Canvas 插值 + Gemini 截图把 F1 历史比赛做成了实时赛道可视化

while True:
    if playing and frame_index < len(frames):
        await websocket.send_json({"type": "frame", **prepare_frame(frames[frame_index])})
        frame_index += 1
        if frame_index >= len(frames):
            playing = False
            await websocket.send_json({"type": "finished"})
            continue
        # base_interval = 0.5 秒，speed 控制倍速
        remaining = base_interval / speed
        while remaining > 0 and playing:
            chunk = min(remaining, 0.05)
            await check_command(chunk)  # 每 50ms 检查一次命令
            remaining -= chunk
    else:
        await check_command(1.0)

const play  = () => { send("play");        setState(s => ({ ...s, playing: true })); };
const pause = () => { send("pause");       setState(s => ({ ...s, playing: false })); };
const seek  = (time: number) => send(`seek:${time}`);
const setSpeed = (speed: number) => send(`speed:${speed}`);

// 插值持续时间故意比帧间隔（500ms）长，确保下一帧到达时上一段动画尚未结束
const BASE_INTERP_MS = 750;
 
// 新帧到来时，记录当前视觉位置作为起点，新坐标作为终点
entry.prevX = entry.prevX + (entry.targetX - entry.prevX) * t; // 当前视觉位置
entry.targetX = drv.x;  // 新目标位置
entry.startTime = now;
entry.duration = BASE_INTERP_MS / Math.max(speed, 0.25);
 
// requestAnimationFrame 循环里，每帧计算当前插值位置
const t = Math.min(elapsed / entry.duration, 1);
const x = entry.prevX + (entry.targetX - entry.prevX) * t;
const y = entry.prevY + (entry.targetY - entry.prevY) * t;

VISION_MODEL = "google/gemini-2.0-flash-001"
 
# 提示词要求 Gemini 识别：当前圈数、间隔显示模式（LEADER/INTERVAL）、每位车手的位置和时间差
# 并统一换算成"与领先者的累计时间差"格式
payload = {
    "model": VISION_MODEL,
    "messages": [{"role": "user", "content": [
        {"type": "text", "text": EXTRACT_PROMPT},
        {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{b64}"}},
    ]}],
    "temperature": 0,  # 关闭随机性，要求确定性输出
}

def _match_frame(frames, extracted):
    target_lap = extracted.get("lap", 1)
    # 先过滤出前后一圈范围内的候选帧，减少比对量
    candidate_frames = [
        (i, f) for i, f in enumerate(frames)
        if abs(f.get("lap", 0) - target_lap) <= 1
    ]
    best_score = float("inf")
    for idx, frame in candidate_frames:
        score = 0.0
        for abbr, target_gap in target_gaps.items():
            frame_gap = get_driver_gap(frame, abbr)
            if frame_gap is None:
                score += 50.0  # 找不到这位车手，加惩罚分
            else:
                score += abs(frame_gap - target_gap)  # 时间差的绝对误差
        if score < best_score:
            best_score = score
            best_idx = idx
    return frames[best_idx]

git clone https://github.com/adn8naiagent/F1ReplayTiming.git
cd F1ReplayTiming
docker compose up