] += speed
        count[y1:y2, x1:x2] += 1.0
    prev_centroids = cur_centroids

    overlay, heatmap_color = render_speed_heatmap(frame, speed_sum, count, HEATMAP_ALPHA)
    cv2.putText(overlay, f"Detections: {len(dets)}", (10, 30),
                cv2.FONT_HERSHEY_SIMPLEX, 0.8, (255, 255, 255), 2)
    writer.write(overlay)

cap.release()
writer.release()

if heatmap_color is not None:
    cv2.imwrite("heatmap.jpg", heatmap_color)
    print("Saved: heatmap.jpg")

print(f"Processed {frame_idx} frames, {total_dets} total detections", flush=True)

Device targets:

• "CPU" -- default, works on all Intel platforms.
• "GPU" -- Intel integrated or discrete GPU.
• "NPU" -- Intel NPU (validate with benchmark_app -d NPU).

Expected Output

DLStreamer Sample

The pipeline below runs the FP16 YOLO26 detector via gvadetect. A buffer probe estimates each object's speed from its per-frame displacement and overlays a speed-colored heatmap (red = fast, blue = slow) on each frame before encoding to output_dlstreamer.mp4.

Notes on running this sample:

• Use the FP16 IR (yolo26n_openvino_model/yolo26n.xml or detr-resnet-50_openvino_model/detr-resnet-50.xml). Class names are read automatically from the model's embedded metadata.yaml by DLStreamer 2026.0+ -- no external labels-file is required.

• Export PYTHONPATH so the DLStreamer Python module is importable:

  source /opt/intel/openvino_2026/setupvars.sh
  source /opt/intel/dlstreamer/scripts/setup_dls_env.sh
  export PYTHONPATH=/opt/intel/dlstreamer/python:\
  /opt/intel/dlstreamer/gstreamer/lib/python3/dist-packages:${PYTHONPATH:-}
  

The full sample is in dlstreamer_sample.py; run it with python3 dlstreamer_sample.py or copy the code below.

# SPDX-License-Identifier: MIT
# Copyright (C) Intel Corporation
#
# Heatmap Generation: DLStreamer pipeline sample.
# See README.md for prerequisites and inputs. Run: python3 dlstreamer_sample.py

import gi

gi.require_version("Gst", "1.0")
gi.require_version("GstAnalytics", "1.0")
from gi.repository import Gst, GLib, GstAnalytics

import numpy as np

Gst.init([])

# Import cv2 after Gst.init to avoid GStreamer re-initialization conflicts.
import cv2

INPUT_VIDEO = "test_video.mp4"
HEATMAP_ALPHA = 0.55
MATCH_DIST = 80.0   # max px between frames to treat detections as the same object
MAX_SPEED = 20.0    # px/frame that maps to full red


def render_speed_heatmap(frame, speed_sum, count, alpha):
    """Color traffic by average speed: blue = slow, yellow = medium,
    red = fast. Only regions where vehicles were seen are tinted, so
    empty background keeps its original color."""
    avg = np.zeros_like(speed_sum)
    seen = count > 0
    avg[seen] = speed_sum[seen] / count[seen]
    avg = cv2.GaussianBlur(avg, (0, 0), sigmaX=15)
    presence = cv2.GaussianBlur(seen.astype(np.float32), (0, 0), sigmaX=15)
    norm = np.clip(avg / MAX_SPEED, 0, 1)  # 0 = slow (blue), 1 = fast (red)
    color = cv2.applyColorMap((norm * 255).astype(np.uint8), cv2.COLORMAP_JET)
    weight = (np.clip(presence, 0, 1) * alpha)[..., np.newaxis]
    overlay = frame.astype(np.float32) * (1 - weight) + color.astype(np.float32) * weight
    return overlay.astype(np.uint8)


# Select the model: "yolo26n" (default) or "detr-resnet-50".
MODEL = "yolo26n"

# For CPU: change device=GPU to device=CPU.
# For NPU: change device=GPU to device=NPU (batch-size=1, nireq=4 recommended).
pipeline_str = (
    f"filesrc location={INPUT_VIDEO} ! decodebin3 ! "
    "videoconvert ! video/x-raw,format=BGR ! "
    f"gvadetect model={MODEL}_openvino_model/{MODEL}.xml "
    "device=GPU "
    "threshold=0.4 ! queue ! "
    "appsink name=sink emit-signals=false sync=false"
)
pipeline = Gst.parse_launch(pipeline_str)
sink = pipeline.get_by_name("sink")
pipeline.set_state(Gst.State.PLAYING)

speed_sum = None
count = None
prev_centroids = []
writer = None
frame_idx = 0
total_dets = 0

while True:
    sample = sink.emit("pull-sample")
    if sample is None:
        break
    buf = sample.get_buffer()
    caps = sample.get_caps().get_structure(0)
    width = caps.get_value("width")
    height = caps.get_value("height")

    if speed_sum is None:
        speed_sum = np.zeros((height, width), dtype=np.float32)
        count = np.zeros((height, width), dtype=np.float32)

    ok, mapinfo = buf.map(Gst.MapFlags.READ)
    if not ok:
        continue
    frame = np.ndarray((height, width, 3), dtype=np.uint8,
                       buffer=mapinfo.data).copy()
    buf.unmap(mapinfo)
    frame_idx += 1

    rmeta = GstAnalytics.buffer_get_analytics_relation_meta(buf)
    cur_centroids = []
    det_count = 0
    if rmeta is not None:
        idx = 1
        while True:
            ok_od, od = rmeta.get_od_mtd(idx)
            if not ok_od:
                break
            _, x, y, w, h, _ = od.get_location()
            x1, y1 = max(0, int(x)), max(0, int(y))
            x2, y2 = min(width, int(x + w)), min(height, int(y + h))
            cx, cy = (x1 + x2) / 2, (y1 + y2) / 2
            cur_centroids.append((cx, cy))

            # Speed = displacement from the nearest detection last frame.
            speed = 0.0
            if prev_centroids:
                d = min(np.hypot(cx - px, cy - py) for px, py in prev_centroids)
                if d <= MATCH_DIST:
                    speed = d
            speed_sum[y1:y2, x1:x2] += speed
            count[y1:y2, x1:x2] += 1.0
            det_count += 1
            idx += 1
    prev_centroids = cur_centroids
    total_dets += det_count

    overlay = render_speed_heatmap(frame, speed_sum, count, HEATMAP_ALPHA)

    if writer is None:
        writer = cv2.VideoWriter(
            "output_dlstreamer.mp4", cv2.VideoWriter_fourcc(*"mp4v"),
            30.0, (width, height))
    writer.write(overlay)
    print(f"Frame {frame_idx}: detections={det_count}", flush=True)

pipeline.set_state(Gst.State.NULL)
if writer:
    writer.release()
print(f"Processed {frame_idx} frames, {total_dets} total detections", flush=True)

Device targets:

• device=GPU -- default in the sample code.
• device=CPU -- change device=GPU to device=CPU.
• device=NPU -- change device=GPU to device=NPU; use batch-size=1 and nireq=4 for best NPU utilization.

Expected Output

---

License

Licensed under the MIT License. See LICENSE for details.

References

• YOLO26 Documentation
• DETR Model Card (facebook/detr-resnet-50)
• End-to-End Object Detection with Transformers (DETR paper)
• Ultralytics Heatmap Guide
• OpenCV Color Maps
• OpenVINO Documentation
• Intel DLStreamer