📸 Lost + Found Dual IMX296 External Trigger Driver

Generates the external trigger for two InnoMaker CAM-IMX296RAW global-shutter camera modules from a single Raspberry Pi Pico 2 (RP2350), so both sensors expose at the same instant.

---

🎯 Why this exists

Two IMX296 modules left free-running do not expose together. Measured on a Pi 5 rig via picamera2 SensorTimestamp, the two sensors sat 21.54 ms apart — a fixed phase offset, constant to the sample, and not something software can close: it is when the sensors expose, not when the host asks for a frame.

For stereo, that offset turns any relative motion into false disparity:

Subject motion	Apparent error (f ≈ 968 px, Z ≈ 510 mm)
5 mm/s	0.20 px
20 mm/s	0.82 px
40 mm/s	1.64 px

It is worse on a moving platform, where the camera is what moves: at 1 m/s a rover accumulates 21 mm of ego-motion between its two eyes on every frame — a bias no calibration can remove, because it is not a property of the rig's geometry.

Driving both sensors from one trigger removes the offset at the source.

---

🎛️ Hardware Description & Specifications

The camera's J3 header is an opto-isolated trigger input (on-board 200 Ω series resistor, LED forward voltage ≈ 1.25 V). Exposure is controlled entirely by the pulse:

• Polarity: idle HIGH, exposure while LOW
• Exposure: LOW pulse width + 14.26 µs (sensor-added offset)
• Frame rate: set by how often you pulse — 30 Hz of pulses gives 30 fps
• Minimum exposure: ~30 µs
• Drive current: (3.3 − 1.25) / 200 ≈ 10.25 mA per camera

⚠️ Do not parallel both cameras onto one GPIO. That is 20.5 mA, over the RP2350's 12 mA per-pin recommendation. This driver uses one pin per camera.

---

🔌 Wiring Diagram (Pico 2)

Camera Pin	Pico 2 GPIO	Pico 2 Physical Pin	Notes
CAM0 J3 Trig+	GP16	Pin 21	PWM slice channel A
CAM1 J3 Trig+	GP17	Pin 22	PWM slice channel B (same slice)
Both J3 Trig−	GND	Pin 23	Common ground for both opto inputs

Three adjacent physical pins, so it leaves the board as one 3-wire bundle.

Why GP16 + GP17 specifically

They are channels A and B of one PWM slice. A slice has a single counter, so giving both channels the same compare value makes their edges align in hardware — no software timing, no interrupt latency, nothing to drift. Any A/B pair on one slice works (GP0+GP1, GP2+GP3, …); lf_imx296_trigger_init() rejects pins that do not qualify rather than silently giving you two loosely-related signals.

Powering the Pico from the host's USB commons the grounds and forfeits the isolation the optocouplers provide. Harmless, but a separate supply keeps it if you want it.

---

🚀 Usage Example

#include "pico/stdlib.h"
#include "lf_imx296_trigger.h"
#include <stdio.h>

#define TRIG_PIN_A 16
#define TRIG_PIN_B 17

int main() {
    stdio_init_all();

    lf_imx296_trigger_t trig;
    if (!lf_imx296_trigger_init(&trig, TRIG_PIN_A, TRIG_PIN_B)) {
        printf("Failed to initialize trigger driver\n");
        return -1;
    }

    // 30 fps, 8 ms exposure
    lf_imx296_trigger_configure(&trig, 30.0f, 8000.0f);
    lf_imx296_trigger_start(&trig);

    while (true) {
        tight_loop_contents();
    }
}

For capture-on-demand — a rover grabbing a pair only while stationary — leave the generator stopped and call lf_imx296_trigger_fire_once() per pair. That path also has no minimum rate, unlike continuous mode.

---

🖥️ Host-side setup (Raspberry Pi)

The Pico is only half of it. On the host, before the pulses start:

1. Raise libcamera's frame timeout. In trigger mode no pulse means no frames, and libcamera will abort the stream on its default timeout. In timeout.yaml, uncomment camera_timeout_value_ms and set it to 60000, then:

export LIBCAMERA_RPI_CONFIG_FILE=timeout.yaml

2. Put each sensor into trigger mode. This is an I2C register write to the sensor (0x30AE = 0x01), not a kernel module parameter:

sudo python3 i2c.py trigger on --bus <CAM0 i2c bus>
sudo python3 i2c.py trigger on --bus <CAM1 i2c bus>

Bus numbers are board-specific — check i2cdetect for the sensor at 0x1a rather than copying them from the vendor manual.

⚠️ Auto-exposure stops working in trigger mode. Brightness is set by the pulse width alone, so expect to tune exposure once by eye. The demo's serial interface exists for exactly that: e 8000 to try 8 ms, no reflashing.

3. Verify. Re-measure the inter-eye SensorTimestamp skew. It should collapse from tens of milliseconds to microseconds. That is a direct before/after on the real failure mode rather than a proxy for it.

---

🧪 Demo

Build and flash example.c for an interactive serial console:

f <hz>   set frame rate      (e.g. 'f 30')
e <us>   set exposure        (e.g. 'e 8000')
s        start pulsing
x        stop pulsing
1        fire a single exposure (must be stopped)
?        show status

---

📎 References

• InnoMaker CAM-IMX296RAW user manual, ch. 2.5 (J3 pinout) and ch. 4 (external trigger)
• https://github.com/INNO-MAKER/cam-imx296raw-trigger